Loading...
Welcome to Smart Agriculture 中文

Table of Content

    30 March 2021, Volume 3 Issue 1
    Topic--Frontier Technology and Application of Agricultural Phenotype
    Research Status and Prospect on Height Estimation of Field Crop Using Near-Field Remote Sensing Technology | Open Access
    ZHANG Jian, XIE Tianjin, YANG Wanneng, ZHOU Guangsheng
    2021, 3(1):  1-15.  doi:10.12133/j.smartag.2021.3.1.202102-SA033
    Asbtract ( 1909 )   HTML ( 357)   PDF (1983KB) ( 957 )  
    Figures and Tables | References | Related Articles | Metrics

    Plant height is a key indicator to dynamically measure crop health and overall growth status, which is widely used to estimate the biological yield and final grain yield of crops. The traditional manual measurement method is subjective, inefficient, and time-consuming. And the plant height obtained by sampling cannot evaluate the height of the whole field. In the last decade, remote sensing technology has developed rapidly in agriculture, which makes it possible to collect crop height information with high accuracy, high frequency, and high efficiency. This paper firstly reviewed the literature on obtaining plant height by using remote sensing technology for understanding the research progress of height estimation in the field. Unmanned aerial vehicle (UAV) platform with visible-light camera and light detection and ranging (LiDAR) were the most frequently used methods. And main research crops included wheat, corn, rice, and other staple food crops. Moreover, crop height measurement was mainly based on near-field remote sensing platforms such as ground, UAV, and airborne. Secondly, the basic principles, advantages, and limitations of different platforms and sensors for obtaining plant height were analyzed. The altimetry process and the key techniques of LiDAR and visible-light camera were discussed emphatically, which included extraction of crop canopy and soil elevation information, and feature matching of the imaging method. Then, the applications using plant height data, including the inversion of biomass, lodging identification, yield prediction, and breeding of crops were summarized. However, the commonly used empirical model has some problems such large measured data, unclear physical significance, and poor universality. Finally, the problems and challenges of near-field remote sensing technology in plant height acquisition were proposed. Selecting appropriate data to meet the needs of cost and accuracy, improving the measurement accuracy, and matching the plant height estimation of remote sensing with the agricultural application need to be considered. In addition, we prospected the future development was prospected from four aspects of 1) platform and sensor, 2) bare soil detection and interpolation algorithm, 3) plant height application research, and 4) the measurement difference of plant height between agronomy and remote sensing, which can provide references for future research and method application of near-field remote sensing height measurement.

    Study on the Micro-Phenotype of Different Types of Maize Kernels Based on Micro-CT | Open Access
    ZHAO Huan, WANG Jinglu, LIAO Shengjin, ZHANG Ying, LU Xianju, GUO Xinyu, ZHAO Chunjiang
    2021, 3(1):  16-28.  doi:10.12133/j.smartag.2021.3.1.202103-SA004
    Asbtract ( 847 )   HTML ( 85)   PDF (2085KB) ( 371 )  
    Figures and Tables | References | Related Articles | Metrics

    Plant micro-phenotype mainly refers to the phenotypic information at the tissue, cell, and subcellular levels, which is an important part of plant phenomics research. In view of the problems of low efficiency, large error, and few traits of traditional methods for detecting kernel microscopic traits, Micro-CT scanning technology was used to carry out precise identification of micro-phenotype on 11 varieties of maize kernels. A total of 34 microscopic traits were obtained based on CT sequence images of 7 tissues, including seed, embryo, endosperm, cavity, subcutaneous cavity, endosperm cavity and embryo cavity. Among the 34 microscopic traits, 4 traits, including endosperm cavity surface area, kernel volume, endosperm volume ratio and endosperm cavity specific surface area, were significantly different among maize types (P-value<0.05). The surface area of endosperm cavity and kernel volume of common maize were significantly higher than those of other types of maize. The specific surface area of endosperm cavity of high oil maize was the largest. The endosperm cavity of sweet corn had the smallest specific surface area. The endosperm volume ration of popcorn was the largest. Furthermore, 34 traits were used for One-way ANOVA and cluster analysis, and 11 different maize varieties were divided into four categories, of which the first category was mainly common maize, the second category was mainly popcorn, the third category was sweet corn, and the fourth category was high oil maize. The results indicated that Micro-CT scanning technology could not only achieve precise identification of micro-phenotype of maize kernels, but also provide supports for kernel classification and variety detection, and so on.

