Table of Content

30 July 2026, Volume 8 Issue 4

Previous Issue   

Overview Article

Research Progress and Prospects of Cloud-Edge-Device Integrated Middleware for Agricultural Product Quality Control |

WANG Ting, WANG Na, CUI Yunpeng, LIU Juan

2026, 8(4):  1-16.  doi:10.12133/j.smartag.SA202510002

Asbtract ( 152 )   HTML ( 10)   PDF (2426KB) ( 10 )  

Figures and Tables | References | Related Articles | Metrics

[Significance] Against the backdrop of consumption upgrading and growing health awareness, consumer concerns about agricultural products are shifting from mere safety to freshness, taste, and brand reputation, making quality a core target of modern agricultural upgrading and accelerating the transition of production modes toward precise and intelligent control. However, the traditional centralized cloud‑computing architecture dominated by a central cloud platform is increasingly constrained in weak‑network environments, low‑latency control, and local autonomy, and thus can hardly support the closed‑loop requirements of "timely detection-rapid response-continuous optimization" in quality control. Consequently, cloud-edge-device integrated collaborative architectures are regarded as an important development direction, and it is necessary to systematically sort out their key technologies and application foundations in the agricultural domain so as to provide unified theoretical and technical support for the design of integrated middleware oriented to agricultural product quality control. [Progress] Starting from the current application status of cloud-edge-device integration in agriculture, the paper focuses on three typical scenarios—field planting, facility horticulture, and livestock farming—and summarizes the functional division of perception, computation, and control among cloud, edge, and device, as well as the differences in business requirements under various geographical conditions and production modes. On this basis, it identifies multiple technical challenges in practical deployments, including heterogeneous adaptation difficulties caused by the coexistence of numerous industrial and IoT protocols, poor interoperability due to inconsistent data definitions and encodings, resource imbalance arising from the mismatch between edge computing capacity and task loads, and coarse‑grained, inefficient cloud-edge collaborative scheduling. To address these challenges, the paper synthesizes relevant research on artificial intelligence, edge computing, federated learning, and blockchain, and proposes improvement ideas for cloud-edge-device collaboration from the perspectives of resource‑allocation optimization, edge‑intelligent inference, privacy protection, and trusted sharing. Furthermore, targeting the concrete requirements of intelligent quality control for agricultural products, it constructs a cloud-edge-device integrated technical framework: At the data layer, it integrates standardized mapping of quality data and a multi‑source data dictionary system; at the access layer, it realizes heterogeneous device‑protocol conversion and unified access; at the computing layer, it achieves dynamic matching between computing resources and services through intelligent task scheduling and load balancing, multi‑level caching, model hot‑updating, and federated deployment. At the collaboration level, it proposes a "device-crop-model" trinity abstraction that brings physical devices, agronomic objects, and algorithmic models under unified middleware management, thereby masking underlying hardware differences while supporting elastic adaptation and capability reuse across diverse scenarios. [Conclusions and Prospects] The cloud-edge-device integrated collaboration and its adaptive middleware constitute the key infrastructure and connective hub of intelligent quality‑control systems for agricultural products, providing reusable architectural concepts and technical references for the development and engineering implementation of systems for precise quality perception, automatic control, and intelligent decision‑making. Looking ahead, further in‑depth research is needed on cross‑scenario data and protocol standard systems oriented to quality indicators and control actions, on the integrated application of edge intelligence and large models, on the enhancement of privacy‑protection and trusted‑computing mechanisms, and on large‑scale engineering demonstrations and validation, so as to continuously improve the intelligence level, operational reliability, and scalability of agricultural product quality‑control systems and to promote the transition of related technologies from pilot applications to large‑scale deployment.

Research Progress on Intelligent Measurement Technologies for Livestock Body-Size Phenotypes Based on 3D Point Clouds |

WANG Haiyan, LI Zechen, CAO Jusong, LI Xuan, LI Jiawei, LI Guoliang, XU Xuewen

2026, 8(4):  17-34.  doi:10.12133/j.smartag.SA202605012

Asbtract ( 89 )   HTML ( 5)   PDF (9041KB) ( 5 )  

Figures and Tables | References | Related Articles | Metrics

[Significance] Livestock body-size parameters are important phenotypic indicators for evaluating growth and development, body condition, production performance, genetic evaluation, and breeding selection. Traditional manual measurement is time-consuming, labor-intensive, operator-dependent, and may induce stress and safety risks during animal restraint. Although two-dimensional imaging enables non-contact measurement, its lack of depth information limits the accurate estimation of girth-related traits and local body morphology. By directly describing three-dimensional body-surface structures, point clouds provide a promising basis for automated and objective livestock phenotyping. The aim of this review is to systematically summarize advances in three-dimensional point cloud-based livestock body measurement, to clarify the main technical workflow and methodological routes, to compare their characteristics and application scenarios, and to identify key challenges and research priorities for intelligent three-dimensional phenotyping and smart livestock farming. [Progress] Three-dimensional point cloud technology can reconstruct livestock body structures while retaining the advantages of non-contact measurement. It can be used for measuring linear traits, such as body length, body height, and body width; girth-related traits, including chest girth, abdominal girth, rump girth, and cannon circumference; and local morphological features such as rump shape, hindquarter fullness, and limb and hoof structure. With pigs, cattle, and sheep as the main research objects, this review examines the technical workflow of point cloud acquisition, preprocessing, completion, segmentation, keypoint localization, and body-size calculation. Livestock point cloud acquisition sensors and fixed and mobile acquisition schemes are summarized. Preprocessing methods, including multi-view registration, filtering and denoising, posture normalization, and point cloud completion, are reviewed. These procedures can reduce background interference, motion distortion, occlusion, data loss, and posture variation, thereby improving measurement consistency. Two technical routes for body-size measurement are discussed. In the first route, anatomical keypoints are directly localized from RGB images, depth images, or point clouds using geometric features, projection analysis, curvature, skeleton models, or deep learning-based detectors. This route is suitable for linear measurements under standardized postures. In the second route, the animal is first segmented into anatomically relevant regions, after which keypoints or measurement sections are identified within the corresponding regions. By narrowing the search space, segmentation-based methods can reduce interference from unrelated body parts and are particularly suitable for variable postures, partial occlusion, and girth-related measurements. After keypoint localization or section determination, linear traits are generally calculated using Euclidean distances, whereas girth traits are estimated through contour extraction, curve fitting, and accumulated curve length. Overall, a development trend from manual annotation and rule-based geometric analysis toward automated methods integrating keypoint detection, point cloud segmentation, regional constraints, posture adaptation, and cross-species transfer is identified. [Conclusions and Prospects] Existing studies indicate that three-dimensional point cloud-based livestock body measurement has promising application prospects for automated, non-contact, and high-precision phenotype acquisition. However, several challenges remain, including unstable point cloud quality in complex farming environments; insufficient robustness, transferability, and edge-deployment capability of algorithms; limited multi-species and multi-posture datasets; and inconsistent definitions, annotations, and evaluation metrics. Future research should focus on improving acquisition accuracy, synchronization, and environmental adaptability and developing low-cost, lightweight, and reliable systems. Multimodal fusion of depth cameras, LiDAR, and RGB images, together with mobile platforms and multi-view acquisition, should be investigated for continuous measurement of group-housed animals. Robust, lightweight, and transferable models should be developed through self-supervised or unsupervised pretraining, semi-supervised learning, transfer learning, model pruning, knowledge distillation, and lightweight feature extraction. Open datasets should cover multiple species, breeds, growth stages, postures, and farming scenarios, accompanied by unified keypoint definitions, anatomical annotations, data partitions, and error metrics. In addition, body-size measurements should be integrated with body condition scoring, local morphological evaluation, growth and development monitoring, production performance analysis, genetic evaluation, and breeding selection. This will promote the transition from geometric parameter extraction to intelligent three-dimensional phenotypic analysis, ultimately achieving accurate, real-time, and high-throughput automatic livestock body measurement for precision feeding and smart livestock farming.

Information Perception and Acquisition

Rapid Prediction Method for Steady-State CO₂ Concentration in Plant Photosynthetic Measurements with Uncertainty-Adaptive Adjustment |

LUO Mingyang, DAI Hangyu, TANG Hao, WU Yingkui, GUO Ya

2026, 8(4):  35-46.  doi:10.12133/j.smartag.SA202512001

Asbtract ( 72 )   HTML ( 4)   PDF (2066KB) ( 14 )  

