Table of Content

30 November 2025, Volume 7 Issue 6

Previous Issue   

Special Issue--Remote Sensing + AI Empowering the Modernization of Agriculture and Rural Areas

Current Status and Trends of Application Scenarios and Industrial Development in the Agricultural Low-Altitude Economy |

HE Yong, DAI Fushuang, ZHU Jiangpeng, HE Liwen, WANG Yueying

2025, 7(6):  1-17.  doi:10.12133/j.smartag.SA202507014

Asbtract ( 315 )   HTML ( 12)   PDF (2623KB) ( 25 )  

Figures and Tables | References | Related Articles | Metrics

[Significance] The agricultural low-altitude economy has rapidly emerged as a transformative force in rural modernization, catalyzed by advancements in intelligent technologies and gradual improvements in low-altitude airspace governance. It encompasses a wide array of applications where aerial platforms—particularly unmanned aerial vehicles (UAVs)—play a pivotal role in reshaping agricultural production, environmental monitoring, and rural management systems. Systematically analyzing the development trajectory and implementation pathways of agricultural low-altitude economy holds considerable significance for promoting cross-sector integration, accelerating policy and technological innovation, and enabling the widespread adoption of intelligent solutions across agriculture-related industries. [Progress] This review draws upon an extensive body of literature and applies keyword clustering analysis to systematically explore the development of the agricultural low-altitude economy. Based on 13 570 peer-reviewed articles retrieved from the Web of Science database between 2000 and 2024, the study reveals a rapid growth trajectory, particularly since 2011, driven by technological breakthroughs in UAVs, sensors, and intelligent analytics. Five representative application domains were identified: smart farming, animal husbandry, forestry, fisheries, and rural governance. Through the integration of bibliometric tools and structured keyword combinations, the study captures both the evolution of research focus and the expansion of technical capabilities. The results demonstrate that agricultural low-altitude economy has progressed from early-stage feasibility validation toward large-scale, multi-functional applications. At the same time, it has catalyzed the development of an emerging industrial framework encompassing equipment manufacturing, aerial service provision, operational support systems, and talent development. These trends highlight the growing maturity and strategic relevance of agricultural low-altitude economy as a technological enabler for modern agriculture. In the context of smart farming, low-altitude technologies are extensively utilized for precision sowing, variable-rate fertilization, real-time crop health monitoring, and pest and disease detection. UAV-based remote sensing facilitates the creation of high-resolution field maps and spatially explicit data layers that support data-driven and site-specific decision-making in modern agricultural management. In smart livestock systems, drones are employed for livestock monitoring, early disease detection, and fence-line inspections, particularly in large, remote pasture areas with limited ground accessibility. Smart forestry applications include forest fire early warning, forest inventory updates, and dynamic monitoring of ecological changes, enabled by low-altitude hyperspectral, LiDAR, and thermal imaging technologies. For smart fisheries, UAVs and amphibious drones support water quality sensing, pond surveillance, and feeding behavior analysis, thereby enhancing aquaculture productivity, animal welfare, and environmental sustainability. In the broader context of smart rural governance, agricultural low-altitude economy technologies assist with infrastructure inspections, land use monitoring, rural logistics coordination, and even public security surveillance, contributing to comprehensive, intelligent rural revitalization. Alongside scenario-based applications, this paper also summarizes the current structure of the agricultural low-altitude economy industry chain, which is preliminarily composed of four key components: manufacturing, flight operations, supporting services, and integrated service platforms. In the manufacturing segment, the development and production of specialized agricultural UAVs, multi-rotor drones, and fixed-wing VTOL aircraft are advancing rapidly, enabling adaptation to various terrains and crop types. The flight operations segment is expanding with increasing market participation, offering aerial spraying, broadcasting, surveying, and inspection services with improved timeliness and precision. Supporting services include airspace coordination, meteorological forecasting, equipment maintenance, and safety supervision. Additionally, comprehensive service systems have emerged, integrating digital platform management, third-party evaluation, data analysis, and agricultural technical advisory services. Talent development and training form a crucial pillar of agricultural low-altitude economy's development. A growing number of vocational institutions and drone enterprises are providing structured training programs for agricultural drone pilots, technicians, and system operators. These programs aim to enhance operational capabilities, ensure safety compliance, and foster human capital that can support the sustainable growth of agricultural low-altitude economy applications across rural regions. [Conclusions and Prospects] Agricultural low-altitude economy is transitioning from isolated, technology-specific applications to an integrated, platform-oriented ecosystem that blends equipment, data, services, and governance. To fully realize its potential, several strategic directions should be pursued. First, there is a pressing need to refine regulatory frameworks to accommodate the unique operational characteristics of rural low-altitude airspace, including dynamic zoning policies, UAV registration standards, and risk management protocols. Second, sustained investment in core technologies, such as autonomous flight control, multi-modal sensing, and AI-based analytics, are essential for overcoming current technical limitations and unlocking new capabilities. Third, agricultural low-altitude economy's infrastructure backbone should be reinforced through the deployment of distributed drone ports, mobile ground control stations, and secure data networks, particularly in underserved regions. Additionally, establishing a national-scale training and certification framework is critical for ensuring an adequate supply of skilled professionals, while innovative funding models such as public-private partnerships and scenario-based insurance can enhance accessibility and scalability. Pilot projects and demonstration zones should also be expanded to validate and refine agricultural low-altitude economy systems under diverse agroecological and socio-economic conditions. In summary, agricultural low-altitude economy offers strategic leverage for advancing China's goals in smart agriculture, green development, and rural revitalization. With cross-sector collaboration and proactive policy support, agricultural low-altitude economy can evolve into a resilient and inclusive ecosystem that fosters agricultural transformation, boosts productivity, and enhances environmental sustainability.

Low-Altitude Technology Empowering Smart Agriculture: Technical System, Application Scenarios, and Challenge Recommendations |

LAN Yubin, WANG Chaofeng, SUN Heguang, CHEN Shengde, WANG Guobin, DENG Xiaoling, WANG Yuanjie

2025, 7(6):  18-34.  doi:10.12133/j.smartag.SA202506025

Asbtract ( 138 )   HTML ( 33)   PDF (2746KB) ( 37 )  

Figures and Tables | References | Related Articles | Metrics

[Significance] Agricultural low altitude agricultural technology, with unmanned aerial vehicles (UAVs) as its primary platform, integrates 5G communication, artificial intelligence, and the Internet of Things to support data acquisition, analysis, and decision making throughout agricultural production. These advances are driving a transition from traditional experience based management toward a model in which data serve as the primary basis for decisions. As low altitude technology continues to advance in communication capacity, payload performance, and onboard processing, agricultural operations are undergoing profound changes. UAVs once used mainly for crop protection spraying have gradually evolved into multifunctional platforms capable of data collection, crop growth monitoring, image interpretation, precise input application, and operational assistance. Supported by a three dimensional integrated framework, which includes vertical integration, horizontal expansion, and spatio-temporal coordination, low altitude systems are reshaping field management structures, operational modes, and decision making processes. This transformation is accelerating the digitalization, networking, and intelligent upgrading of agriculture. The aim is to provide theoretical guidance and technical pathways for the broader application of low altitude technology in agriculture and to support the exploration of sustainable development models and industrial layouts for the low altitude agricultural economy. [Progress] The core contribution of low altitude technology to smart agriculture lies in establishing a complete sensing, decision, execution, and feedback cycle and implementing a four level structure comprising infrastructure, core technologies, application support, and scenario deployment. The infrastructure layer relies on 4G/5G networks, RTK high precision positioning, and ground based sensor systems. Multi source data are acquired through UAV mounted multi-spectral and thermal sensors working in coordination with ground monitoring devices to capture information on crop conditions and field environments. The core technology layer utilizes edge computing, cloud platforms, and analytical models to support growth assessment, pest and disease warnings, and other forms of analysis. At the application layer, UAVs operate in collaboration with ground equipment to implement precise crop protection, seeding, and irrigation, while also extending to field monitoring and agricultural logistics. This paper focuses on agricultural low altitude agricultural technology, summarizes its mechanisms and systematically reviews the associated technical system from the perspectives of operational equipment, low altitude remote sensing and recognition, data processing and analysis, and precision operation and supervision. It further examines key functions enabling agricultural intelligence. Drawing on recent research and representative cases, the paper discusses practical applications in depth. In smart orchards, for example, the South China Agricultural University Smart Patrol system combined with the Lichi Jun model can deliver early pest and disease warnings two to three weeks before outbreak and support yield estimation. In ecological unmanned farms, integrated sky, air, and ground monitoring enables autonomous operation across plowing, planting, management, and harvesting. In production operations, agricultural UAVs have accumulated over 7.5 billion mu (500 million hectare) times of service area globally, covering nearly one third of China's cultivated land area, saving approximately 210 million tons of water, and reducing carbon emissions by approximately 25.72 million tons. In logistics scenarios, transport assisted by UAVs in mountainous orchards improves efficiency more than tenfold while keeping damage rates below three percent. [Conclusions and Prospects] Sensors remain fundamental tools for capturing agricultural information and reflecting crop growth conditions. Developing highly generalizable technical modules helps lower application barriers and improve operational efficiency, while fusing multi scale data partially compensates for the limitations of single source information. Despite rapid progress, the low altitude agricultural economy still faces challenges including technological maturity, application cost, standardization, industrial integration, and workforce development. Based on an analysis of these challenges, this paper proposes building a three dimensional integrated technology framework featuring vertical integration, horizontal expansion, and spatio-temporal coordination; promoting the improvement and unification of technical standards; constructing an integrated industry ecosystem spanning research, manufacturing, application, and service; and strengthening policy support, industry norms, and talent training systems. These measures are expected to accelerate the emergence of new drivers of growth in the low altitude agricultural economy.