    Estimation of Maize Leaf Area Index and Aboveground Biomass Based on Hyperspectral Data | Open Access
    SHU Meiyan, CHEN Xiangyang, WANG Xiqing, MA Yuntao
    2021, 3(1):  29-39.  doi:10.12133/j.smartag.2021.3.1.202102-SA004
    Asbtract ( 644 )   HTML ( 43)   PDF (1944KB) ( 364 )  
    Figures and Tables | References | Related Articles | Metrics

    In order to assess maize growth status accurately and quickly for improving maize precise management, field experiment was conducted in Gongzhuling research station, Jilin Academy of Agricultural Sciences, Jilin province. Experimental design included 3 planting densities and 5 maize materials. The near-ground hyperspectral data and the unmanned aerial vehicle (UAV) hyperspectral images were obtained when maize were during V11-V12 stage. The application abilities of the hyperspectral data obtained from the two phenotyping platforms were compared and analyzed in the estimation of maize leaf area index (LAI) and aboveground biomass. In this study, 21 commonly used spectral vegetation indices were constructed based on ground hyperspectral data, and then the estimation models of maize LAI and aboveground biomass were established based on ground hyperspectral full-bands, UAV hyperspectral full-bands and vegetation indices and partial least square regression method, respectively. According to the variance estimation of regression coefficients, the important bands of LAI and aboveground biomass were selected, and the partial least square method was also used to establish the estimation model of maize LAI and aboveground biomass based on important bands. The results showed that the canopy spectral reflectance of the same maize material increased with the increase of planting density in the near infrared bands. Among the 5 maize materials under the same planting density, the canopy spectral reflectance of wild type material was the lowest in the visible and near infrared bands. For LAI, the model constructed based on vegetation indices had the best estimation result, with R2, RMSE and rRMSE values of 0.70, 0.92 and 15.94%. For aboveground biomass, the model constructed based on the sensitive spectral bands (839-893 nm and 1336-1348 nm) had the best estimation results, with R2, RMSE and rRMSE values of 0.71, 12.31 g and 15.89%, which showed that there was information redundancy in hyperspectral bands in the estimation of aboveground biomass, and the estimation accuracy could be improved by reducing the number of spectral bands and selecting sensitive spectral bands. In summary, the UAV hyperspectral images have a good application ability in the estimation of maize LAI and aboveground biomass, and can quickly and effectively extract the parameters information of maize growth. For specific parameters, sensitive spectral bands selected can provide reliable basis for the development and practical application of multi-spectrum in the future. The study can provide a reference for the use of hyperspectral technology in the management of precision agriculture at the community scale.

    Vertical Heterogeneity Analysis of Biochemical Parameters in Oilseed Rape Canopy Based on Fast Chlorophyll Fluorescence Technology | Open Access
    ZHANG Jiafei, WAN Liang, HE Yong, CEN Haiyan
    2021, 3(1):  40-50.  doi:10.12133/j.smartag.2021.3.1.202103-SA005
    Asbtract ( 535 )   HTML ( 15)   PDF (3024KB) ( 324 )  
    Figures and Tables | References | Related Articles | Metrics

    Accurate acquisition of crop canopy biochemical information is of great significance for monitoring crop growth and guiding precise fertilization. Previous vertical distribution researches of crop biochemical information were mainly based on hyperspectral inversion, which was lack of the association of plant photosynthesis physiology. This study mainly investigated the vertical distribution characteristics of biochemical parameters such as chlorophyll, carotenoid, dry matter, and water content in the oilseed rape canopy under different nitrogen treatments at the mid-seedling stage. The photosynthetic performance of leaves was measured by using fast chlorophyll fluorescence technology, and linear regression and principal component analysis were further implemented to explore the internal relationship between fluorescence response and biochemical parameters. The results showed that: (1) The chlorophyll content, carotenoid content, dry matter and water content of the rape canopy at the mid-seedling stage all showed a parabolic vertical distribution, while the ratio of chlorophyll to carotenoids content gradually decreases with the leaf position and nitrogen treatments, which was the same as the vertical distribution pattern of fluorescence parameters such as driving force comprehensive performance (DFTotal) and end electron chain quantum yield (φRo) and other fluorescence parameters could be used to diagnose nitrogen stress; (2) JIP-test parameters, especially DFTotal, had a good performance to evaluate the chlorophyll/carotenoids, chlorophyll and dry matter content of oilseed rape leaves; (3) Nitrogen deficiency would weaken the PSII and PSI performance of oilseed rape leaves at the mid-seedling stage, and the maximum photochemical efficiency (φPo) could be used to diagnose nitrogen stress. There was a significant difference in the PSI performance, namely electron transfer efficiency at the end acceptors of leaves in the different leaf position, hence the comprehensive performance parameter DFTotal could be an effective characterization of the vertical heterogeneity of canopy biochemical parameters. These findings indicated the feasibility of applying the rapid chlorophyll fluorescence technology to crop biochemical information heterogeneity monitoring and provided new ideas and technical support for guiding precise fertilization and achieving high-quality and high-yield.