Figures and Tables | References | Related Articles | Metrics

[Objective] The gas exchange method is the most widely applied and most direct quantitative measurement technique in plant photosynthesis research. Plant photosynthetic measurement systems based on the gas exchange method usually rely on the "steady-state paradigm" to determine the photosynthetic rate and other physiological parameters. During the measurement process, the system needs to wait until the CO₂ concentration reaches a steady state. However, this process is affected by factors such as plant physiology, leaf chamber volume, and sensor response characteristics, which severely restricts measurement efficiency. To this end, a new method is proposed that predicts the steady-state CO₂ concentration in advance based on the dynamic response sequence of the sensor and adaptively determines the measurement termination time using predictive uncertainty. [Methods] A dual-branch encoding network, DBE-TSNet (Dual-Branch Encoder for Time-Series Network), was constructed. The overall architecture of the network consisted of dual feature encoders and an aggregation encoder. The dual feature encoders adopted a structurally symmetric and parameter-independent branch design. They combined differential enhancement, cross-channel convolution-linear channel transformation, and global average pooling to model the dynamic response processes of the CO₂ concentration decrease and increase stages perceived by the system, respectively. After concatenating the features of the two branches, the aggregation encoder generated a four-dimensional output vector through a two-layer multilayer perceptron decoder: predicted decrease value (t̂₁), predicted decrease uncertainty (σ̂₁²), predicted increase value (t̂₂), and predicted increase uncertainty (σ̂₂²). During training, a Gaussian negative log-likelihood loss function was adopted to simultaneously learn the steady-state prediction together with its associated predictive uncertainty, while sample weights were adaptively adjusted accordingly. By setting uncertainty thresholds, adaptive truncation of the input sequence was realized, enabling determination of the prediction timing and the steady-state value, thereby improving measurement efficiency. A photosynthetic measurement system was built based on a single CO₂ sensor, using a time-shared differential measurement mode. Gas exchange data was collected from eight plant species (corn, potato, radish, bok choy, lettuce, loquat, orange tree, and paper mulberry) under three light intensity levels (low, medium, and high) for model training and validation. [Results and Discussions] By comparison with six typical time-series prediction models, including one-dimensional convolutional neural network (1D-CNN) and recurrent neural network (RNN), the DBE-TSNet model achieved the best performance in the steady-state prediction task: The MAE of the decreasing stage was 1.22 μmol/mol, and that of the increasing stage was 2.07 μmol/mol; the R² values reached 0.996 and 0.984, respectively. To evaluate the relationship between model predictive reliability and output uncertainty, all test samples were divided into ten intervals according to the uncertainty output by the model, and the relationships between each interval and mean absolute error (MAE), root mean square error (RMSE) and mean absolute percentage error (MAPE), and the input segment ratio were counted. The results showed that as uncertainty gradually increased, the prediction error gradually increased, resulting in a significant decline in model predictive reliability. Based on this analysis, the uncertainty thresholds of the decreasing stage and increasing stage were determined to be 0.026 7 and 0.014 6, respectively. Early prediction experiments based on the set thresholds showed that under the premise of maintaining the error below 1 μmol/mol, the model saved about 128.87 s of measurement time on average, shortened the measurement cycle by about 51%, and significantly improved measurement efficiency. [Conclusions] DBE-TSNet integrated dual-branch encoding and uncertainty estimation mechanisms to achieve accurate and early prediction of steady-state CO₂ concentration in plant photosynthetic measurement, effectively solving the problem of long steady-state measurement time in the gas exchange method and transforming the traditional "waiting for steady state" process into a data-driven decision-making process of "dynamic steady-state prediction". The model showed stable performance under multiple plant species and multiple light conditions, with good generalization ability, and could provide key technical support for improving the efficiency of plant photosynthetic measurement systems based on the gas exchange method.

Estimation of Winter Wheat Chlorophyll Content Based on Hyperspectral Remote Sensing and Cross-Regional Transfer Learning |

LI Yiyue, FEI Shuaipeng, LI Lei, JIA Yidan, WANG Duoxia, ZHANG Bohan, XIAO Yonggui, MENG Yaxiong

2026, 8(4):  47-59.  doi:10.12133/j.smartag.SA202601026

Asbtract ( 107 )   HTML ( 10)   PDF (121087KB) ( 13 )  

Figures and Tables | References | Related Articles | Metrics

[Objective] Hyperspectral remote sensing enables accurate estimation of crop chlorophyll content, a key indicator of photosynthetic capacity and nitrogen nutrient status. However, models trained in one region (source domain) often suffer significant performance degradation when applied to another region (target domain) due to "domain shift" caused by differences in soil, climate, and management practices. This study proposed a robust adaptive transfer learning framework (RATL) to mitigate domain shift in cross-regional chlorophyll estimation of winter wheat and achieve accurate inversion. [Methods] A total of 1 491 paired samples of canopy hyperspectral reflectance and leaf soil and plant analyzer development (SPAD) values were collected during the late grain filling stage of winter wheat in the 2023－2024 growing season from Xinxiang and Zhoukou in Henan province. The degree of domain shift between the two regions was quantified using multiple metrics, including maximum mean discrepancy (MMD), Wasserstein distance, and correlation differences. The RATL framework consists of three synergistic modules: (1) Adaptive feature selection, which combines source-target correlation weighting and stability constraints to select 100 core features from the original 1 734-dimensional feature set, with preferential selection of red-edge bands (680－750 nm); (2) daptive feature weight calculation, which fuses random forest importance scores from both domains and feature stability metrics to assign weights to each feature, guiding the model to focus on domain-invariant features; and (3) domain adaptation training, which employs a two-stage XGBoost regressor (pre-training on source domain followed by fine-tuning with 30% of target domain samples) and incorporates a domain discriminator loss (balancing parameter λ=0.1) to encourage learning of domain-invariant representations. Four experimental scenarios were designed: source-only validation (Scenario A), direct transfer (Scenario B), transfer learning comparison with Transfer Component Analysis and Correlation Alignment(Scenario C), and the ideal upper bound using combined data from both regions (Scenario D). Shapley additive explanations (SHAP) analysis was employed to interpret model decision mechanisms, and quantile regression forests were used to generate 90% prediction intervals for uncertainty assessment. [Results and Discussions] Quantitative domain shift analysis revealed significant distribution differences between Xinxiang and Zhoukou maximum mean discrepancy was 0.20, P<0.001, confirming the challenge of domain shift in cross-regional modeling. Direct transfer exhibited substantial performance degradation on the target domain (R²=0.61). Traditional transfer learning methods TCA and CORAL achieved only marginal improvements (R²≈0.64－0.65, transfer gains<7%), proving insufficient for addressing complex domain shift. In contrast, RATL achieved optimal performance on the target domain (R²=0.75, root mean square error was 7.49), representing a 23.40% improvement over direct transfer and reaching 105.60% of the ideal performance achieved with combined data from both regions. SHAP analysis demonstrated that RATL successfully shifted the model's reliance from environmentally sensitive features to red-edge vegetation indices (modified Simple Ratio at 705 nm, normalized difference vegetation index,Green Index) that exhibit consistent responses across regions, enhancing both physiological rationality and regional adaptability of model decisions. Uncertainty quantification showed that RATL achieved higher prediction interval coverage (79.52%) compared to direct transfer (77.94%), while adaptively widening intervals for extreme SPAD values to provide more realistic and reliable uncertainty estimates. These results demonstrate that RATL's multi-module collaborative strategy effectively mitigates agricultural hyperspectral domain shift, significantly improving model interpretability and reliability while maintaining prediction accuracy, outperforming traditional linear alignment methods. [Conclusions] The proposed RATL framework achieves high-precision and high-reliability cross-regional chlorophyll inversion with only a small set of target samples through the synergistic effects of adaptive feature selection, feature weight adjustment, and domain adaptation training. The framework effectively focuses on physiologically meaningful domain-invariant features and provides reliable uncertainty estimates, offering a practical technical solution for regional-scale agricultural remote sensing monitoring. Future work will extend RATL to address temporal domain shift (e.g., across different growth stages) and explore lightweight model implementations for real-time applications.

Recognition of Wheat Waterlogging Stress and Regulation Effect Based on UAV Multispectral Images and VGG21 Model |

LIANG Wanjie, LIU Xiaojun, WU Qian, SUN Chuanliang, LEI Tianjie, XU Deze, ZHENG Xingfei, ZHU Guanglong, TANG Quan

2026, 8(4):  60-69.  doi:10.12133/j.smartag.SA202601017

Asbtract ( 81 )   HTML ( 3)   PDF (2299KB) ( 10 )  

Figures and Tables | References | Related Articles | Metrics

[Objective] In recent years, global warming and increasingly frequent extreme precipitation events have intensified waterlogging stress in agriculture, establishing it as a major abiotic threat to food security.To achieve scientific and precise control of wheat waterlogging disaster and increase yield, a rapid, accurate, and non-destructive method for identifying wheat waterlogging stress and regulation effect was proposed. [Methods] Field trials of control (CK), waterlogging stress, silicon-fertilizer regulation, and amino-acid regulation were conducted during the jointing-booting and flowering-filling stages of wheat. During the experiments, unmanned aerial vehicle (UAV) multispectral images of the wheat canopy were collected using DJI P4 Multispectral. The multispectral image had five bands, namely blue (B), green (G), red (R), red-edge (RE) and near-infrared (NIR), with reflectivity values of (450±16), (560±16), (650±16), (730±16), and (840±26) nm, respectively. At the end of the experiments, grain yield was measured for each treatment and used to evaluate the effectiveness of the regulation measures. The multispectral images were processed, which included radiometric correction, geometric correction, image alignment, segmentation, etc. Then a dataset was established that includes four categories: CK, waterlogging stress, silicon fertilizer regulation, and amino acid regulation. The stratified sampling was used for dataset partitioning. According to the ratio of 7:1.5:1.5, all original multispectral images of each category were randomly divided into the training, validation and testing set, respectively. Then, the original images were segmented into data blocks of 48×48 pixel size for model training and testing. A wheat waterlogging stress and regulation effect recognition model using visual geometry group (VGG) 21 was established by adding convolutional layers, adjusting the convolutional layers of convolutional block, replacing activation functions, and optimizing model hyperparameters based on VGG19. The VGG21 model included 4 convolutional blocks and the convolutional layers were 3, 3, 6, and 6, respectively. To verify the performance of the VGG21 model, the accuracy, precision, recall and F₁-Score of the ResNet50, Swin-Transformer, VGG19, VGG21, and VGG23 models were compared under the same conditions, and the confusion matrix and Grad-CAM (Gradient-Weighted Class Activation Mapping) were employed to evaluate the CK, waterlogging stress, and silicon fertilizer regulation model. [Results and Discussions] The performance comparison results showed that the VGG model performed better on multispectral image data recognition than the ResNet50 and Swin-Transformer model, and the VGG21 model demonstrated the most favorable overall performance. The comprehensive analysis of model recognition performance showed that the accuracy of the VGG21 model was more than 91% for silicon fertilizer regulation and waterlogging stress. And the precision, recall, and F₁-Score for silicon fertilizer regulation samples reached 96.42%, 89.53%, and 92.85%, respectively. When the silicon fertilizer regulation and stress recognition model was used for amino acid regulation sample identification, the accuracy, precision, recall, F₁-Score were comparable to the amino acid regulation and stress recognition model. A recognition model for CK, waterlogging stress, and silicon fertilizer regulation was constructed using VGG21, and the precision, recall, and F₁-Score for silicon fertilizer regulated samples achieved 95.13%, 88.71%, and 91.80%, respectively. The confusion matrix of the CK, waterlogging stress, and silicon fertilizer regulation model indicated that there was a high proportion of misclassifications between CK and waterlogging stress samples, resulting in a low accuracy of model, which only reached 77.45%. The heatmap analysis results of the CK, waterlogging stress, and silicon fertilizer regulation model indicated that after passing through the fourth convolutional block, the concentrated regions of the output features for misclassified CK and waterlogging stress samples became more dispersed, and even disappear. The CK and waterlogging stress samples with unfocused region heatmap accounted for a large proportion in the dataset, which led to poor recognition of these two types of samples by the regulation model. The comparison between recognition results of VGG21 model and the yield of regulation experiment showed that the silicon fertilizer had a better regulation effect, and the wheat regulated by silicon fertilizer had significant spectral and image features with high recognition accuracy. [Conclusions] The results demonstrated that UAV multispectral imaging technology and VGG21 model were feasible and useful for recognizing waterlogging stress and the regulatory effect of wheat. This technique could provide theory and technical support for integrated waterlogging management of wheat and for refined, intelligent wheat production management. However, the quantitative relationship between the identification results of regulation effect and yield was still unclear. In future research, it will be necessary to quantify the increased yield of regulation measures, and then use statistical methods to quantify the quantitative relationship between the yield-increasing regulation effect and identification indicators of recognition model.