Current Status and Development Trend of Low-Altitude Economy Industry in Orchards |

WANG Xuechang, XU Wenbo, ZHENG Yongjun, YANG Shenghui, LIU Xingxing, SU Daobilige, WANG Zimeng

2025, 7(6):  35-57.  doi:10.12133/j.smartag.SA202506008

Asbtract ( 401 )   HTML ( 22)   PDF (2540KB) ( 30 )  

Figures and Tables | References | Related Articles | Metrics

[Significance] The low-altitude economy in orchards represents a key emerging direction in the integrated development of new-quality productive forces in agriculture. As a burgeoning industry driving the high-quality development of the fruit sector, it relies on the integration of advanced equipment manufacturing, the application of smart agriculture technologies and the expansion of consumer-centric ecosystems. These elements contribute to building a full-cycle industrial chain encompassing orchard production, management and services. This fosters the coordinated development of the entire low-altitude value chain and supports the formation of a closed-loop industrial ecosystem. This paper systematically reviews the key technological pathways and development trends of the orchard low-altitude economy across three dimensions: upstream equipment manufacturing, midstream operational processes and downstream service systems. The aim is to provide strategic reference for technological innovation and industrial planning in related fields. [Progress] In the upstream segment, research and industrial development are increasingly focused on lightweight and multifunctional aerial platforms tailored to the complex terrain of mountainous orchards. By utilizing carbon fiber composites, high energy-density batteries and hybrid power systems, these platforms achieve significant reductions in weight and improvements in flight endurance. The integration of artificial intelligence (AI) computing chips, light detection and ranging (LiDAR) and multispectral sensors equips drones with advanced capabilities for precise fruit tree recognition, obstacle avoidance in complex landscapes and multimodal environmental perception. With centimeter-level real-time kinematic (RTK) positioning and multi-sensor fusion flight control algorithms, operational safety and autonomy have been greatly enhanced. Furthermore, low-altitude infrastructure, such as distributed takeoff and landing points and mobile battery-swapping stations, based on integrated 5G-Advanced and BeiDou navigation communication systems, is being systematically deployed. This provides strong support for continuous unmanned operations in hilly and mountainous orchards. The midstream segment, encompassing the pre-production, in-production and post-production stages, serves as the core scenario for value realization in the low-altitude economy. In the pre-production stage, high-resolution remote sensing imagery, combined with machine learning models such as extreme gradient boosting (XGBoost) and convolutional neural networks, enables detailed diagnostics of soil nutrients, micro-topography and vegetation cover. These insights support the precise planning of digital orchards. During the in-production stage, monitoring models based on indices such as normalized difference vegetation index (NDVI) and leaf area index (LAI) facilitate real-time assessment of tree vigor and early detection of pests and diseases, enhancing the management of plant health and growth conditions. Intelligent systems that integrate target recognition, path optimization and electric atomizing nozzles allow for precise, demand-driven application of pesticides and fertilizers, thereby improving resource efficiency and reducing environmental impact. Additionally, collaborative multi-UAV (unmanned aerial vehicle) operations and ground-aerial collaboration, optimized through genetic algorithms and digital twin models, further enhance task scheduling, flight path planning and energy utilization. In the post-production stage, drones equipped with robotic arms or vacuum suction grippers, coupled with thermal imaging, are increasingly effective in fruit identification and targeted harvesting, achieving higher levels of automation and reliability. At the same time, low-altitude logistics networks, supported by autonomous navigation and multi-sensor obstacle avoidance technologies, are addressing the last-mile challenges in cold-chain transportation. This significantly shortens the time window from field to sorting center, improving overall supply chain efficiency. At the downstream service level, the orchard low-altitude economy has evolved beyond single-equipment sales into a diversified service ecosystem. This emerging model centers on pilot training, drone insurance, equipment leasing and the integration of orchard tourism, forming a new type of business landscape. On one hand, standardized pilot training programs and operational quality evaluation systems have enhanced both talent development and safety assurance. On the other hand, risk control models developed by insurers based on operational data, along with "rent-to-own" financing schemes, have effectively lowered entry barriers for farmers. Moreover, the rise of integrated low-altitude agri-tourism models is steadily boosting the brand value of fruit products and generating new income streams through cultural and tourism-related activities. [Conclusions and Prospects] As a vital carrier of new-quality productive forces in agriculture, the orchard low-altitude economy has established a comprehensive industrial chain encompassing equipment manufacturing, operational systems and service platforms. This integrated structure is driving the transformation of orchard management toward greater intelligence, precision and sustainability. Despite current challenges such as limited equipment endurance and underdeveloped service systems, the sector is expected to achieve continuous breakthroughs through the development of high-payload aerial platforms, the integration of data-driven operational systems, the construction of diversified service ecosystems, and the refinement of relevant policies and standards. With the gradual opening of low-altitude airspace and the rapid iteration of core technologies, the orchard low-altitude economy is poised to become a key driver of agricultural modernization and rural revitalization.

Remote Sensing Approaches for Cropland Abandonment Perception in Southern Hilly and Mountainous Areas of China: A Review |

LONG Yuqiao, SUN Jing, WEN Yanru, WANG Chuya, DONG Xiuchun, HUANG Ping, WU Wenbin, CHEN Jin, DING Mingzhong

2025, 7(6):  58-74.  doi:10.12133/j.smartag.SA202505022

Asbtract ( 67 )   HTML ( 0)   PDF (4023KB) ( 5 )  

Figures and Tables | References | Related Articles | Metrics

[Significance] Cropland abandonment in hilly and mountainous regions is a pronounced manifestation of land‐use marginalization, with profound implications for both regional food security and ecosystem service provision. In southern China, this issue is particularly acute due to a confluence of factors including early and rapid urbanization, rugged topographic relief, complex multi‐cropping systems, and substantial rural‐to‐urban labor migration, which have driven widespread abandonment of steep, fragmented terraces. This trend presents a profound dual dilemma: On one hand, the cessation of cultivation diminishes local grain production capacity, amplifies pressure on existing cropland, and threatens national food supplies; On the other hand, the secondary succession of spontaneous vegetation on these deserted parcels offers significant carbon sequestration potential and contributes to biodiversity recovery. Yet, accurately mapping these spatio-temporal patterns is severely hampered by persistent cloud cover and the landscape's complexity. This leaves decision-makers without the timely, high‐resolution maps needed to track abandonment dynamics, uncover their socioeconomic and environmental drivers, and craft land-use policies that holistically balance agricultural output, carbon storage, and landscape resilience. [Progress] Drawing from literature published since 2014, this paper systematically reviews remote sensing‐based methods for cropland abandonment, revealing a clear developmental trajectory. Methodologically, the approaches have evolved along two parallel paths. First, the monitoring paradigm has shifted from early "state comparison" methods, such as post-classification comparison of discrete multi-temporal images, to modern "process tracking" approaches. These leverage dense time series, utilizing phenology‐aware algorithms such as LandTrendr and BFAST to identify abrupt or gradual breaks in the vegetation trajectory, thus capturing the dynamics of abandonment and distinguishing it from short-term fallows. Second, the identification algorithm has progressed from traditional machine learning classifiers and object-based image analysis (OBIA), which depend on hand‐crafted features, towards sophisticated deep learning frameworks capable of automatically learning complex spatio-temporal signatures. Concurrently, data pre-processing techniques have advanced significantly, with harmonic analysis, Savitzky-Golay filtering, and the integration of Synthetic Aperture Radar (SAR) data now routinely applied to reconstruct continuous, high-quality time series. Furthermore, this review provides a critical synthesis of common methodological issues, focusing on the spatio-temporal representativeness bias in ground validation samples and the multiple sources of uncertainty stemming from cloud cover, mixed pixels, and phenological variability. [Conclusions and Prospects] Despite considerable advances, persistent challenges continue to limit operational monitoring. Looking forward, the field must evolve from descriptive mapping toward a truly predictive and decision‐ready framework. This transformation hinges on five interlinked frontiers. First, it requires forging the seamless integration of diverse data streams: Fusing optical imagery, radar backscatter, and terrain models within cloud computing environments to yield uninterrupted, high‐resolution time series that capture both abrupt and gradual land‐use changes. Second, it necessitates the establishment of an extensive, stratified ground‐truth network; by systematically sampling high-risk, transitional, and reference plots and collecting synchronized measurements, researchers can iteratively recalibrate classification models and improve their resilience across the region's landscape mosaic. Third, on the algorithmic frontier, hybrid approaches that embed expert‐defined phenological rules within deep learning architectures offer a promising path to robustly disentangle permanent abandonment from temporary fallows and to quantify a continuous "abandonment intensity". Fourth, the deployment of fully automated and reproducible processing pipelines on cloud platforms like Google Earth Engine will democratize access to near-real‐time monitoring and enhance reproducibility through open-source workflows. Finally, anchoring detection within dynamic simulation frameworks (e.g., agent‐based models) driven by historical trajectories and key drivers will allow for the projection of future abandonment of "hotspots". Layering these projections with multi‐criteria risk assessments will yield spatially explicit risk maps to guide precision interventions—such as targeted recultivation subsidies or ecological restoration efforts—enabling sustainable land stewardship that simultaneously safeguards food security and enhances ecosystem resilience.