    Cotton Phenotypic Trait Extraction Using Multi-Temporal Laser Point Clouds | Open Access
    YANG Xu, HU Songtao, WANG Yinghua, YANG Wanneng, ZHAI Ruifang
    2021, 3(1):  51-62.  doi:10.12133/j.smartag.2021.3.1.202102-SA003
    Asbtract ( 665 )   HTML ( 40)   PDF (2561KB) ( 383 )  
    Figures and Tables | References | Related Articles | Metrics

    To cope with the challenges posed by the rapid growth of world population and global environmental changes, scholars should employ genetic and phenotypic analyses to breed crop varieties with improved responses to limited resource environments and soil conditions to increase crop yield and quality. Therefore, the efficient, accurate, and non-destructive measurement of crop phenotypic traits, and the dynamic quantification of phenotypic traits are urgently needed for crop phenotypic research, and breeding as well as for modern agricultural development. In this study, cotton plants were taken as research objects, and the multi-temporal point cloud data of cotton plants were collected by using three-dimensional laser scanning technology. The multi-temporal point clouds of three cotton plants at four time points were collected. First, RANSAC algorithm was implemented for main stem extraction on the original point cloud data of cotton plants, then region growing based clustering was carried out for leaf segmentation. Plant height was estimated by calculating the end points of the segmented main stem. Leaf length and width measurements were conducted on the segmented leaf parts. In addition, the volume was also estimated through the convex hull of the original point cloud of plant cotton. Then, multi-temporal point clouds of plants were registered, and organ correspondence was constructed with the Hungarian method. Finally, dynamic quantification of phenotypic traits including plant volume, plant height, leaf length, leaf width, and leaf area were calculated and analyzed. The overall performance of the approaches achieved a matching rate through a series of experiments, and the traits extracted by using of point cloud showed high correlation with the manually measured ones. The relative error between plant height and manual measurement results did not exceed 1.0%. The estimated leaf length and width on point clouds were highly correlated with the manually measured ones, and the coefficient of determination was nearly 1.0. The proposed 3D phenotyping methodology can be introduced and used to other crops for phenotyping.

    Foxtail Millet Ear Detection Approach Based on YOLOv4 and Adaptive Anchor Box Adjustment | Open Access
    HAO Wangli, YU Peiyan, HAO Fei, HAN Meng, HAN Jiwan, SUN Weirong, LI Fuzhong
    2021, 3(1):  63-74.  doi:10.12133/j.smartag.2021.3.1.202102-SA066
    Asbtract ( 960 )   HTML ( 78)   PDF (2620KB) ( 466 )  
    Figures and Tables | References | Related Articles | Metrics

    谷穗的检测和计数对于预测谷子产量和育种至关重要。但是,传统的谷穗计数主要基于人工统计,既费时又费力。为解决上述问题,本研究首先建立了一个包含784张图像和10,000个谷穗样本的谷穗检测数据集。提出了一种基于YOLOv4和自适应锚框调整的谷穗检测方法,可快速准确地检测特定框中的谷穗。通过自适应地调整锚框,可生成符合谷穗目标的候选框,从而提升检测的准确率。为验证该方法的有效性,采用了多个标准,包括平均精度(mAP),F1得分(F1-Score),精度(Precision)和召回率(Recall)进行评价。此外,设计了对比试验验证所提出方法的有效性,包括与其他模型(YOLOv2,YOLOv3和Faster-RCNN)进行比较来评估模型的性能,评估模型在不同交并比(IOU)取值下的性能,评估模型在自适应锚框调整下的谷穗检测性能,评估引起模型评价标准变化的原因,以及评估模型在不同原始输入图像尺寸下的性能。试验结果表明,YOLOv4获得了良好的谷穗检测性能。YOLOv4的mAP达到78.99%,F1-score达到83.00%,Precision达到87%和Recall达到79.00%,在所有评价标准上均比其他比较模型高出8%。试验结果表明,该方法具有较好的准确性和高效性。

    Tassel Segmentation of Maize Point Cloud Based on Super Voxels Clustering and Local Features | Open Access
    ZHU Chao, WU Fan, LIU Changbin, ZHAO Jianxiang, LIN Lili, TIAN Xueying, MIAO Teng
    2021, 3(1):  75-85.  doi:10.12133/j.smartag.2021.3.1.202102-SA001
    Asbtract ( 720 )   HTML ( 32)   PDF (1993KB) ( 333 )  
    Figures and Tables | References | Related Articles | Metrics