County-Level Winter Wheat Yield Prediction Based on Phenological-Stage Progressive Multi-Source Data Fusion |

WANG Yi, XIONG Baowei, WANG Gang, SHAO Guomin, ZHANG Liyuan, LI Guang

2026, 8(4):  70-84.  doi:10.12133/j.smartag.SA202605002

Asbtract ( 118 )   HTML ( 10)   PDF (2784KB) ( 14 )  

Figures and Tables | References | Related Articles | Metrics

[Objective] Winter wheat yield formation is jointly affected by multiple factors, including canopy structure, growth status, photosynthetic function, and meteorological conditions. The contributions of remote sensing, meteorological data, and photosynthetic information to yield prediction differ among phenological stages. However, the influence of information accumulation at different phenological stages on county-level winter wheat yield estimation remains unclear, which restricts further improvement in early prediction capability and agricultural interpretability of yield prediction models. Therefore, a phenological-stage progressive multi-source data fusion method was proposed for county-level winter wheat yield prediction, to improve the prediction accuracy of county-level winter wheat yield, clarify the influence of information accumulation at different phenological stages on yield estimation performance, and provide methodological support for dynamic regional yield prediction and monitoring of key growth stages. [Methods] 100 major winter wheat-producing counties of Henan province were selected as the study area. Remote sensing variables, meteorological variables, and solar-induced chlorophyll fluorescence (SIF)-based photosynthetic variables from 2013 to 2022 were integrated to construct a county-level multi-source time-series feature dataset covering the period from the tillering stage to the maturity stage of winter wheat. Among these variables, remote sensing variables were used to characterize canopy structure and crop growth status, meteorological variables were used to reflect heat and water conditions during the growth period, and SIF variables were used to characterize changes in crop photosynthetic function. Based on the winter wheat planting distribution mask and county-level administrative boundaries, county-scale time-series features of each variable were extracted. According to the developmental sequence of tillering, overwintering, greening, jointing, heading, anthesis, and maturity stages, the model input window was progressively expanded to form input sequences under different phenological-stage accumulation conditions. For model construction, a convolutional neural network-long short-term memory-transformer (CNN-LSTM-Transformer) model was adopted. The CNN was used to extract local variation features between adjacent time steps, the LSTM was used to capture continuous temporal dependencies among stages, and the Transformer Encoder was used to model global associations among time steps in the input sequence. To evaluate model performance, the CNN-LSTM-Transformer model was compared with random forest (RF), Extreme Gradient Boosting (XGBoost), and CNN-LSTM models. In addition, a leave-one-year-out validation strategy was used to analyze the interannual generalization ability of the CNN-LSTM-Transformer model under different year conditions. [Results and Discussions] In the test set based on county-level sample partitioning, the CNN-LSTM-Transformer model outperformed the RF, XGBoost, and CNN-LSTM models and achieved the best yield estimation performance. The coefficient of determination (R²), root mean square error (RMSE), and mean absolute percentage error (MAPE) were 0.827, 588.25 kg/hm², and 8.02%, respectively. Compared with the CNN-LSTM model, the R² of the CNN-LSTM-Transformer model increased by 0.054, while RMSE and MAPE decreased by 85.69 kg/hm² and 1.54 percentage points, respectively. These results indicated that introducing the Transformer Encoder helped enhance the model's ability to represent global multi-source time-series information. The phenological-stage accumulation analysis showed that model prediction accuracy continuously improved as growth-stage information gradually increased. Prediction accuracy was relatively low at the tillering and overwintering stages, whereas model performance improved most obviously from the greening stage to the heading stage. This indicated that multi-source information during this period could more fully reflect rapid canopy development, enhanced photosynthetic capacity, and the yield formation process of winter wheat. Therefore, the period from greening to heading was identified as the key stage for the rapid improvement of county-level winter wheat yield prediction ability. The residual analysis across yield zones showed that the CNN-LSTM-Transformer model alleviated, to some extent, the overestimation in low-yield areas and the underestimation in high-yield areas. The leave-one-year-out validation results showed that the proposed method maintained good prediction stability across different test years. The spatial distribution results further showed that the predicted yield could effectively reflect the county-level spatial pattern of winter wheat yield in Henan province, with lower values in the western region and higher values in the central and eastern regions. [Conclusions] Overall, the proposed phenological-stage progressive multi-source data fusion method effectively integrated remote sensing, meteorological, and SIF-based photosynthetic information, improved the prediction accuracy of county-level winter wheat yield, and revealed the influence of information accumulation at different phenological stages on yield estimation performance. The CNN-LSTM-Transformer model collaboratively represented local variation features, stage-wise temporal dependencies, and global long-term associations, thereby effectively characterizing multi-source time-series dynamic changes during winter wheat yield formation. The results showed that the period from the greening stage to the heading stage was the key period for the rapid improvement of county-level winter wheat yield prediction ability. This study could provide a reference for early winter wheat yield estimation, key growth-stage monitoring, and regional grain production management.

Information Processing and Decision Making

Lightweight Detection Method for Grading Fresh Cut Dianthus caryophyllus L. Based on Flor-YOLO |

LI Chuanmeng, YANG Jie, ZHANG Xiaoyu

2026, 8(4):  85-99.  doi:10.12133/j.smartag.SA202512007

Asbtract ( 224 )   HTML ( 26)   PDF (14802KB) ( 16 )  