Progress in Soil Moisture Retrieval under Crop Canopy Cover Based on Multi-polarization SAR Data |

SUN Rong, GAO Han, JIANG Yujie, LI Qiaochu, WU Haoyu, WU Shangrong, YU Shan, XU Lei, YU Liangliang, ZHANG Jie, BAO Yuhai

2025, 7(6):  75-95.  doi:10.12133/j.smartag.SA202509009

Asbtract ( 43 )   HTML ( 2)   PDF (2235KB) ( 4 )  

Figures and Tables | References | Related Articles | Metrics

[Significance] Soil moisture is a critical parameter in surface water cycling and agricultural productivity, playing an essential role in crop growth monitoring, yield estimation, and field management. Synthetic aperture radar (SAR), with its all-weather capabilities and multi-polarization advantages, is highly sensitive to the structural, orientational, and moisture characteristics of crops and soil, making it a key remote sensing tool for soil moisture monitoring. However, under crop cover, surface scattering signals are confounded by vegetation scattering, and the spatial heterogeneity of crop and soil properties further complicates the scattering process. These factors make it challenging to directly apply traditional methods for agricultural soil moisture retrieval. The separation of scattering contributions from the crop canopy and underlying soil remains a significant research challenge. To address this, the present paper systematically reviews the state-of-the-art advancements in soil moisture retrieval under crop cover across three dimensions: data resources, scattering theory, and retrieval applications. [Progress] This review offers a comprehensive assessment of multi - polarization SAR - based agricultural soil moisture retrieval technology from the viewpoints of data, theory, and application, emphasizing future optimization. In terms of data resources, the paper presents a comprehensive summary of spaceborne multi - polarization SAR data. It compares key imaging parameters (e.g., frequency band, polarization mode, spatial resolution, and incidence angle) and analyzes their impacts on agricultural soil moisture retrieval. Research shows that, under single - source data conditions, long - wavelength bands, small incidence angles, and co - polarization modes are less prone to canopy scattering interference. Under multi - modal data conditions, integrating multi - band, multi - angle, and multi - polarization SAR data can more effectively distinguish between vegetation and surface scattering contributions. Regarding theoretical and technical progress, the paper tracks the development of scattering models, reviews existing soil and vegetation scattering models, and contrasts the applicability of physical, empirical, and semi - empirical models. It also emphasizes the advantages of coupled modeling approaches. Moreover, the paper examines various solution methods for scattering models, focusing on local and global optimization algorithms. In the application context, this paper evaluates the performance of multi - polarization SAR in soil moisture retrieval across different crop and soil conditions, using wheat, corn, rapeseed, and soybean as typical crops. It discusses the influence of different crop types (e.g., differences in leaf and stem structure) and phenological stages on retrieval accuracy. The paper compares the applicability of soil scattering models and retrieval methods under various soil surface roughness and soil texture conditions (e.g., sandy and loamy soils) and examines their retrieval accuracy under different soil scenarios. Additionally, it reviews the improvements in retrieval performance through multi - source data fusion, including optical - SAR combinations and active - passive remote sensing fusion. It also synthesizes the main challenges and future directions for multi - source data fusion strategies, especially with regard to scale effects. [Conclusions and Prospects] Based on the reviewed advancements, the paper identifies key technical challenges, including discrepancies in monitoring range and scale among spaceborne, airborne, and ground-based data, difficulties in adapting scattering models to crop morphology, and the lack of standardized validation protocols for retrieval results. Looking ahead, the paper envisions the potential for future technological progress driven by multi-modal big data and artificial intelligence. This review highlights critical insights, addresses key bottlenecks, and drives the development of intelligent, adaptive, high-resolution, and high-precision soil moisture retrieval systems in multi-polarization SAR soil moisture retrieval.

LightTassel-YOLO: A Real-Time Detection Method for Maize Tassels Based on UAV Remote Sensing |

CAO Yuying, LIU Yinchuan, GAO Xinyue, JIA Yinjiang, DONG Shoutian

2025, 7(6):  96-110.  doi:10.12133/j.smartag.SA202505021

Asbtract ( 234 )   HTML ( 13)   PDF (4491KB) ( 11 )  

Figures and Tables | References | Related Articles | Metrics

[Objective] The accurate identification of maize tassels is critical for the production of hybrid seed. Existing object detection models in complex farmland scenarios face limitations such as restricted data diversity, insufficient feature extraction, high computational load, and low detection efficiency. To address these challenges, a real-time field maize tassel detection model, LightTassel-YOLO (You Only Look Once) based on an improved YOLOv11n is proposed. The model is designed to quickly and accurately identify maize tassels, enabling efficient operation of detasseling unmanned aerial vehicles (UAVs) and reducing the impact of manual intervention. [Methods] Data was continuously collected during the tasseling stage of maize from 2023 to 2024 using UAVs, establishing a large-scale, high-quality maize tassel dataset that covered different maize tasseling stages, multiple varieties, varying altitudes, and diverse meteorological conditions. First, EfficientViT (Efficient vision transformer) was applied as the backbone network to enhance the ability to perceive information across multi-scale features. Second, the C2PSA-CPCA (Convolutional block with parallel spatial attention with channel prior convolutional attention) module was designed to dynamically assign attention weights to the channel and spatial dimensions of feature maps, effectively enhancing the network's capability to extract target features while reducing computational complexity. Finally, the C3k2-SCConv module was constructed to facilitate representative feature learning and achieve low-cost spatial feature reconstruction, thereby improving the model's detection accuracy. [Results and Discussions] The results demonstrated that LightTassel-YOLO provided a reliable method for maize tassel detection. The final model achieved an accuracy of 92.6%, a recall of 89.1%, and an AP@0.5 of 94.7%, representing improvements of 2.5, 3.8 and 4.0 percentage points over the baseline model YOLOv11n, respectively. The model had only 3.23 M parameters and a computational cost of 6.7 GFLOPs. In addition, LightTassel-YOLO was compared with mainstream object detection algorithms such as Faster R-CNN, SSD, and multiple versions of the YOLO series. The results demonstrated that the proposed method outperformed these algorithms in overall performance and exhibits excellent adaptability in typical field scenarios. [Conclusions] The proposed method provides an effective theoretical framework for precise maize tassel monitoring and holds significant potential for advancing intelligent field management practices.

Corn Borer Pests Infestations Detection Method Using Low-Altitude Close-Range UAV Imagery |

ZHAO Jun, NIE Zhigang, LI Guang, LIU Jiayu

2025, 7(6):  111-123.  doi:10.12133/j.smartag.SA202505006

Asbtract ( 180 )   HTML ( 17)   PDF (3745KB) ( 8 )  

Figures and Tables | References | Related Articles | Metrics

[Objective] The detection of corn borer infestations is essential for improving corn yield and quality, as corn borer pests pose a significant threat to global corn production. In traditional agricultural practices, identifying corn borer infestations relies on manual field inspections or trapping tools, which are labor-intensive, time-consuming, and difficult to implement over large areas. These methods are further limited by their susceptibility to human error and inability to meet the demands of modern precision agriculture. To address these challenges, a method for detecting corn borer infestations using low-altitude, close-range imagery captured by unmanned aerial vehicles (UAVs) was investigated. By focusing on detecting boreholes rather than insect bodies, this approach overcomes the difficulties of detecting corn borers, which are nocturnal and often concealed within plant tissues, thereby enhancing the applicability of field-based detection and aligning with practical field conditions. [Methods] Based on the YOLOv11 (You Only Look Once v11) object detection algorithm, a model named YOLO-ESN was introduced, for corn borer infestation detection. The YOLO-ESN model was optimized through multiple modifications. In the Backbone, an enhanced lightweight attention (ELA) mechanism was incorporated to increase sensitivity and improve the extraction of small visual features, such as boreholes, by modeling spatial dependencies in horizontal and vertical directions using one-dimensional convolutions. In the Neck, a C3k2-Spatial and channel reconstruction convolution (C3k2-SCConv) module was introduced to reduce the number of model parameters while improving feature fusion efficiency through spatial and channel reconstruction, suppressing redundant information. In the Head, a small-object detection branch, termed the P2 detection head, was added, enabling YOLO-ESN to directly utilize shallow, high-resolution features from early network layers to enhance the detection of fine-grained targets like boreholes. Additionally, a combined loss function of normalized Wasserstein distance (NWD) and efficient intersection over union (EIoU) was employed to optimize bounding box regression accuracy, addressing gradient vanishing issues for small targets and improving target localization stability and robustness. A decision tree algorithm was applied to classify infestation severity levels based on borehole detection results, and heatmaps were generated to visualize the spatial distribution of corn borer infestations across the field. [Results and Discussions] Multiple experiments were conducted using a constructed dataset of corn borer infestation images. The results demonstrated that YOLO-ESN achieved an mAP@50 of 88.6% and an mAP@50:95 of 40.5%, representing an improvement of 7.6 and 4.9 percentage points, respectively, compared to the original YOLOv11 model. The total number of parameters in YOLO-ESN was reduced by 11.52%, contributing to a lighter model suitable for UAV deployment. Ablation studies evaluated individual contributions: incorporating the ELA mechanism alone improved mAP@50 by 0.3 percentage points, and the parameters are reduced by 10.57%; replacing the C3k2 module with C3k2-SCConv reduced parameters by 2.5% while increasing mAP@50 by 0.9 percentage points; adding the P2 detection head enhanced mAP@50 and mAP@50:95 by 4.1 and 1.2 percentage points, respectively; and introducing the NWD+EIoU loss function improved mAP@50 and mAP@50:95 by 1.9 and 1.2 percentage points, respectively. Comparative experiments demonstrate that YOLO-ESN outperforms a range of mainstream object detection models, including Faster R-CNN, SSD, YOLOv8, YOLOv11, and YOLOv12. YOLO-ESN achieves an mAP@50 and an mAP@50:95, surpassing Faster R-CNN by 14.9 and 9.7 percentage points, respectively, and SSD by 17.8 and 11.4 percentage points, respectively. With a compact parameter size of 8.37 M, YOLO-ESN delivers excellent detection accuracy and generalization, striking a strong balance between performance and efficiency. Although its inference speed (32.48 frame/s) was slightly slower than YOLOv12 (75.44 frame/s), it offered a superior trade-off between accuracy and efficiency. These results validated YOLO-ESN as a lightweight, high-performing solution for small object detection tasks, such as dense small targets in remote sensing images. The decision tree algorithm classified infestation severity with high accuracy, achieving F₁-Scores of 0.906, 0.803, and 0.842 for mild, moderate, and severe infestations, respectively. Heatmaps generated from borehole detection results enabled spatial visualization of infestation severity, providing a scientific basis for quantitative monitoring and targeted pesticide application in field infestations. [Conclusions] The proposed YOLO-ESN model has more advantages in overall detection accuracy and running speed. While improving the lightweight degree and deployment efficiency of the model, it also shows better recognition ability in small target detection, and can accurately locate the wormhole area on the corn leaf, effectively improving the bounding box regression accuracy and feature extraction efficiency. Compared with the traditional insect recognition method, the use of wormholes as detection objects is more in line with the actual field situation, effectively avoiding the problems of insect occlusion and strong concealment, and improving the availability of field image data and algorithm robustness. The heat map generated by the model detection results can also effectively display the distribution changes of insect pests in farmland, providing a scientific basis for precision pesticide spraying and farmland management. Overall, this study provides an effective solution for the intelligent detection of corn borer pests, has strong versatility and promotion prospects, and can provide strong technical support for precision agriculture and smart farmland management.

Robust UAV-Based Method for Peanut Plant Height Estimation Using Bare-Soil Invariant Constraints |

SONG Mingxuan, BAI Bo, YANG Juntao, ZHANG Yutao, LI Sa, LI Zhenhai, WAN Shubo, LI Guowei

2025, 7(6):  124-135.  doi:10.12133/j.smartag.SA202509029

Asbtract ( 33 )   HTML ( 1)   PDF (3004KB) ( 1 )  

Figures and Tables | References | Related Articles | Metrics

[Objective] Peanuts plant height is a key structural trait for assessing crop growth and nitrogen response. Accurate and efficient height acquisition is essential for monitoring canopy vigor, supporting genotype selection, and enabling precision management. However, conventional ground control point (GCP)-based methods require substantial field deployment and are highly sensitive to local misregistration between multi-temporal digital surface models (DSMs) and digital elevation models (DEMs). In low-stature, prostrate peanut canopies on uneven terrain, such residual elevation errors propagate directly into the canopy height model, severely reducing estimation accuracy. To overcome these limitations, a robust unmanned aerial vehicle (UAV)-based method is developed for peanut plant height estimation using bare-soil invariant constraints. The workflow incorporates crop-mask-assisted fine registration to optimize DSM-DEM alignment and eliminates the need for dense GCP distribution. [Methods] Field experiments were conducted at a peanut experimental station in Wangbian community, Ningyang county, Tai'an city, Shandong Province, China, using multiple UAV platforms (DJI Mavic 3 Multispectral and DJI MATRICE 350 RTK equipped with a DJI Zenmuse P1 camera), two growth stages (42 d and 49 d after sowing, DAS), and two nitrogen fertilization levels (high nitrogen and low nitrogen). To validate the peanut plant height estimates, representative plants in each plot were selected before and after each UAV image acquisition, and manual measurements from the ground surface to the canopy apex were recorded as the reference plant height. High-resolution digital orthomosaic (DOM) images were then generated from the UAV data, and peanut canopy regions were extracted using the excess green (ExG) index. To mitigate threshold instability caused by variable illumination and soil background conditions, a fixed empirical threshold was combined with an adaptive strategy that integrated Otsu's between-class variance method and the median absolute deviation (MAD), thereby ensuring robust canopy segmentation across growth stages and nitrogen treatments. After canopy extraction, the peanut canopy mask derived from the DOM was used to remove corresponding pixels from the DSM on a per-pixel basis. The DSMs with and without canopy points were then separately used for 3D reconstruction, yielding a canopy point cloud and a bare-soil point cloud. This bare-soil point cloud and a bare-soil DSM acquired before crop emergence (used as the DEM reference) were jointly input into the iterative closest point (ICP) algorithm to solve for a three-dimensional rigid transformation matrix. The resulting matrix was used to jointly optimize translations and rotations along the X, Y, and Z directions. It was applied uniformly to the DSM containing peanut canopy points, thereby achieving fine-scale alignment between the DSM and DEM at the block level. Following registration, the DEM was used as the ground reference, and the canopy height model was constructed by differencing the DSM and DEM pixel by pixel. The 95th percentile (P95) of canopy height within each plot, derived from the canopy height histogram, was used as the representative plant height to reduce the influence of local noise on the statistics. [Results and Discussions] The results showed that varying the ExG threshold among 0.05, 0.10 and 0.15 had only a limited effect on overall plant height estimation accuracy, with the best performance observed at 0.10. At this threshold, the Mavic 3 platform achieved an R² of 0.864 7 and a root-mean-square error (RMSE) of 2.57 cm. In contrast, the P1 platform achieved an R² of 0.918 6 and an RMSE of 2.05 cm, indicating that the proposed threshold selection strategy provided a good balance between accuracy and robustness. Error analysis across different canopy-height percentiles showed that, as the percentile increased from P90 to P99, R² and RMSE exhibited a typical concave pattern, first improving and then degrading. Among these percentiles, P95 yielded the highest R² and the lowest RMSE, representing the best trade-off between noise suppression and canopy-top information retention; therefore, P95 was adopted as the representative plant height for this method. Under the P95-based definition of plant height, the traditional GCP method produced R² values of only 0.592 3－0.669 9 and RMSE values of 4.60~4.94 cm, and the "GCP+ICP" workflow, in which canopy points were not removed prior to ICP registration, was most strongly affected by noise in the point clouds, with R² dropping below 0.3 in some cases. In contrast, the proposed method maintained R² values of 0.864 7~0.918 6 and RMSE values of 2.05~2.57 cm across both platforms, markedly improving the agreement between estimated and measured plant height relative to the traditional GCP-based approach. Further platform-specific analysis showed that, owing to its higher spatial resolution, the P1 platform reconstructed a more complete canopy-top structure and yielded better plant height estimates than the Mavic 3 platform at each growth stage. Nevertheless, when combined with the proposed plant height extraction workflow, the Mavic 3 platform still achieved reliable performance (R² > 0.817 7) in regions with different nitrogen contents, confirming the method's multi-platform applicability. From the perspective of canopy cover and nitrogen level, as the crop progressed from 42 to 49 DAS, the peanut canopy gradually approached full closure, the proportion of high-value pixels in the canopy height model increased, the canopy-top point cloud in the DSM became more continuous, and plant height estimation accuracy improved accordingly. Under high nitrogen treatment, the canopy was denser and structurally more complete than under low nitrogen treatment, resulting in slightly higher R² and slightly lower RMSE on both platforms; however, these differences remained within a controllable range, demonstrating that the bare-soil-based registration workflow was robust to fertility differences and that the proposed method was stable and transferable across growth stages and fertility conditions. [Conclusions] Overall, the proposed method for estimating peanut plant height substantially alleviates the constraints imposed by the misregistration of residual DSM and DEM on plant height inversion for low-stature, prostrate crops. It achieves centimetre-level accuracy for plant height retrieval across platforms and nitrogen treatments. By significantly reducing the dependence on densely distributed GCPs and offering a simple, reproducible, and low-cost processing pipeline, the method provides a scalable technical route for monitoring peanut nitrogen responses, deriving high-throughput agronomic structural traits, and measuring plant height in other low-stature, prostrate crops.