Figures and Tables | References | Related Articles | Metrics

[Objective] Carnation (Dianthus caryophyllus L.) is one of the most economically valuable cut flower crops worldwide. Postharvest openness is a key quality indicator influencing pricing, logistics tolerance, and shelf life. However, manual grading is inefficient and subjective due to dense petal overlap and complex edge structures. With the shift toward large-scale production and rising labor costs, accurate automated grading has become essential. Existing object detection models face a trade-off between computational efficiency and feature fidelity: High-precision architectures are computationally expensive for edge deployment, while lightweight models often lack sufficient feature representation. Additionally, conventional spatial downsampling introduces spectral aliasing, leading to the loss of high-frequency petal texture information and limiting the separability of adjacent openness grades. Therefore, a lightweight yet detail-preserving detection framework is required. To address this need, Flor-YOLO (Flower openness recognition You Only Look Once) was proposed by integrating frequency-domain perception with structural re-parameterization for efficient and accurate carnation openness grading. [Methods] Based on the YOLO11n baseline, the Flor-YOLO architecture was proposed with targeted improvements to the backbone, downsampling mechanism, and detection head. Backbone reconstruction: A lightweight LiteChimeraNet was constructed to enhance feature expression under limited computing power. A RepStem re-parameterization module was introduced at the input stage to establish an anti-aliasing mechanism via multi-branch training and single-path inference. Simultaneously, the C3k2_PConv module, utilizing partial convolution, was integrated to reduce memory access cost (MAC) and focus computation on petal foregrounds. Additionally, a RepNCSPELAN4_CAA module embedded with context anchor attention was incorporated in deep layers to capture long-range dependencies of the global flower topology. Frequency-domain downsampling: To mitigate texture aliasing and detail loss caused by spatial downsampling, a WaveletPool module was introduced. Utilizing the 2D discrete wavelet transform (2D-DWT), this module orthogonally decomposed feature maps into low- and high-frequency sub-bands, explicitly preserving high-frequency information in horizontal, vertical, and diagonal directions to alleviate spectral aliasing. Detection head optimization: A lightweight shared detail-enhanced detection head (SDL-Head) was designed. It reduced parameter redundancy through cross-scale weight sharing and incorporated detail-enhanced convolution (DEConv), fusing central and angular difference operators, to boost sensitivity to the geometric morphology of petal edges. Furthermore, a scale-adaptive layer combined with Intersection over Union (IoU)-aware soft labels was applied to improve multi-scale feature alignment. A dataset comprising 1 748 original images of "Hongkang" carnations was collected and expanded to 6 580 samples via hybrid data augmentation. The model was trained on an NVIDIA RTX 4060 GPU for 250 epochs using SGD optimization, and comparative evaluations were conducted against the YOLO series, NanoDet-m, and Hyper-YOLO-t. [Results and Discussions] Ablation studies and comparative experiments on the self-constructed dataset revealed significant performance gains. Ablation analysis: Reconstructing the backbone to LiteChimeraNet reduced floating point operations from 6.3 GFLOPs (baseline) to 1.5 GFLOPs, a decrease of 76.2%, while maintaining stable mean Average Precision (mAP@50), verifying its efficiency in removing background redundancy. Introducing WaveletPool significantly improved mAP@50 by 1.79 percentage points, confirming the critical role of explicitly preserving high-frequency components for serrated texture representation. Integrating SDL-Head further optimized feature alignment, increasing the recall rate to 94.47%. Overall performance: Flor-YOLO achieved a precision of 93.04%, recall of 94.47%, and mAP@50 of 96.10%. Compared to the YOLO11n baseline, these metrics improved by 3.52, 1.34, and 3.25 percentage points, respectively. Meanwhile, parameters and floating point operations were reduced by 51.2% to 1.26 M and 1.1 GFLOPs (82.5% reduction). Flor-YOLO exhibited distinct advantages over YOLOv5n, YOLOv8n, YOLOv9t, YOLOv10n, and YOLOv12n in accuracy, mAP, and inference speed. Mechanism analysis: Spectral energy statistics showed that high-frequency energy intensified with increasing openness grades, aligning with the visual characteristics of petal expansion and wrinkle formation, thus validating the discriminative value of high-frequency information. Grad-CAM++ visualizations further validated that the improved model stably focused on petal edges and flower centers, demonstrating superior robustness over the baseline in complex backgrounds. [Conclusions] By constructing the LiteChimeraNet backbone, incorporating frequency-domain downsampling, and designing a detail-enhanced head, the proposed model effectively enhances the representation of critical details such as petal edges and flower centers while maintaining extremely low computational costs. Comprehensively, Flor-YOLO achieves an optimal balance between accuracy, model size, and real-time performance, demonstrating strong potential for deployment on low-power mobile terminals and embedded sorting equipment. Furthermore, the proposed frequency-aware lightweight design paradigm provides a valuable reference for other agricultural vision tasks relying on subtle textural differences.

Cotton Maturity Detection Algorithm Based on Improved RT-DETR |

SHI Qimeng, WANG Jun, XU Xiaofeng, ZHANG Weiyi

2026, 8(4):  100-113.  doi:10.12133/j.smartag.SA202512013

Asbtract ( 115 )   HTML ( 16)   PDF (2739KB) ( 13 )  

Figures and Tables | References | Related Articles | Metrics

[Objective] Cotton maturity assessment is a vital task in precision agriculture, playing a crucial role in supporting timely irrigation, fertilization, and harvesting decisions. However, traditional monitoring approaches are time-consuming and labor-intensive, and current deep learning-based models often struggle to effectively recognize cotton bolls at varying maturity stages, especially in complex field environments with dense foliage, occlusion, and illumination changes. To address these challenges, a high-accuracy and lightweight computer vision model was proposed for cotton maturity detection. [Methods] An enhanced detection framework named cotton maturity-detection transformer (CM-DETR) was proposed, based on an improved real-time detection transformer (RT-DETR) architecture. CM-DETR incorporated three core architectural innovations that significantly improved both detection accuracy and computational efficiency. First, to construct a lightweight and efficient backbone, a novel feature extraction module named Re-parameterized Group Convolution Spatial Enhancement Lightweight Attention Network (RGCSPELAN) was introduced. This module integrated Progressive Convolution, which captured hierarchical and local contextual features, with Re-parameterized Convolution (RepConv), which reduced computational complexity during inference by transforming multi-branch structures into a single-path representation. The combination effectively enhanced the model's feature representation capabilities and gradient propagation while minimizing the number of parameters and FLOPs. Furthermore, RGCSPELAN was designed with a scalable architecture, allowing its computational capacity to be adjusted via a scaling factor. This ensured compatibility with both small and large models, facilitating flexible deployment across resource-constrained edge devices and high-performance systems alike. Second, to address the issue of small target feature loss, a new module termed Deep Robust Feature Downsampling (DRFD) was proposed. DRFD employed a multi-scale feature fusion strategy by integrating multiple downsampling branches (e.g., convolutional, cut-based, and max-pooling pathways). This design enabled the model to retain fine-grained spatial details while expanding its receptive field. Third, the original loss function in RT-DETR was replaced with Focaler-CIoU, and an adaptive regression optimization strategy integrating sample reweighting and geometric constraints was implemented to improve bounding box localization under complex conditions. [Results and Discussions] Experimental results demonstrated that CM-DETR achieved mAP50 and mAP50~95 scores of 80.8% and 51.1%, respectively, outperforming the baseline model by 3.7 and 1.8 percentage points. Meanwhile, CM-DETR reduced the parameter count and computational cost by 31.7% and 22.8%, respectively, indicating a favorable trade-off between detection accuracy and model efficiency. The incorporation of the DRFD module enhanced the model's sensitivity to small and subtly distinct features related to cotton maturity, improved robustness under diverse field conditions, and enabled more precise detection of cotton bolls at different growth stages. Moreover, the optimized regression strategy contributed to more stable bounding box prediction performance in scenarios involving occlusion, scale variation, and dense foliage. Overall, the proposed architectural improvements effectively strengthened feature representation capability while maintaining lightweight characteristics, thereby demonstrating practical applicability in real-time agricultural environments. [Conclusions] The proposed CM-DETR model provides an efficient and scalable solution for automated cotton maturity detection. By enhancing multi-stage feature recognition, improving small-target sensitivity, and reducing the demand on computational resources, CM-DETR serves as a reliable tool for intelligent decision-making in precision agriculture. Its practical deployment can support more accurate timing for irrigation, fertilization, and harvesting, thereby contributing to improved crop management and yield optimization.

MaMoNet-Based Detection Model for Moisture Content and Hardness of Fresh Corn Using Spatially Resolved Spectroscopy |

XU Min, ZHAO Xin, CHEN Yanping, ZHU Qibing, HUANG Min

2026, 8(4):  114-126.  doi:10.12133/j.smartag.SA202512025

Asbtract ( 127 )   HTML ( 9)   PDF (1701KB) ( 10 )  

Figures and Tables | References | Related Articles | Metrics

[Objective] Moisture content and kernel hardness are key indicators for evaluating the eating quality and harvest maturity of fresh corn. Accurate and simultaneous prediction of the two parameters is essential for quality grading and post-harvest management. However, conventional near-infrared spectroscopy (NIRS) methods are mostly based on single-point measurements, which are easily affected by husk shielding and kernel heterogeneity, resulting in limited prediction accuracy, especially for multi-attribute estimation. Spatially resolved spectroscopy, by acquiring spectral information from multiple measurement channels, can better reflect internal quality differences of corn kernels. Nevertheless, the high dimensionality and strong correlation of spatially resolved spectral data, together with the heterogeneity between different prediction tasks, pose significant challenges to traditional modeling approaches. Therefore, the purpose of this research was to develop an effective multi-task prediction model that can fully exploit spatially resolved spectral information while balancing feature sharing and task specificity for moisture content and kernel hardness prediction in husked fresh corn. [Methods] Husk-on fresh corn samples were collected across different maturity stages to ensure sufficient variability in moisture content and kernel hardness. A multi-channel visible-near infrared spatially resolved spectroscopy system was constructed to acquire spectral signals from four spatial measurement channels at different source-detector distances. Each corn ear was segmented into multiple positions, and spectral data were collected from different spatial locations to comprehensively capture internal quality information. In total, 500 valid spectral samples were obtained and randomly divided into training, validation, and test sets at a ratio of 7:1:2. Before modeling, the raw spectral data were preprocessed using z-score standardization to eliminate scale differences among channels and wavelengths and to improve numerical stability during training. The proposed MaMoNet (Mamba-MMoE Network) model is composed of three main modules: a one-dimensional convolutional neural network (1D-CNN) for local feature extraction, a Mamba-based sequence modeling module for global dependency learning, and a Multi-gate Mixture-of-Experts (MMoE) module for task-adaptive feature allocation. First, the 1D-CNN module was employed to extract low-level local spectral patterns and reduce feature redundancy. The input spectral data were processed by two successive one-dimensional convolutional layers and down sampling was then applied to compress the spectral length, resulting in a more compact representation and reducing the computational burden for subsequent sequence modeling. Next, the compressed spectral features were fed into the Mamba module to capture long-range dependencies along the spectral dimension. Mamba is a selective state space model that enables efficient modeling of long spectral sequences by dynamically updating hidden states, allowing global spectral evolution patterns to be effectively learned with linear computational complexity. Finally, the output features of the Mamba module were input into an MMoE module to support multi-task learning of moisture content and kernel hardness. Multiple expert networks were shared across tasks, and task-specific gating networks were used to adaptively weight expert outputs, enabling flexible feature sharing and task-specific representation learning. The task-specific features were then passed to individual regression heads to generate the final predictions for moisture content and kernel hardness. To comprehensively evaluate the effectiveness of the proposed approach, MaMoNet was compared with several representative multi-task convolutional neural network models. In addition, ablation experiments were conducted by selectively removing the Mamba module or the MMoE module to analyze their individual contributions to overall performance. [Results and Discussions] Experimental results demonstrated that the proposed MaMoNet model consistently outperformed all comparison models on the test set for both prediction tasks. For moisture content prediction, MaMoNet achieved a coefficient of determination (R²) of 0.91, a root mean square error (RMSE) of 4.98%, and a residual predictive deviation (RPD) of 3.40, indicating excellent predictive accuracy and robustness. For kernel hardness prediction, the corresponding R², RMSE, and RPD values reached 0.89, 3.46 N, and 3.06, respectively, which also surpassed those of the benchmark models. The ablation study further verified the rationality of the proposed model design. Removing both the Mamba and MMoE modules resulted in the poorest performance, whereas introducing either module individually led to noticeable improvements. The best prediction results were achieved when both modules were jointly employed, indicating that their combination was critical for achieving optimal multi-task prediction performance. These results indicated that moisture content and kernel hardness, although correlated, emphasize different spectral characteristics. The flexible feature-sharing mechanism enabled by MMoE allows the model to balance information sharing and task specificity, while the Mamba module ensures effective utilization of long-range spectral information. Together, they contribute to improved generalization performance under limited sample conditions. [Conclusions] This study proposes a MaMoNet model that integrates Mamba-based state space modeling with an MMoE-based multi-task learning strategy for simultaneous prediction of moisture content and kernel hardness in husked fresh corn using spatially resolved spectroscopy. The proposed approach effectively overcomes the limitations of conventional single-point spectral analysis and rigid parameter-sharing multi-task models. Experimental comparisons and ablation analyses confirm that MaMoNet achieves improved accuracy, robustness, and generalization capability. The results demonstrate the potential of the proposed framework for rapid and non-destructive quality assessment of fresh corn and provide useful methodological insights for multi-attribute prediction based on high-dimensional spectral data.