Construction and Evaluation of Lightweight and Interpretable Soybean Remote Sensing Identification Model |

WANG Yinhui, ZHAO Anzhou, LI Dan, ZHU Xiufang, ZHAO Jun, WANG Ziqing

2025, 7(6):  136-148.  doi:10.12133/j.smartag.SA202508025

Asbtract ( 42 )   HTML ( 4)   PDF (7016KB) ( 1 )  

Figures and Tables | References | Related Articles | Metrics

[Objective] Soybean stands as one of the most crucial global crops, serving as a vital source of plant-based protein and vegetable oil while playing an indispensable role in sustainable agricultural systems and global food security. Accurate and timely mapping of soybean cultivation areas is essential for agricultural monitoring, policy-making, and precision farming. However, existing remote sensing methods for soybean identification, such as threshold-based approaches, traditional machine learning, and deep learning, often face challenges related to model complexity, computational efficiency, and interpretability. These limitations collectively highlight the pressing need for a methodological solution that maintains classification accuracy while simultaneously offering computational efficiency, operational simplicity, and interpretable results, a balance crucial for effective agricultural monitoring and policy-making. To address these limitations, a lightweight and interpretable soybean mapping framework was proposed based on Sentinel-2 imagery and a binary logistic regression model in this method. [Methods] Six representative agricultural regions within the primary U.S. soybean production belt were selected to capture the diversity of cultivation practices and environmental conditions across this major production area. The analysis utilized the complete growing season (April－October) Sentinel-2 satellite imagery from 2021 to 2023. The USDA's cropland data layer served as reference data for model training and validation, benefiting from its extensive ground verification and statistical rigor. All Sentinel-2 images undergo rigorous preprocessing, including atmospheric correction, cloud and shadow masking with the scene classification layer, and spatial subsetting to the regions of interest. The Jeffries-Matusita distance was employed as a quantitative metric to objectively identify the optimal temporal window for soybean discrimination. This statistical measure evaluated the separability between soybean and other major crops across the growing season, with calculations performed on 10 d composite periods to ensure data quality and temporal consistency. The analysis revealed that late July to mid-September (Day of Year 210－260) provided maximum spectral separability, corresponding to the soybean's critical reproductive stages (pod setting and filling) when its spectral signature becomes most distinct from other crops, particularly in short-wave infrared regions sensitive to canopy structure and water content. Within this optimally identified window, a binary logistic regression model was implemented that treated soybean identification as a probabilistic classification problem. The model was trained using spectral features from the optimal period through maximum likelihood estimation, creating a computationally efficient framework that required optimization of only a limited number of parameters while maintaining physical interpretability through explicit feature coefficients. [Results and Discussions] The comprehensive evaluation showed that the integrated approach balanced classification performance and operational practicality optimally. The temporal optimization identified late July to mid-September as the peak discriminative period, which matches soybean's reproductive phenological stages (when its canopy spectral characteristics differ most from other crops). This finding was consistent across three study years and multiple regions, verifying the robustness of the data-driven window selection. The binary logistic regression model, trained on features from this optimal period, performed excellently: In the 2022 model construction region, it achieved 0.90 overall accuracy and 0.79 Kappa coefficient. When applied to independent validation regions in the same year, it maintained strong performance (0.88 overall accuracy, 0.76 Kappa) without region-specific parameter adjustments, demonstrating outstanding spatial transferability. Temporal validation further confirmed the model's robustness: Across the 2021 to 2023 study period, it maintained consistent performance across all regions, with an average accuracy of 0.87 and Kappa of 0.76. This inter-annual stability is notable, despite potential variations in annual weather, management practices, and planting schedules, and highlights the advantage of basing the model on a stable phenological period rather than fixed calendar dates. The model's lightweight architecture offered practical benefits: Compared with complex ensemble or deep learning methods, it only requires optimizing a limited number of parameters. This parsimonious structure enhances computational efficiency, enabling rapid training and deployment over large areas while reducing reliance on extensive labeled datasets—a key advantage in regions lacking sufficient ground truth data. Beyond accuracy and efficiency, the model exhibited exceptional interpretability via its probabilistic framework and transparent feature weighting. Coefficient analysis provided quantifiable insights into feature contributions, revealing that short-wave infrared bands and specific vegetation indices had the highest discriminative power during the optimal temporal window. [Conclusions] An effective soybean mapping approach that balances accuracy with operational practicality through the strategic combination of temporal optimization and binary logistic regression was proposed. The method offers a viable solution for operational agricultural monitoring, especially in resource-constrained environments. Future work can enhance the robustness of the model across multiple regional conditions through cross-regional validation in different climate zones and cropping systems, or by integrating transfer learning with domain adaptation methods. This will improve its potential for global-scale application. Concurrently, integrating additional data, methodologies, and models to achieve end-to-end feature learning should be considered.

Physics-Constrained PROSAIL-cGAN Approach for Spectral Sample Augmentation and LAI Inversion of Winter Wheat |

LU Yihang, DONG Wen, ZHANG Xin, YAN Ruoyi, ZHANG Yujia, TANG Tao

2025, 7(6):  149-160.  doi:10.12133/j.smartag.SA202508026

Asbtract ( 48 )   HTML ( 3)   PDF (1771KB) ( 2 )  

Figures and Tables | References | Related Articles | Metrics

[Objective] The leaf area index (LAI) is a key biophysical parameter that reflects the canopy structure and photosynthetic capacity of crops. However, the inversion of winter wheat LAI from remote sensing data is often constrained by the limited availability of field measurements, leading to insufficient model generalization. Although radiative transfer model (RTM)-based simulations can expand the sample size, discrepancies between simulated and observed spectra persist due to simplified canopy and soil parameterizations. Conversely, purely data-driven generative models such as generative adversarial networks (GANs) can enhance sample diversity but often produce physically inconsistent samples in the absence of biophysical constraints. To address these issues, a physics-constrained PROSAIL-cGAN (conditional generative adversarial network) spectral sample augmentation method was proposed that integrated the PROSAIL model with cGAN to improve the accuracy and robustness of LAI inversion under small-sample conditions, generate physically realistic spectral-parameter pairs and provide reliable data support for remote sensing-based monitoring of winter wheat growth. [Methods] The study area was located in Zouping city, Shandong Province, a major winter wheat production region within the Huang-Huai-Hai Plain. A total of 133 field samples were collected during the jointing stage in April 2025 using an LAI-2200C canopy analyzer, with synchronous canopy spectra acquired. A Sentinel-2A Level-2A image from April 15, 2025, served as the remote sensing source, comprising 13 bands resampled to a spatial resolution of 10 m. The dataset was divided into training (70%) and validation (30%) subsets, with LAI values ranging from 1.646 to 7.505. The proposed method combined the PROSAIL radiative transfer model with a conditional GAN framework. First, PROSAIL was employed to simulate canopy reflectance and corresponding biophysical parameters, including chlorophyll content (C_ab), carotenoid content (C_ar), brown pigment content (C_brown), equivalent water thickness (C_w), dry matter content (C_m), LAI, and leaf inclination distribution (LIDFa). A multi-layer perceptron (MLP) surrogate model was then trained to approximate the forward mapping of PROSAIL, enabling differentiability for integration with deep learning architectures. The cGAN generator received random noise and physical parameters as conditional inputs to produce corresponding canopy reflectance, while the discriminator jointly evaluated authenticity and physical consistency. During adversarial training, physical constraints were incorporated into the generator's loss function to ensure biophysical realism. The generated samples were subsequently filtered based on parameter ranges and discriminator confidence scores. Kernel density overlap between real and generated LAI distributions was used to quantify their statistical consistency. Finally, the enhanced dataset was used to train random forest (RF) and extreme gradient boosting (XGBoost) regression models for the LAI inversion. Model performance was assessed using the coefficient of determination (R²), root mean square error (RMSE), and mean absolute error (MAE), and compared with three baselines: 1) field-measured modeling, 2) the PROSAIL lookup table (LUT) method, and 3) cGAN-only augmentation. [Results and Discussions] The surrogate MLP model accurately reproduced PROSAIL-simulated spectra, achieving R² 0.817, RMSE 0.008 5, and MAE 0.005 5, confirming its feasibility as a differentiable physical proxy. The cGAN-based augmentation achieved a LAI distribution overlap of 0.806 with the measured samples, whereas the PROSAIL-cGAN improved the overlap to 0.827, demonstrating enhanced physical realism and sample diversity. Model comparisons revealed substantial differences in performance. The LUT-based inversion yielded only R² 0.353 0 and RMSE 1.284 0, reflecting its limited adaptability to spectral heterogeneity. Direct regression using field data improved accuracy (R²=0.680 1 for XGBoost and 0.648 8 for RF). Incorporating cGAN-generated samples further enhanced model accuracy (R² 0.745 0 for RF and 0.739 0 for XGBoost). The PROSAIL-cGAN-enhanced RF model achieved the best overall performance, with R² 0.848 8, RMSE 0.540 9, and MAE 0.293 7. The sample-size sensitivity analysis demonstrated that as the number of field samples increased from 27 to 106, R² improved from 0.546 2 to 0.848 8 and RMSE decreased from 1.024 3 to 0.540 9. When the sample size exceeded 79, model performance stabilized, indicating strong robustness. Spatial mapping results showed that LAI values were higher in the central and northern regions (4~7) and lower in the southern mountainous areas (1.5~4), consistent with variations in soil fertility and field management practices. These findings validate the model's applicability for regional-scale monitoring of crop growth. [Conclusions] This study developed a physics-constrained PROSAIL-cGAN spectral sample augmentation method for winter wheat LAI inversion. By integrating a radiative transfer model, a conditional generative network, and a differentiable surrogate, the method effectively generated physically consistent and diverse spectral-parameter samples under small-sample conditions. The PROSAIL-cGAN-based RF model achieved a relatively high inversion accuracy, outperforming traditional LUT and field-only approaches. The proposed method successfully mitigated small-sample limitations, ensured physical interpretability, and improved model generalization. It provides a robust framework for the remote sensing inversion of crop canopy parameters, supporting precision agriculture and dynamic monitoring of crop growth. Future work will focus on optimizing sample generation strategies, integrating multi-temporal satellite data and additional physiological parameters, and coupling with deep or semi-supervised learning techniques to further enhance scalability and applicability across crops and regions.

Method for Estimating Leaf Area Index of Winter Rapeseed Based on Fusion of Vegetation Indices and Texture Features |

LIU Jie, GUO Jiaxin, ZHANG Jiahao, ZHANG Bingchao, XIONG Jie, CAO Jianpeng, WU Shangrong, DENG Yingbin, CHEN Guipeng

2025, 7(6):  161-173.  doi:10.12133/j.smartag.SA202507018

Asbtract ( 32 )   HTML ( 0)   PDF (1434KB) ( 1 )  

Figures and Tables | References | Related Articles | Metrics

[Objective] Leaf area index (LAI) is a vital agronomic parameter that reflects the structure of crop canopies, photosynthetic capacity, and population growth status. It holds significant importance for the precision cultivation and management of winter oilseed rape. Traditional methods for measuring LAI, such as destructive sampling or the use of costly instruments, are often constrained by low efficiency, high costs, and limited adaptability. In contrast, unmanned aerial vehicle (UAV) remote sensing technology offers a novel approach to rapid and non-destructive LAI monitoring due to its advantages in high resolution and flexibility. However, reliance solely on spectral vegetation indices (VIs) frequently results in saturation phenomena at elevated LAI levels, complicating accurate representation of complex canopy structures. Consequently, this study aims to investigate the integration of vegetation indices with texture features (TFs) using machine learning techniques to enhance the estimation accuracy of LAI throughout the entire growth cycle of winter oilseed rape. [Methods] The research was conducted at an experimental site in Gao'an city, Jiangxi province, where 81 plots were established with varying sowing dates, densities, and fertilization treatments. These plots encompassed four critical growth stages: seedling, bud elongation, flowering, and pod development. A DJI Phantom 4 RTK multispectral UAV was employed to acquire image data, while ground-truth LAI data were concurrently collected using an LAI-2200C plant canopy analyzer, resulting in a total of 324 valid samples. From the multispectral imagery obtained, eight vegetation indices, namely normalized difference vegetation index (NDVI), optimized soil adjusted vegetation index (OSAVI), normalized difference red edge (NDRE), enhanced vegetation index (EVI), green normalized difference vegetation index (GNDVI), simple ratio (SR), difference vegetation index (DVI), and canopy infrared reflectance estimate (CIRE), were derived alongside reflectance values from five spectral bands: blue, green, red, red-edge, and near-infrared. Additionally, 40 texture features were extracted based on the Gray-Level Co-occurrence Matrix. To select the ten most representative features with minimal redundancy among these variables, the minimum redundancy maximum relevance (mRMR) algorithm was utilized. Subsequently, three machine learning algorithms, multiple linear regression (MLR), extreme gradient boosting (XGBoost), and support vector machine regression (SVR), were applied to develop models for estimating LAI. To assess model generalizability and mitigate overfitting risks during evaluation processes, a 3-fold GroupKFold cross-validation approach was implemented to ensure that samples originating from the same plot remained intact between training and testing sets. The performance of each model was rigorously evaluated using several metrics, including the coefficient of determination (R²), root mean square error (RMSE), and mean absolute error (MAE). [Results and Discussions] The results indicated that the LAI of winter oilseed rape exhibited a dynamic trend throughout its growth cycle, characterized by being "low at the seedling stage, high at the bud elongation stage, decreasing at the flowering stage, and dropping again at the pod stage". The LAI values ranged from 0.90 to 6.39, with a uniform sample distribution observed within each developmental phase. Most vegetation indices and texture features demonstrated a highly significant correlation with LAI (P < 0.001), with the SR and near-infrared entropy (NIR-entropy) exhibiting the strongest correlation (r = 0.81). In terms of feature selection, vegetation indices such as SR, CIRE, and NDRE along with texture features like NIR-entropy and G-variance maintained high selection frequencies among the top ten mRMR-selected features. This indicated their stable contribution to model construction. Regarding model performance, the fused vegetation and texture features (VTFs) model outperformed all other models evaluated; specifically, the VTFs-SVR model achieved superior estimation accuracy across the entire growth cycle (R²=0.90, RMSE=0.38, MAE=0.27). When compared to models utilizing only vegetation indices or solely texture features, the fused model demonstrated particularly enhanced performance during high-coverage stages such as bud elongation, effectively addressing issues related to spectral saturation. Residual analysis further confirmed that the VTFs model exhibited a more concentrated residual distribution, indicating significantly greater prediction stability than single-feature models. [Conclusions] The fusion of vegetation indices and texture features extracted from UAV-based multispectral imagery, combined with machine learning modeling, particularly the SVR algorithm, enabled high-accuracy, non-destructive estimation of LAI throughout the entire growth cycle of winter oilseed rape. Texture features effectively supplemented canopy structural information, showing strong complementarity, especially during high-LAI stages where spectral data is prone to saturation. Through mRMR feature selection and group-based cross-validation, the model demonstrated good generalizability and practical application potential. This method could provide reliable technical support for monitoring winter oilseed rape growth status and precision agriculture management. Future research would further incorporate canopy structural parameters or multi-temporal features to enhance the model's estimation capability during stages dominated by non-leaf organs.

Remote Sensing Monitoring Method of Cropping Index in Typical Open-Field Vegetable Production Areas |

ZHANG Yunxiang, WU Xuequn, HE Yonglin, MA Junwei

2025, 7(6):  174-184.  doi:10.12133/j.smartag.SA202508024

Asbtract ( 20 )   HTML ( 1)   PDF (32879KB) ( 2 )  

Figures and Tables | References | Related Articles | Metrics

[Objective] Most existing studies on cropping index extraction have primarily focused on cereal crops, whereas investigations targeting vegetable-based cropping systems remain relatively limited. Current national-scale cropping index products are largely designed for major cereal-producing regions, with model parameters calibrated according to the phenological characteristics and growth cycles of cereal crops. Such parameterization neglects the distinctive multi-season rotation patterns of open-field vegetable cultivation, leading to reduced accuracy in capturing the actual cropping dynamics of vegetable-growing areas. Consequently, these limitations hinder a comprehensive understanding of cropland utilization characteristics in regions characterized by intensive vegetable production. This study aims to improve the extraction method of the cropping index for vegetable cultivation systems and to reveal the characteristics of cropland use in typical regions with intensive vegetable multiple cropping. [Methods] Sentinel-2 MSI surface reflectance (SR) data released by the European space agency (ESA) was employed. Greedy algorithm was used to identify the optimal grid and orbit combination (47QRG, 61) covering Tonghai county, Yunnan Province. Based on this configuration, a total of 362 images acquired from 2020 to 2024 were compiled to construct a 5-day, 10 m spatial resolution time series. The normalized difference vegetation index (NDVI) time series was smoothed and reconstructed using the Whittaker smoothing (WS) method. The cropland extent was defined using the cropland mask from the GLC_FCS10 land cover dataset. Active croplands and greenhouse-covered areas were further identified using the vegetation-soil-pigment indices and synthetic-aperture radar (SAR) time-series images (VSPS) and the advanced plastic greenhouse index (APGI). Non-active croplands and greenhouse areas were excluded to refine the open-field cropland boundaries. Subsequently, the second-order difference method was applied to detect NDVI peaks in the reconstructed time series, with rule-based constraints used to eliminate false peaks. The number of valid peaks per pixel was then used to calculate the annual cropping index of Tonghai county from 2020 to 2024, and its spatial distribution and spatiotemporal variations were analyzed. [Results and Discussions] Compared with conventional 10-day median or maximum value compositing approaches, the time-series reconstruction based on specific grid and orbit combinations provides a more accurate representation of crop growth dynamics and peak patterns. Validation using 338 ground samples of cropping index obtained from field surveys in 2024 demonstrated an overall accuracy of 89.94%, a Kappa coefficient of 0.84, mean absolute error (MAE) of 0.11, and root mean square error (RMSE) of 0.36, indicating satisfactory reliability of the extracted results. From 2020 to 2024, the average cropping indices of croplands in Tonghai county were 221.45%, 217.80%, 275.37%, 232.41%, and 237.50%, respectively, reflecting a generally high level of land-use intensity. In 2020, 2021, 2023, and 2024, double cropping systems dominated, with triple cropping being secondary, whereas in 2022, triple cropping became predominant. Multi-season cropping (≥3 seasons) was mainly concentrated along the urban zones adjacent to Qilu lake, where abundant water resources provide favorable conditions for open-field vegetable cultivation. Interannual variations in the cropping index were largely driven by the alternation between double- and triple-cropping systems. Specifically, from 2020 to 2021, the cropping index decreased by 3.67%; cropland areas with decreased, unchanged, and increased indices accounted for 31.08%, 40.23%, and 28.69% of the total cropland area, respectively, with 10.45% of croplands shifting from triple to double cropping. From 2021 to 2022, the index increased substantially by 57.57%; decreased, unchanged, and increased areas accounted for 13.79%, 33.02%, and 53.18%, respectively, with 17.60% of croplands converting from double to triple cropping. Between 2022 and 2023, the index decreased by 42.96%, with corresponding area proportions of 47.96%, 35.06%, and 16.99%, and 16.62% of croplands shifting from triple to double cropping. From 2023 to 2024, the index slightly increased by 5.09%, with 27.49%, 39.19%, and 33.32% of croplands showing decreases, stability, and increases, respectively; 11.21% of croplands converted from double to triple cropping. Overall, the interannual variations were mainly influenced by the mutual transitions between double- and triple-cropping systems. [Conclusions] The results provide valuable theoretical and technical references for cropland resource management, optimization of regional vegetable production, and the promotion of sustainable agricultural development in Tonghai county.