DC-YOLO: A Behavior Detection Model for Hu Sheep Addressing Occlusion and Illumination Variations |

LI Xiaxi, JI Ronghua, CHANG Hongrui, ZHANG Suoxiang

2026, 8(4):  127-147.  doi:10.12133/j.smartag.SA202601021

Asbtract ( 134 )   HTML ( 15)   PDF (4980KB) ( 7 )  

Figures and Tables | References | Related Articles | Metrics

[Objective] In the livestock industry, animal behavior serves as a critical indicator for evaluating physiological health and providing environmental early warning. To address the challenges of severe multi-object occlusion, complex illumination, and class imbalance in intensive farming, a detail-contextual attention-YOLO (DC-YOLO) model was developed for the 24/7 automatic detection of daily behaviors in housed Hu sheep, thereby facilitating precision management. [Methods] For severe multi-object occlusion, an adaptive detail-contextual attention (DCAttention) mechanism was introduced to construct the DCAC3K2 module, and bounding box similarity soft-NMS (BS-NMS) was adopted to prevent incorrect removal of occluded bounding boxes. For complex illumination, a Light-Encoder was pre-trained on illumination conditions and its parameters were transferred to DC-YOLO. For class imbalance, a comprehensive classification-quality focal loss was designed to adaptively increase loss weights for occluded and minority-class samples. The dataset, collected through 24/7 surveillance at Hu sheep farms, comprised 505 annotated images depicting 5 daily behaviors: drinking, eating, lying, licking, and standing. [Results and Discussions] DC-YOLO achieved mAP@50 of 91.4%, improving by 7.8 percentage points over YOLOv12. Moreover, DC-YOLO had 2.29 M parameters, an 8.74% reduction compared to YOLOv12. On CPU, the inference time of DC-YOLO was reduced to 115.5 ms and the frame rate increased to 8.50 f/s, corresponding to improvements of 33.2% and 48.9%, respectively. [Conclusions] Experimental results demonstrate that DC-YOLO effectively mitigates detection challenges caused by severe occlusion, complex illumination, and class imbalance while maintaining high inference efficiency. Consequently, it provides effective technical support for tracking and analyzing behaviors in intensive Hu sheep farming.

Temporal Action Localization of Mounting Behavior in Dairy Goats Based on Improved AdaTAD |

WANG Jiayuan, LI Qitong, LUO Yuantao, YANG Shuqin, WANG Zhenhua, NING Jifeng, WANG Meili

2026, 8(4):  148-163.  doi:10.12133/j.smartag.SA202601012

Asbtract ( 76 )   HTML ( 8)   PDF (2150KB) ( 5 )  

Figures and Tables | References | Related Articles | Metrics

[Objective] Accurate temporal localization of mounting behavior in dairy goats is important for intelligent reproductive management, as event frequency, onset time, and duration provide useful evidence for heat monitoring and mating decisions. Unlike simple behavior recognition, temporal localization in untrimmed videos enables fine-grained, time-resolved records for practical farm use. However, real-world mounting behavior is usually brief and sporadic, with few informative frames in long video streams. Moreover, weak discrimination from similar non-target interactions, together with occlusion, viewpoint variation, and background motion, often degrades boundary-aware representation learning and leads to unstable start-end localization. To enhance localization accuracy and stability while maintaining practical efficiency for deployment, an improved AdaTAD-based end-to-end temporal action localization approach was proposed for mounting behavior in dairy goats. [Methods] The proposed approach adopted AdaTAD as the baseline end-to-end temporal action localization framework and introduced two complementary improvements, explicit key-frame guidance and multi-scale motion modelling, while retaining the original detection head and post-processing pipeline for generating temporal action instances. First, visual prompt tuning (VPT) was incorporated to provide task-conditioned guidance to backbone feature extraction in a parameter-efficient manner. Specifically, a small number of learnable prompt tokens were inserted into the Transformer backbone with backbone parameters frozen. Through multi-head attention interactions between prompt tokens and patch tokens, the prompts steered attention towards mounting-relevant temporal regions, strengthened feature responses at critical frames and in boundary neighbourhoods, and improved the separability between brief target segments and abundant background frames. Second, a multi-scale motion adapter (MSMA) was introduced to model motion patterns at different temporal scales and improve robustness to diverse scene dynamics. MSMA employed parallel multi-scale temporal depthwise separable convolution branches to capture short-, mid-, and longer-range temporal variations, enhancing representations of subtle short-duration micro-actions as well as relatively complete action processes. Residual connections and nonlinear mappings further stabilised feature injection and gradient propagation, enabling multi-scale dynamics to be integrated into backbone features with limited additional optimisation burden. Overall, VPT focused on boundary-relevant attention guidance, whereas MSMA emphasised multi-scale temporal dynamics modelling; Together, they formed a complementary design within the end-to-end localization pipeline. [Results and Discussions] Comparative experiments showed that the proposed method achieves an average mAP (mean Average Precision@[0.3:0.1:0.7]) of 81.72%, improving upon the baseline AdaTAD by 5.00 percentage points, indicating that incorporating VPT and MSMA enhanced overall localization performance. At a temporal Intersection over Union (tIoU) threshold of 0.7, the proposed method attained 68.85%, exceeding AdaTAD by 4.06 percentage points, demonstrating that the performance gain was preserved under stricter temporal boundary-consistency criteria. Further comparisons with representative approaches, including TadTR, VSGN, AFSD, ActionFormer, TriDet, DyFADet, and Re2TAL, showed average mAP improvements of 38.82, 33.83, 25.29, 4.09, 2.83, 1.20, and 6.06 percentage points, respectively, demonstrating stronger overall competitiveness. In terms of efficiency, the model ran at 65.78 f/s with 27.941 million trainable parameters, indicating that the accuracy gains were achieved while maintaining a relatively low parameter overhead and practical runtime efficiency. Overall, task-guided prompting and multi-scale temporal modelling improved key temporal feature representations with limited parameter increments, thereby benefiting localization of short, sporadic behaviors. [Conclusions] By combining VPT for boundary-relevant attention guidance with an MSMA for multi-scale temporal dynamics modelling, the proposed AdaTAD-based end-to-end temporal action localization method improves localization accuracy and maintains stable advantages under stricter boundary-consistency requirements, while preserving practical inference efficiency. The method could provide temporal information for reproductive behavior monitoring and decision support, and offer a feasible basis for building individual-level, time-resolved management systems in real farming environments.