Remote Sensing Extraction Method of Rice-Crayfish Fields Based on Dual-Branch and Multi-Scale Attention |

ZHANG Yun, ZHANG Lumin, XU Guangtao, HAO Jiahui

2025, 7(6):  185-195.  doi:10.12133/j.smartag.SA202507032

Asbtract ( 43 )   HTML ( 2)   PDF (3231KB) ( 1 )  

Figures and Tables | References | Related Articles | Metrics

[Objective] Rice-crayfish co-culture represents a highly efficient ecological agricultural system that simultaneously provides substantial economic returns and ecological benefits. Accurately obtaining spatial distribution information on rice-crayfish fields is of great importance for promoting the optimal allocation of agricultural resources, supporting ecological protection, and facilitating sustainable agricultural development. In regions characterized by complex terrain, fragmented plots, and diverse planting structures, traditional extraction approaches are often influenced by spectral confusion and indistinct field boundaries, making it difficult to achieve high-precision identification of rice-crayfish fields. The aim of this research is to develop a deep learning model that integrates multi-temporal and multi-scale feature information to improve the accuracy of remote sensing identification of rice-crayfish fields. [Methods] Multi-temporal GF-2 satellite imagery was employed as the primary data source, and a deep learning-based extraction method named the dual-branch attention pyramid network (DBAP-Net) was developed. The proposed model was established upon the U-Net framework and designed with a dual-branch encoder architecture to extract temporal features from different phenological stages, thereby fully capturing the spectral variations of rice-crayfish fields between the paddy flooding and rice-growing periods. The convolutional block attention module (CBAM) was embedded in each encoder layer and in the skip connections to adaptively adjust feature weights along both the channel and spatial dimensions, enhancing the network's capacity to emphasize critical spatial structures of rice-crayfish fields while effectively suppressing background noise and redundant information. During the feature fusion stage, an atrous spatial pyramid pooling (ASPP) module was incorporated to aggregate contextual information from multiple receptive fields through multi-scale atrous convolutions, improving the model's capability for multi-scale spatial information representation. [Results and Discussions] A quantitative performance evaluation of each DBAP-Net component was conducted through ablation experiments. The results demonstrated that the introduction of a dual-branch structure improved overall accuracy (OA), F₁-Score, intersection over union (IoU), and Matthews correlation coefficient (MCC) by 0.48, 0.57, 0.94, and 0.99 percentage points, respectively. Incorporating the CBAM module further enhanced these metrics by 0.52, 0.85, 1.40, and 1.20 percentage points, while the addition of the ASPP module yielded further increases of 0.59, 1.09, 1.81, and 1.49 percentage points, respectively. The DBAP-Net model achieved the highest comprehensive performance, with an OA of 94.45%, F₁-Score of 91.79%, IoU of 84.82%, and MCC of 87.60%. These values represented respective improvements of 1.83, 2.55, 4.25, and 3.98 percentage points compared with the baseline U-Net model. These findings indicated that each enhancement module made a substantial contribution to improving both feature representation and spatial boundary delineation. DBAP-Net was further compared with five representative semantic segmentation networks, U-Net, PSPNet, DeepLabV3+, SegFormer, and TransUNet, to comprehensively evaluate its generalization and segmentation performance. The results demonstrated that DBAP-Net consistently achieved higher overall accuracy, precision, F₁-Score, and IoU than all other comparison models. Specifically, compared with U-Net, PSPNet, and DeepLabV3+, the F₁-Scores increased by 2.55, 2.98, and 2.75 percentage points, while the IoU values improved by 4.25, 4.96, and 4.57 percentage points, respectively. In comparison with the more recent models, SegFormer and TransUNet, DBAP-Net's F₁-Score was higher by 3.09 and 1.67 percentage points, and its IoU was enhanced by 5.12 and 2.81 percentage points. Visualization of the segmentation results further revealed that DBAP-Net produced clear segmentation boundaries with complete fields, significantly reducing misclassification and omission rates. In contrast, other models exhibited varying degrees of boundary blurring and fragmentation. When applied across the entire study area, DBAP-Net demonstrated strong robustness and stability. The model achieved an overall accuracy of 96.00%, with a Kappa coefficient of 0.920. The producer's accuracy and user's accuracy were 95.13% and 97.38%, respectively. Compared with traditional approaches such as the seasonal water-difference method, random forest classification, and temporal index thresholding, DBAP-Net significantly improved both extraction precision and spatial completeness, particularly under conditions of complex terrain and fragmented agricultural landscapes. [Conclusions] DBAP-Net, by integrating multi-temporal spectral and multi-scale spatial information, significantly improves the accuracy and completeness of remote sensing extraction of rice-crayfish fields from high-resolution remote sensing imagery. The model provides a reliable and adaptable technical framework for fine-scale monitoring, precise mapping, and sustainable management of rice-crayfish co-culture systems, offering valuable methodological support for agricultural resource assessment and ecological protection.

Parcel-Scale Crop Distribution Mapping Based on Stacking Ensemble Learning |

XIE Wenhao, ZHANG Xin, DONG Wen, ZHENG Yizhen, CHENG Bo, TU Wenli, SUN Fengqing

2025, 7(6):  196-209.  doi:10.12133/j.smartag.SA202509003

Asbtract ( 71 )   HTML ( 4)   PDF (8396KB) ( 6 )  

Figures and Tables | References | Related Articles | Metrics

[Objective] With the widespread availability of high-resolution and multi-source remote sensing data, remote sensing-based crop classification has played an increasingly vital role in agricultural monitoring, yield estimation, and land use management. However, traditional pixel-level classification methods often struggle to achieve stable, high-precision classification under conditions of intra-plot heterogeneity, spectral confusion, and noise interference.Therefore, this study aimed to improve parcel-level crop classification accuracy and spatial consistency by constructing a multi-source feature fusion and ensemble learning framework, which exploits complementary spectral, spatial, temporal, and productivity characteristics to enhance robustness and generalization in multi-crop classification tasks. [Methods] To enhance field-level classification accuracy and spatial consistency, a crop classification method integrating field-scale feature extraction, feature selection, and Stacking ensemble learning was proposed and validated. This approach aimed to fully leverage the complementarity of spectral, spatial, and temporal information through feature engineering and model fusion. The study area was located in Feicheng city, Shandong province. The data included multi-temporal Sentinel-2 optical imagery, Sentinel-1 SAR data, Gaofen remote sensing imagery, and parcel vector samples with crop_type attributes. All imagery underwent radiometric and atmospheric correction, projection registration, and cropping during preprocessing to ensure spatial consistency and temporal correspondence across sensors. The dataset was constructed at the field level, comprising 3 200 fields for model training and independent validation. This study systematically constructed four types of meta-features on the plot scale: raw bands and vegetation indices; spatial meta-features, including texture, morphology, and structural indicators calculated from high-resolution imagery to reflect internal spatial heterogeneity; temporal sequence meta-features, extracting vegetation indices, backscatter, and harmonic/temporal statistics from multi-temporal optical and SAR imagery to characterize crop growth cycles; crop primary productivity features to highlight differences in carbon fixation and biomass accumulation among crops. Subsequently, for the high-dimensional, multi-source feature set, a combined strategy of LightGBM and recursive feature elimination (RFE) was employed for feature importance assessment and selection. This retained a subset of features most critical to classification, enhancing model generalization and computational efficiency. Within the classification framework, a Stacking-based ensemble learning model was constructed. Base learners included random forest (RF), eXtreme gradient boosting (XGB), support vector machine (SVM), gradient boosting (GB), categorical boosting (CatBoost), adaptive boosting (ADA), back propagation (BP), K-nearest neighbors (KNN), and light gradient boosting machine (LightGBM). These base models learn and represent plot features from distinct perspectives, fully exploring nonlinear relationships among spectral, spatial, and temporal characteristics. During the meta-learner selection phase, to compare the impact of different feature fusion strategies on classification performance, XGBClassifier, LightGBMClassifier, MLPClassifier, and LogisticRegression were selected as meta-learners for experimental comparison. By contrasting the classification outcomes of different meta-models under the same base model outputs, their contributions to improving feature fusion accuracy and stability differences were analyzed. During model training, hierarchical cross-validation was employed to mitigate bias caused by class imbalance. Overall accuracy (OA), Kappa coefficient, and F₁-Score served as primary evaluation metrics, while recall and precision rates for each crop category underwent systematic analysis. [Results and Discussions] The findings indicated that feature selection significantly impacted classification performance. By integrating LightGBM with feature selection strategies, a subset of 102 optimal features was identified. This subset included gross primary production (GPP), spectral features, vegetation indices, textural features, temporal features, and harmonic features. This approach effectively mitigated feature redundancy and multicollinearity issues, enhancing model stability and generalization capability. Among these, GPP-related features and vegetation indices from key growth stages demonstrated high discriminative power in distinguishing crop categories, fully reflecting the close coupling between remote sensing features and crop phenological information. The Stacking ensemble strategy demonstrated outstanding classification performance. Among various meta-learners, the Stacking model with XGBClassifier as the final learner achieved the highest classification accuracy (OA = 95.66%, Kappa = 0.900 6), showcasing exceptional ensemble generalization capability. It performed particularly well in identifying major crops like maize while maintaining good adaptability for less common crops. The method's advantage extended beyond accuracy gains to its comprehensive integration of complementary spectral, temporal, and spatial feature processing capabilities across base learners. The meta-learner adaptively synthesized multi-model outputs, enhancing classification stability and spatial consistency. Compared to traditional pixel-level classification followed by parcel reclassification, direct feature extraction and classification based on vector parcels effectively avoided edge blending and noise interference inherent in pixel-level methods, significantly improving parcel recognition stability and accuracy. Experimental results demonstrated that parcel-level classification outperformed pixel-level strategies in overall accuracy and Kappa coefficient, with superior spatial consistency and robustness in classification outcomes. [Conclusions] The "optimal feature subset + Stacking ensemble learning + parcel-level classification" method developed in this research demonstrates outstanding accuracy and stability in multi-source remote sensing crop identification, providing an efficient and feasible technical pathway for parcel-level classification in complex agricultural landscapes. Future work will integrate high-resolution time-series data with deep learning models to further enhance the method's cross-regional adaptability and crop monitoring capabilities.

Analysis of the Spatiotemporal Evolution and Driving Forces of Rural Settlements in Relation to Terrain Differences |

LIU Miao, ZHANG Jiayi, LI Zhenhai, CHEN Jing

2025, 7(6):  210-224.  doi:10.12133/j.smartag.SA202509027

Asbtract ( 34 )   HTML ( 0)   PDF (4559KB) ( 0 )  

Figures and Tables | References | Related Articles | Metrics

[Objective] With the in-depth implementation of the rural revitalization strategy and the rapid progress of China's urbanization, the urban-rural spatial structure has undergone significant restructuring. Rural settlements, as the fundamental spatial units of rural areas, have witnessed remarkable changes in scale, layout, and function. Accurately identifying their spatiotemporal evolution and clarifying the driving forces is essential for comprehending urban-rural transformation, optimizing territorial spatial planning, and promoting coordinated development. However, current research still has limitations. Many studies primarily concentrate on describing spatial patterns, overlooking the underlying processes and regional disparities. Comparative analyses between different topographic regions, such as plains and hilly areas, are inadequate, resulting in an incomplete understanding of the impacts of terrain. Moreover, investigations into multi-scale and multi-factor driving mechanisms are relatively weak. Therefore, the aim of this research is to systematically uncover the spatiotemporal evolution of rural settlements under various topographic conditions and to quantitatively identify the key drivers shaping these patterns. [Methods] The primary research regions were Laoling City in Shandong Province, representing typical plain terrain, and Yi'an District in Anhui Province, characterized by hilly landforms. This selection fully takes into account how variations in topographic conditions influence the long - term evolution of rural settlement patterns. Based on the remote - sensing mapping results of rural settlements in 2002, 2012, and 2022, a systematic analysis of the spatial distribution characteristics and temporal differentiation of settlement patterns was conducted. Using geographic information system (GIS) spatial analysis as the analytical foundation and integrating methods such as landscape pattern indices, centroid migration analysis, and spatial pattern change detection, the spatiotemporal evolution trajectories of rural settlements were revealed under contrasting geomorphic settings from the perspectives of overall spatial configuration, internal structural features, and dynamic change processes. In addition, the geographical detector model was employed to quantitatively assess ten potential driving factors, including natural environmental conditions, socio - economic development indicators, transportation accessibility, and location - related attributes. [Results and Discussions] On the county scale, in the plain area, the largest patch index increased from 0.88 to 2.46, while the average nearest neighbor ratio (NNR) decreased from 0.99 to 0.90. This indicates that the settlement size expanded, the structure became more centralized, and the degree of clustering continuously strengthened. In contrast, in the hilly area, the patch density (PD) decreased from 9.16 to 2.77, and the NNR increased from 0.50 to 0.69. This suggests that the number of settlements declined and their spatial structure evolved from highly clustered to relatively dispersed. At the village scale, there were significant differences in the evolution trends between the two regions. In the plain area, changes in rural settlement areas were relatively balanced, with similar proportions of villages experiencing expansion and contraction. Settlements mainly exhibited a block - like distribution, extending along roads. In contrast, in the hilly area, the expansion of rural settlements was more pronounced, with over 70% of villages showing an increase in area. Settlements primarily displayed a linear distribution pattern, extending along rivers and valleys. Over the 20 - year period, the driving mechanisms of rural settlement evolution in the plain area shifted from being dominated by natural and economic factors to being dominated by land resource and safety factors. The driving power (q value) of distance to cultivated land and distance to the urban center increased by 0.51 and 0.39, respectively, becoming the main growth factors. In the hilly area, settlement evolution became increasingly constrained by topography, water resources, and geological safety. The driving power of distance to rivers increased from 0.22 to 0.45, and the dominant driving interaction shifted from "cultivated land-scenic spot" to "geological hazard point–scenic spot", reflecting a more complex driving mechanism. [Conclusions] Different topographic regions exhibit distinct spatial pattern characteristics and evolutionary driving mechanisms at both the county and village scales. Rural settlements in plain areas tend to demonstrate higher degrees of clustering, more regular morphologies, and relatively stable evolutionary processes. In contrast, settlements in hilly areas are more scattered and fragmented due to topographic constraints and resource limitations, and their evolutionary processes are more intricate. This study not only deepens the understanding of rural settlement evolution but also offers scientific support for the localized development of smart agriculture and the reconstruction of rural spatial systems.

Cross-validation Study on the Authenticity of Agricultural Insurance Underwriting Based on Multi-Source Satellite Remote Sensing Data: Taking Multi-Season Rice in M ​​County, S Province as A Case |

CHEN Ailian, ZHANG Rusheng, LI Ran, ZHAO Sijian, ZHU Yuxia, LAI Jibao, SUN Wei, ZHANG Jing

2025, 7(6):  225-236.  doi:10.12133/j.smartag.SA202507034

Asbtract ( 25 )   HTML ( 2)   PDF (1911KB) ( 1 )  

Figures and Tables | References | Related Articles | Metrics

[Objective] Rice, wheat, and corn, account for over 50% of government-subsidized premiums. Therefore, ensuring the authenticity of insurance underwriting data of three major staple crops is crucial for safeguarding fiscal funds and promoting the high-quality development of agricultural insurance. Currently, verifying the authenticity of underwriting data relies on remote sensing technology to achieve high-precision, high-efficiency, and low-cost crop identification. However, in the multi-season rice-growing areas of southern China, remote sensing identification still suffers from insufficient accuracy and delayed timeliness. This study, targeting the actual business needs of agricultural insurance, explores a "fast, accurate, and low-cost" identification method for multi-season rice in the hilly areas of Southern China. Cross-validation of underwriting data authenticity is conducted based on case studies, providing technical support for fiscal fund security and the high-quality development of agricultural insurance. [Methods] The high-resolution remote sensing data of China was integrated with internationally available data. First, a deep learning algorithm and high-resolution imagery were used to rapidly extract cultivated land parcels as classification units. Combined with field sampling and samples derived from high resolution imagery, rice classification and identification were performed on the Google Earth Engine (GEE) platform using Sentinel-1 radar data and Sentinel-2 multispectral data. Three methods, random forest, support vector machine, and classification and regression tree, were compared, and the optimal model result was selected for cross-validation of rice insurance data. Validation metrics included four categories: area difference (AD), the difference between remote sensing and insured area, cover ratio (CR), crop insurance coverage, overlapping ratio (OR), overlap of policy parcels, and crop proportion (CP), crop proportion within policy parcels. [Results and Discussions] Based on the cultivated land units extracted using deep learning, a classification feature set was constructed by integrating Sentinel-1 radar polarimetric signatures from March to October with Sentinel-2 multi-spectral reflectance and NDVI from July and August. The random forest model achieved 0.93 identification accuracy, meeting the accuracy and cost requirements for verifying mid- and late-season rice insurance data. However, due to the lack of multi-spectral data at key time phases required for early rice identification, its identification timeliness is poor and is only suitable for post-warning and deterrence the following year. Cross-validation results show that the county's overall insured area is basically the same as the remote sensing identification area, but there are significant differences in township scale: Among the 33 townships, 14 towns have AD more than 10 000 hectare, indicating false insurance. As for CR, 10 townships have a crop insurance coverage rate of more than 1, 1 township has a CR less than 0.4, and another 2 townships have not carried out insurance. As for policy parcels, the 31 townships with insurance records, 10 did not provide plot data; among the 21 townships that provided data, 3 townships had an overlap rate of more than 40% for more than 50% of the policy plots, and 14 townships had an overlap rate of more than 40% for more than 20% of the policy plots. These results suggest that some areas may have problems such as false insurance, duplicate insurance, or non-standard operations, and regulatory authorities need to intervene and verify in a timely manner. [Conclusions] A technical system suitable is developed for rapid remote sensing identification and cross-validation with insurance data. Its feasibility in practical regulatory applications has been verified, providing effective methodological support for authenticity verification and precise regulation of agricultural insurance underwriting.