Multi-Scenario Simulation of Land Use and Spatiotemporal Evolution of Carbon Storage in Plain Agricultural Region Based on The Coupled PLUS-InVEST Model |

WANG Gaocheng, LIU Jian, LI Shasha, ZHANG Tingting, WANG Ailing

2026, 8(4):  164-179.  doi:10.12133/j.smartag.SA202512028

Asbtract ( 114 )   HTML ( 4)   PDF (4211KB) ( 2 )  

Figures and Tables | References | Related Articles | Metrics

[Objective] Carbon storage is a key indicator for measuring ecosystem functionality, with significant differences in carbon sequestration capacity among various land use types. Changes in land use directly lead to variations in terrestrial ecosystem carbon storage. Therefore, an in-depth analysis and prediction of the spatiotemporal distribution patterns of land use change and carbon storage can provide a scientific basis for achieving carbon sequestration targets and optimizing land use structures. [Methods] Taking Gaotang county, a typical plain agricultural county in Shandong province as the study area, based on land use types extracted from remote sensing data from 2009 to 2023, the spatiotemporal evolution patterns of land use were first analyzed. The InVEST model was then employed to estimate carbon storage and identify the spatial distribution patterns of carbon storage across different land categories. Driving factors were selected from the perspectives of natural conditions, socio-economic development, and locational conditions. Five scenarios: natural development, cultivated land protection, urban development, ecological protection, and sustainable development, were established. A coupled PLUS-InVEST framework was constructed. First, the PLUS model was used to simulate land use patterns under five scenarios for 2035. Then, the simulated land use maps were input into the InVEST model to estimate carbon storage and to compare changes across different scenarios. This coupling enables an integrated analysis linking policy scenarios, spatial land use patterns, and carbon storage responses. [Results and Discussions] From 2009 to 2023, significant changes occurred in land use types, with the most notable transformation being the conversion from cultivated land to forest land. Overall, cultivated land area fluctuated and decreased by 3 184.10 hm²; forest land in the southwestern region increased by 1 988.74 hm²; and construction land in the county center decreased by 109.89 hm². Total carbon storage increased by 1.14×10⁵ t. Carbon storage was significantly correlated with the spatial distribution of various land types, exhibiting a pattern of higher values in the southwest and lower values in the northeast. Carbon storage in the county center was relatively low, while the southwestern region, rich in forest land resources, had the highest carbon storage. Under the natural development scenario, all land use types except cultivated land showed a decrease in area. Under the cultivated land protection scenario, cultivated land area reached 69 116.78 hm², an increase of 8.23% compared to 2023, while forest land experienced the largest decrease. Under the urban development scenario, both cultivated land and construction land increased, whereas all other land types decreased. Under the ecological protection scenario, cultivated land, forest land, and grassland increased, while all other land types decreased. Under the sustainable development scenario, cultivated land area was slightly smaller than that under the cultivated land protection scenario, while forest land and grassland areas were similar to those under the ecological protection scenario. The spatial distribution characteristics of carbon storage were similar across all scenarios, consistently showing a pattern of lower values in the central area and higher values in the surrounding areas. Under the urban development scenario, total carbon storage was the lowest among all scenarios, decreasing by 3.01×10⁵ t compared with the 2023 level. [Conclusions] This research reveals that construction land expansion is the main cause of carbon storage loss, while ecological restoration measures can effectively increase carbon storage. The spatial distribution of different land categories remains relatively stable across scenarios, with carbon storage exhibiting a pattern of lower values in the central urban area and higher values on the periphery. Strategies pursuing only economic development or focusing solely on cultivated land protection have inherent limitations. The sustainable development scenario can better balance cultivated land protection and ecological conservation, dynamically adjust the relationship between the two, and achieve their long-term coordinated development. This research could provide a reference for achieving carbon peak, carbon neutrality, and sustainable development goals in plain agricultural regions.

Machine Learning Reveals the Driving Effects of Soil Nutrients on Maize Yield Formation: A Case Study of Yunnan, China |

SUN Jiaze, QU Mingshan, YANG Jizhong, LOU Chuixin, LI Guangwei, ZHANG Zhonglili

2026, 8(4):  180-191.  doi:10.12133/j.smartag.SA202508029

Asbtract ( 79 )   HTML ( 3)   PDF (2260KB) ( 6 )  

Figures and Tables | References | Related Articles | Metrics

[Objective] Maize production in Yunnan‑representative mountainous regions is strongly shaped by rugged terrain, fragmented cropland, and steep vertical climatic gradients, which drive pronounced spatial heterogeneity in soil nutrients and yield. Unlike homogeneous plains, yield variation here stems from the combined effects of soil fertility, climate, and topography rather than any single nutrient factor, making traditional statistics or isolated soil indicators inadequate. To this end, an interpretable machine learning framework integrating soil nutrients and spatial information was developed to enhance maize yield prediction and identify the dominant drivers of yield disparities across sub‑regions of such landscapes. [Methods] Soil testing and formula fertilization data from Yunnan province during 2005—2020 were combined with 5 km gridded maize yield data. Four soil nutrient indicators were selected, including soil organic matter, total nitrogen, available phosphorus, and available potassium. Longitude and latitude were also used as input variables. In this study, these two variables were not treated simply as location labels. They were included because spatial position can partly reflect regional environmental gradients that are difficult to describe using soil nutrient data alone, such as differences in elevation, heat and moisture conditions, terrain fragmentation, and cropping background. Based on these variables, particle swarm optimization was used to tune the main hyperparameters of the XGBoost model, forming a PSO-XGBoost yield prediction model. The optimized model was compared with LightGBM, CatBoost, random forest, support vector regression, and the original XGBoost model. Considering that nearby samples may share similar environmental conditions, random partitioning alone may overestimate model performance. Therefore, spatial block cross-validation was further used to examine the model's ability to predict yield under spatial separation. In addition, longitude and latitude were removed in an ablation experiment to test the contribution of spatial background information. Finally, SHAP was used to interpret the trained model, quantify the contribution of each variable, analyze the response of yield prediction to soil nutrients and spatial factors, and identify the dominant limiting factors in different maize-growing areas. [Results and Discussions] PSO-XGBoost showed the best performance among the tested models. Under random data partitioning, the model achieved an R² of 0.91, indicating that it could effectively capture the relationship among soil nutrients, spatial background, and maize yield. Compared with LightGBM, CatBoost, and the unoptimized XGBoost model, PSO-XGBoost gave more accurate and stable predictions. This result suggested that particle swarm optimization improved the parameter configuration of XGBoost and made it more suitable for yield prediction in complex mountainous regions. Under spatial block cross-validation, the R² decreased to 0.847. Although this value was lower than that obtained from random partitioning, it still showed good predictive performance under a stricter validation strategy. This also indicated that the model did not rely only on the similarity between neighboring samples, but retained a certain ability to predict yield in spatially separated areas. The ablation experiment further confirmed the importance of spatial information. After longitude and latitude were removed, the model R² dropped sharply to 0.698. This decline showed that maize yield patterns in Yunnan cannot be well explained by soil nutrient indicators alone. Spatial variables probably contain information related to elevation-induced climatic differences, regional hydrothermal conditions, terrain fragmentation, and differences in the cropping environment. The SHAP results led to a similar conclusion. Longitude and latitude had relatively large contributions, suggesting that regional environmental gradients played an important role in yield formation. Among the soil nutrient variables, soil organic matter, total nitrogen, available phosphorus, and available potassium were all related to maize yield, but their roles differed across Yunnan. In some areas, yield was more closely associated with organic matter and nitrogen supply, while in others phosphorus or potassium appeared to be more limiting. This spatial difference suggested that the main constraints on maize production were not the same across the province. For this reason, a province-wide fertilization scheme would be difficult to apply effectively. Nutrient management in Yunnan should instead be adjusted according to local soil conditions and the environmental background of different mountainous regions. [Conclusions] Maize yield in Yunnan province is affected by both soil nutrient conditions and broader spatial environmental factors. Soil fertility is still an important basis for yield formation in mountainous farmland, but it does not explain the spatial yield differences on its own. By combining PSO-XGBoost with SHAP, this study provides a way to improve yield prediction while also tracing the main factors related to regional yield variation. The framework can be used to identify areas where nutrient constraints are more likely to occur, as well as areas where terrain, climate background, or other regional environmental factors may have a stronger influence. These results can support more targeted fertilization, cultivated land quality improvement, and sustainable maize production in the mountainous agricultural areas of Yunnan.

Estimation of Citrus Transpiration in a Savanna Valley Based on Feature Selection and Optimization Algorithms |

WU Mingqing, LI Weikang, WANG Jing, LI Jing, PEI Rentao, HUANG Haitao, QIU Huanghuang, LEI Chujing, ZHAO Duo, HUANG Jingtao, GAO Zhiyong

2026, 8(4):  192-203.  doi:10.12133/j.smartag.SA202511007

Asbtract ( 5 )   HTML ( 1)   PDF (2662KB) ( 0 )  

Figures and Tables | References | Related Articles | Metrics

[Objective] As the core physiological process determining crop water demand, transpiration profoundly influences irrigation-district water regulation and allocation strategies. During the dry season, water stress is particularly acute in the Yunnan savanna region. For smart, sensor-driven irrigation management, reliable orchard-scale transpiration estimation is essential for real-time irrigation scheduling and water allocation. However, traditional mechanistic models face challenges such as difficulties in parameter acquisition, high sensitivity to specific parameters, and limited applicability in complex terrains. To address these issues in the development of smart irrigation districts, a high-precision citrus transpiration estimation model was constructed with low parameter dependence. [Methods] A citrus orchard with 10-year-old trees in the savanna region was selected as the study site. Three years of stem sap flow observations and synchronous environmental-factor data were collected. Sap flow was accumulated to the daily scale and used to estimate field-scale transpiration. A Random Forest–based wrapper feature selection method was applied to identify an optimal subset of key predictors from 16 original environmental parameters and the day of year (DOY). Specifically, the wrapper method iteratively removed the variable with the lowest feature importance and rebuilt the model, continuing until only one variable remained. During this process, model performance under different numbers of variables was recorded for each iteration, and the best feature subset was determined based on performance. In addition, a heatmap was used to quantify linear correlations among variables to better understand redundant parameters removed by the model. After identifying the optimal feature combination, the built-in feature importance method of Random Forest was used for importance evaluation. Based on these parameters, extreme gradient boosting (XGB), support vector regression (SVR), and random forest (RF) models were established. Particle swarm optimization (PSO), bayesian optimization (BO), and the asynchronous successive halving algorithm (ASHA) were introduced for hyperparameter tuning, resulting in 12 simulation models in total. The dataset was split into 80% for training and 20% for validation. During training, 10-fold cross-validation was adopted, and mean absolute error (MAE) was used as the fitness function. Model performance was comprehensively evaluated using the coefficient of determination (R²), root mean square error (RMSE), MAE, and uncertainty metrics. [Results and Discussions] The RF wrapper feature selection reduced the 16 meteorological input parameters to four: actual vapor pressure (e_a), soil water content (VWC), wind speed at 2 m (u₂), and mean air temperature (T_a). In addition, the DOY was readily obtainable and was identified as the most critical feature, effectively characterizing the unique seasonal climatic variations of the Savanna Valley and citrus phenological characteristics; meanwhile, e_a, VWC, T_a, and u₂ mainly characterize short-term atmospheric evaporative demand and soil moisture constraints. While substantially reducing data requirements, the proposed method still achieved high prediction accuracy. Among all model combinations, XGB-ASHA performed best, reaching R²=0.91 and RMSE=0.19 mm/d for daily-scale transpiration prediction. Hyperparameter optimization improved model robustness, further reducing RMSE and MAE compared with non-optimized baselines. Compared with commonly used mechanistic models in the region, the proposed machine-learning approach does not require complex parameter calibration, is less sensitive to input errors, and shows better application feasibility. Uncertainty evaluation further indicates that this framework can provide not only accurate point predictions but also reliable confidence information for operational decision-making. [Conclusions] The proposed model requires monitoring only four environmental parameters to achieve high-accuracy, operational daily-scale transpiration estimation for citrus orchards in the savanna region. The optimal configuration, XGB-ASHA, is suitable for embedding into smart irrigation systems, supporting water-saving irrigation scheduling and irrigation-district-scale water management. Future work will integrate multi-source data and explore more advanced learning models to further improve transferability and generalization across orchards and years.

Multi-Scale Heterogeneous Feature Synergistic Model for Cotton Leaf Disease Detection |

SHEN Xueli, ZHANG Yue, JIN Haibo, ZHANG Xuxu

2026, 8(4):  204-216.  doi:10.12133/j.smartag.SA202601027

Asbtract ( 109 )   HTML ( 10)   PDF (2420KB) ( 3 )  

Figures and Tables | References | Related Articles | Metrics

[Objective] Detecting cotton leaf diseases in natural field environments is challenging due to diverse image interferences, variable leaf‑spot sizes, and the demand for fast computation on mobile phones and other small‑scale devices. Nevertheless, existing lightweight models cannot well balance detection accuracy and computational efficiency, especially for detecting small lesions and suppressing noise surrounding leaves. To address these challenges, MHSF‑DETR (Multi‑Scale Heterogeneous Synergistic Feature DETR), an improved detection model based on the RT‑DETR framework, is proposed. It aims to achieve high‑precision, low‑power diagnosis in complex agricultural scenarios. [Methods] The primary innovation of this study consisted of the complete reconfiguration of the feature extraction and fusion architectures. Firstly, a hierarchical context-selective perception network (HCSP-Net) was constructed as the backbone to replace conventional architectures. This backbone employed a differentiated processing strategy tailored to the depth of the feature maps: In the early parts of the process where the features were simple, it used something called micro-macro spatial context attention (M²-SCA). This module used a channel semantic filter then a dual stream spatial perception structure to actively capture high frequency textures of micro-lesions and preserve macro semantics so that fine details were not lost when downsampled. At the deep feature stage, a competitive selection fusion (CSF) module was added. Unlike the traditional static summing approach, CSF created a dynamic competition arbitration system that flexibly balances local importance versus overall coherence via soft competition gates, making the semantics sharper and filtering away irrelevant background noise. Secondly, to tackle the spatial and semantic misalignment that was commonly seen in cross-level feature fusions, a learnable weighted context fusion (LWC-Fusion) module was created inside the neck network. This module used global amplitude dynamic weighting to learn autonomously the best blending ratios, so that deep semantic features were aligned precisely with shallow geometry. Moreover, to solve the problem of artifacts appearing at irregular leaf boundaries caused by traditional zero-padding convolutions, an edge-aware reconstruction mechanism (EARM) was proposed. By using edge-refined convolution (ER-Conv) and the edge-refined convolution C3 module (ER-ConvC3), which integrated reflection padding and partial convolution techniques, the model successfully curtailed invalid edge noise and diminished computational redundancy without sacrificing the geometric continuity of features. [Results and Discussions] Empirical benchmarks demonstrated that the proposed MHSF-DETR achieved a superior balance between detection performance and computational efficiency. Compared to the RT-DETR-R18 baseline, MHSF-DETR yielded a significant 3.2 percentage points increase in mean average precision (mAP), while simultaneously reducing parameters by 22.42% and GFLOPs by 13.29%. When benchmarked against mainstream detectors, MHSF-DETR consistently outperformed models such as YOLOv5m, YOLOv10m, and RT-DETR-R50. Although YOLOv8m maintained a marginal mAP50 lead in specific scenarios, its exorbitant computational overhead rendered it less practical for real-time deployment compared to MHSF-DETR. It successfully matched the lean efficiency of YOLOv10m but excelled in detection accuracy. Furthermore, extensive ablation studies confirmed that these performance gains stemmed from the structural synergy among the HCSP-Net backbone, the LWC-Fusion neck, and specialized reconstruction modules, rather than isolated component upgrades. These results validated the effectiveness of MHSF-DETR design in optimizing feature extraction and fusion, offering a highly efficient solution for resource-constrained object detection tasks. [Conclusions] MHSF‑DETR addresses the long‑standing accuracy‑efficiency trade‑off in cotton disease monitoring. By integrating hierarchical perception, adaptive fusion and edge refinement, the model mitigates scale disparity and resource‑constraint challenges. It provides a feasible lightweight template for real‑time diagnosis on agricultural edge devices to support deployment in smart‑farming systems. Future work will expand validation to other plant organs (bolls, stems) and conduct rigorous embedded‑hardware field tests to assess real‑world robustness.

LiteFocus-Net: A Three-Point Lightweight Enhancement Framework for Small-Target Detection of Corn Leaf Diseases and Pests |

YANG Yu, ZHANG Yibo, MAO Bo, ZHANG Lei

2026, 8(4):  217-237.  doi:10.12133/j.smartag.SA202605017

Asbtract ( 95 )   HTML ( 19)   PDF (2562KB) ( 20 )  

Figures and Tables | References | Related Articles | Metrics

[Objective] Early detection of corn leaf diseases and pests is important for field monitoring, precision pesticide application, and yield protection. In practical field environments, early lesions, insect bodies, eggs, and spore clusters often occupy only a very small part of the image. Their visual appearance is affected by weak texture, irregular shape, and complex backgrounds, making the visual boundary between disease targets and normal leaf tissue less distinguishable. When lightweight detectors based on the YOLO series are applied, fixed convolution kernels lack flexibility for targets with large scale differences. After repeated downsampling, small objects lose representation in deep feature maps, while larger targets dominate the regression loss. These factors make it difficult to balance detection accuracy and deployment efficiency on edge devices. To improve the recognition capability for small targets under complex field conditions, LiteFocus-Net, a lightweight enhancement framework, was developed based on YOLOv11n. [Methods] The framework improved the baseline model from three closely related aspects: adaptive backbone feature extraction, deep feature detail recovery, and scale-aware regression supervision. In the backbone, an Adaptive Kernel Lightweight Block AKL-Block was designed to replace part of the original feature extraction structure. Instead of computing multiple convolution branches simultaneously during inference, AKL-Block used global average pooling and a lightweight gating module to estimate the selection probabilities of three candidate depthwise separable convolution kernels with different receptive fields. A Gumbel-Softmax strategy was used during training to keep the kernel selection process differentiable, while only the kernel with the highest selection probability is activated during inference. In this way, the model could adjust its effective receptive field according to the input feature distribution without introducing the redundant computation usually caused by parallel multi-branch structures. To alleviate the loss of small-object details in deep layers, a Feature Decomposition and Reconstruction module (FDR) was introduced after the P5 feature layer. The deep feature was divided into a structure branch and a detail branch. The structure branch was upsampled by bilinear interpolation to preserve global semantic information, such as leaf shape, lesion distribution, and large disease regions. The detail branch used a lightweight pixel-shuffle reconstruction operation to enhance local responses related to lesion edges, insect contours, and spore textures. The reconstructed feature was then fused with the corresponding P4 feature in the neck network, allowing the model to reuse deep semantic information while selectively strengthening detail-sensitive responses. For the regression loss, a Scale-Aware Gradient Boosting Loss, termed SAGB-Loss, was constructed to increase the training contribution of small targets. The loss combined feature-level weighting and target-area-aware weighting. The feature-level term assigned larger weights to shallow layers that are more closely related to small-object detection, while the area-aware term used a continuous exponential function to smoothly enhance the regression gradients of small targets. Experiments were carried out on a field corn leaf image dataset containing 10 324 images and eight categories of disease and pest targets. The dataset was divided into training, validation, and test subsets at a ratio of 6:2:2. [Results and Discussions] LiteFocus-Net achieved 74.97% mean average precision (mAP) at an Intersection over Union (IoU) threshold of 0.5, with 2.52 million parameters and 5.3 GFLOPs. Compared with the YOLOv11n baseline, the average precision for small objects (AP_s) increased from 28.1% to 32.77%, giving an improvement of 4.7 percentage points, while the computational cost decreased by 15.9%. Ablation experiments confirmed the contribution of each component. AKL-Block reduced the overall computational cost and improved small-target accuracy, FDR further enhanced detail-sensitive features from deep layers, and SAGB-Loss increased the regression contribution of small targets without causing an obvious decline in medium- and large-object detection. Comparisons with Inception-style and selective-kernel multi-scale modules showed that AKL-Block achieved a more favorable accuracy-complexity trade-off under lightweight deployment constraints. Comparisons with full-channel pixel-shuffle reconstruction also indicated that FDR provided a practical balance between detail recovery and additional computational cost. Deployment tests on RK3588 and Jetson Orin NX further showed that LiteFocus-Net maintained real-time inference capability after INT8 quantization, which indicated its potential for field inspection and agricultural edge intelligence applications. [Conclusions] The results suggest that LiteFocus-Net improves small-target detection of corn leaf diseases and pests without relying on a larger model scale. Future work will focus on expanding cross-region, cross-variety, and cross-growth-stage samples, and on exploring region-level dynamic kernel selection, video-based continuous detection, and weakly supervised annotation strategies to improve robustness in long-term field deployment.

Digital Economy

Trusted Data Space for the Agricultural Industry Chain: Theoretical Framework, Operating Mechanism, and Implementation Path |

ZHANG Xin, CHEN Mingyang, ZHAO Zhiyao, CHI Cheng, WANG Xiaoyi, XU Jiping

2026, 8(4):  238-254.  doi:10.12133/j.smartag.SA202605005

Asbtract ( 156 )   HTML ( 6)   PDF (2852KB) ( 25 )  

Figures and Tables | References | Related Articles | Metrics

[Objective] Data across the agricultural industrial chain are generated throughout production, processing, storage, logistics, sales and supervision. Such data feature dispersed stakeholders, long‑link processes, multi‑source heterogeneity, and high privacy‑ and business‑information sensitivity. Nevertheless, current agricultural big‑data platforms and standalone technologies including blockchain‑enabled traceability, privacy‑preserving computation and federated learning cannot systematically resolve bottlenecks in cross‑entity data circulation, such as data ownership confirmation, controlled utilization, audit tracing and value‑oriented collaboration. Accordingly, this paper proposes a trusted data‑space framework tailored for agricultural production, circulation, supervision and service collaboration scenarios to satisfy requirements for trusted cross‑entity data circulation. It elaborates the framework's layered architecture, core operational mechanisms and implementation paths, offering systematic references for agricultural data sovereignty protection, controlled data sharing, audit tracing and value collaboration. [Methods] Based on the national standard Technical Architecture of Trusted Data Space and the characteristics of the agricultural industry chain, including multiple stakeholders, long-chain processes, limited computing capacity at edge nodes, and data heterogeneity, a domain-adaptation approach was adopted to construct the overall framework and analyze its operating mechanisms. The framework design focused on semantic interoperability, connector-based controlled interaction, privacy-preserving computation, and blockchain-based evidence preservation. A self-developed integrated service platform was then used for preliminary scenario-based analysis, combining about 120 000 food safety sampling and monitoring records from a city during 2023－2025 with the business workflow of an organic farm. The analysis covered standardized data access and governance, risk profiling, on-chain evidence preservation, trusted traceability, permission control, abnormal request interception, and audit tracing. [Results and Discussions] A four-layer architecture was established, consisting of an infrastructure and data resource layer, a trusted data space core layer, a data capability support layer, and an agricultural industry chain application ecosystem layer. The operating mechanism formed a closed loop jointly driven by deep semantic interoperability, connector-based controlled interaction, value co-creation through privacy-preserving computation, and dynamic trust supported by blockchain and smart contracts. Under this framework, ontology models, metadata, and knowledge graphs supported concept alignment, structural mapping, and contextual disambiguation; connectors, digital contracts, and usage control policies constrained usage purposes, invocation frequency, field scope, output forms, and prohibited behaviors; privacy-preserving computation and federated learning enabled cross-entity collaborative analysis without centralizing raw plaintext data; and blockchain recorded contract hashes, log hashes, result digests, and abnormal interception records for auditable tracing. The platform-based scenario description showed that the framework could support standardized data access, risk profiling, on-chain evidence preservation, role-based permission control, abnormal operation interception, and audit tracing in agricultural and food safety risk governance scenarios. In the organic-farm workflow, the mechanism was reflected by keeping raw data off-chain, storing key digests on-chain, controlling data use within authorized environments, and retaining auditable records of the process. Compared with conventional centralized agricultural data platforms, this framework emphasized physical distribution with logical integration and extends security protection from access control to continuous usage control after cross-entity interaction. The four-dimensional implementation pathway further indicated that institutional rules, lightweight connectors and privacy-preserving components, specialized data intermediaries, and progressive pilot deployment should advance in coordination. [Conclusions] The proposed trusted data space framework provides a systematic approach to controlled data circulation, data sovereignty protection, auditability, and value collaboration among multiple stakeholders in the agricultural industry chain without requiring centralized aggregation of raw plaintext data. It can serve as a reference for the circulation of agricultural data elements and the collaborative transformation of the agricultural industry chain, while providing a basis for integration with the national data infrastructure system.

Vulnerability of the Ecological Agricultural Product Industry Chain under Digital Technology Empowerment: Archetype Analysis and Resilience Governance |

WANG Cuixia, HU Tao, LI Yaqin, DING Xiong

2026, 8(4):  255-267.  doi:10.12133/j.smartag.SA202604024

Asbtract ( 54 )   HTML ( 1)   PDF (2959KB) ( 0 )  

Figures and Tables | References | Related Articles | Metrics

[Objective] The ecological agricultural product industry chain is subject to deep-seated vulnerabilities during scale expansion, quality improvement, and premium growth, yet its structural roots remain inadequately understood. Although digital technologies are expected to relieve these vulnerabilities, recent evidence suggests that digitalization may also induce new suppression loops, resulting in smallholder exclusion, profit squeeze, and homogenized competition. Existing studies have largely addressed this subject from isolated perspectives such as brand governance, industrial chain resilience, or policy evaluation, without offering a systematic diagnosis of the evolutionary mechanism of vulnerability. The aim of this study is to identify the operational logic of growth ceiling archetypes under digital empowerment, exploring how digital technologies interact with feedback structures, and proposing resilience governance strategies that move beyond "technological fixes". [Methods] The Limits to Growth archetype was employed as the analytical framework. Variable selection followed a three-stage procedure integrating theoretical deduction, bibliometric analysis, and case validation. First, constraining factors were identified across multiple dimensions, including natural resource endowment, ecological carrying capacity, market saturation, and trust thresholds, grounded on the theoretical proposition that growth processes endogenously activate restraining forces. Second, a systematic literature search was conducted using search terms such as ecological agriculture, agricultural product industry chain, and system dynamics. Candidate variables that appeared with high frequency in relevant empirical studies were screened, yielding an initial pool of over 20 variables. Third, field investigations of typical cases such as Gannan navel orange, Wuchang rice, and Hengzhou digital jasmine were performed to test, consolidate, and refine the initial pool, ultimately producing 14 core variables. Causal linkages among these variables were established upon three forms of evidence, namely theoretical logic derived from established economic principles, literature evidence drawn from prior empirical findings, and case facts observed during field investigations. Loop polarities were determined by the parity of negative causal chains, where even numbers indicate reinforcing loops (R) and odd numbers indicating balancing loops (B). On this basis, and in combination with field data from the typical cases, four reinforcing loops that drive growth and seven balancing loops that constrain growth were identified. The moderating effects of three categories of digital tools, namely digital agricultural technology platforms, quality traceability systems, and e-commerce platforms, on these loops were then separately assessed. [Results and Discussions] The results indicated that the sustained growth of the ecological agricultural product industry chain was underpinned by four positive feedback loops: the scale-income loop, the quality-demand loop, the premium-income loop, and the quality-reputation loop. At the same time, however, the growth process activated seven negative feedback loops that imposed constraints from multiple directions, including cost erosion, quality dispersion, market saturation, homogenized competition, low-price substitute diversion, trust erosion, and reputation damage—thus locking the system into three types of Limits to Growth dilemmas relating to scale expansion, quality improvement, and premium growth. Digital technology empowerment enhanced the operational efficiency of the industrial chain by reinforcing the positive loops and weakening the negative ones, yet it failed to remove the structural roots of the growth ceiling. More importantly, digitalization unexpectedly triggered three new suppression loops, namely the smallholder exclusion loop, driven by digital and certification barriers, illustrated by the Hengzhou digital jasmine case where elderly flower farmers were marginalized due to the digital divide; the profit squeeze loop, driven by platform cost transference, as shown in the Shanghai Hema village cooperative case, where high equipment costs and loss rates eroded profit margins; and the homogenized competition loop, driven by standardization orientation, as evidenced in the Gannan navel orange case where farmers abandoned flavor-differentiated production practices to comply with platform specifications. These three loops respectively weakened the operational foundations of the scale-income, premium-income, and quality-reputation loops, thereby shifting the locus of vulnerability from traditional constraints to digitally-induced risks. These findings revealed a dual effect of digital technology empowerment, as mitigating certain existing constraints, digital technologies may simultaneously generate new vulnerabilities through digital divides, cost transference, and standardization pressures. The essential task for resilience governance, therefore, lay in identifying and intervening in the dominant balancing loops that constrains growth, rather than relying exclusively on technological inputs. [Conclusions] First, the three limits to growth archetypes, namely scale expansion, quality improvement, and premium growth, constitute the shared structural roots of vulnerability in the ecological agricultural product industry chain. Second, digital technology empowerment cannot eradicate these structural roots and may give rise to new vulnerabilities. Third, enhancing industrial chain resilience should be grounded in leverage point interventions derived from the system's feedback structure. Corresponding governance strategies for the three types of newly identified risks, namely smallholder exclusion, profit squeeze, and homogenized competition, should include lowering technological entry barriers, establishing benefit-sharing mechanisms, and strengthening differentiated certification and geographical indication protection. Overall, this study provides a system dynamics-based analytical tool and policy leverage points for resilience governance of the ecological agricultural product industry chain in the context of digital technology empowerment.