    Accurate and high-throughput maize plant phenotyping is vital for crop breeding and cultivation research. Tassel-related phenotypic parameters are important agronomic traits. However, fully automatic and fine tassel organ segmentation of maize shoots from three-dimensional (3D) point clouds is still challenging. To address this issue, a tassel point cloud segmentation method based on point cloud super voxels clustering and local geometric features was proposed in this study. Firstly, the undirected graph of the maize plant point cloud was established, the edge weights were calculated by using the difference of normal vectors, and the spectral clustering method was used to cluster the point cloud to form multiple super voxel sub-regions. Then, the principal component analysis method was used to find the two end regions of the plant and based on the observation of the straight direction of the bottom stem regions, the top and bottom regions were distinguished by the point cloud linear features. Finally, the tassel points were identified based on the plane local features of the point cloud. The sub-regions of the top region of the plant were classified into leaf regions, tassel regions, and mixed regions by plane local features of the point cloud, the tassel points in the tassel sub-region, and the mixed region were the finally segmented tassel point clouds. In this study, 15 mature maize plants with 3 point cloud densities were tested. Compared with the ground truth segmented manually, the average F1 scores of the tassel segmentation were 0.763, 0.875 and 0.889 when the point cloud density was 0.8/cm, 1.3/cm, and 1.9/cm, respectively. The segmentation accuracy of this method increased with the increase of plant point cloud density. The increase of point cloud density and the number of point clouds mainly affected the calculation results of point cloud plane features in tassel segmentation. When the number of point clouds was small, the top leaf point cloud was relatively sparse. Therefore, the difference between the plane feature of the leaf point and the plane feature of the tassel point was not obvious, which led to the increase of the misclassification of the point cloud. However, the time complexity of the algorithm was O(n3), so the increase in the density and number of point clouds would lead to a significant increase in the running time. Considering the segmentation accuracy and running time, the research obtained the best effect on the mature maize plants with a point cloud density of 1.3/cm and an average number of 15,000. The segmentation F1 score reached 0.875 and the running time was 6.85 s. The results showed that this method could extract tassels from maize plant point cloud, and provided technical support for the research and application of high-throughput phenotyping and three-dimensional reconstruction of maize.

    Detection and Grading Method of Pomelo Shape Based on Contour Coordinate Transformation and Fitting | Open Access
    LI Yan, SHEN Jie, XIE Hang, GAO Guangyin, LIU Jianxiong, LIU Jie
    2021, 3(1):  86-95.  doi:10.12133/j.smartag.2021.3.1.202102-SA007
    Asbtract ( 812 )   HTML ( 15)   PDF (1652KB) ( 271 )  
    Figures and Tables | References | Related Articles | Metrics

    Automatic grading method of pomelo fruit according to the shape and size is urgently needed in the industry since the work mainly depends on artificial judgment currently. In this research, a method, which detected the vertical and horizontal size of pomelo by using contour coordinate transformation fitting, fruit shape feature extraction and direction angle compensation algorithm, while it determined the shape defects based on fruit shape index, was proposed. The image acquisition system was self-designed and built up with a CMOS camera, a dot matrix LED light source, a plane mirror, the computer, a box and brackets. The image data containing whole surface information of Shatian pomelo samples with different sizes and shapes were collected by this system. The G-B component grayscale image was chosen for denoising and segmentation. The Laplacian edge detection algorithm was implemented to extract the edge pixels of the fruit. The polynomial fitting method was applied to converse the rectangular coordinates to polar coordinates so that the fruit shape description was simplified. The characteristic point polar angle value was used to compensate the random direction of the vertical and horizontal diameters of the sample. Then the vertical and horizontal diameters of fruit were calculated after classifying the sample shapes into the spherical and the pear-like categories. For the involved 168 pomelo samples, the average error, maximum absolute error and average relative error of the vertical diameters were 2.23 mm, 7.39 mm and 1.6% respectively, while these parameters of the horizontal diameters were 2.21 mm, 7.66 mm and 1.4% respectively. The fruit shape discriminant model was established by using BP neural network algorithm based on the seven features extracted from the fitting function and verified by independent validation set including 3 peak heights, 3 peak widths and 1 trough value difference. The total recognition rate of shape identification was 83.7%. The results illustrated that the method had the potential to measuring the pomelo size and shape for grading fast and non-destructively.

    Information Processing and Decision Making
    Improved AODV Routing Protocol for Multi-Robot Communication in Orchard | Open Access
    MAO Wenju, LIU Heng, WANG Dongfei, YANG Fuzeng, LIU Zhijie
    2021, 3(1):  96-108.  doi:10.12133/j.smartag.2021.3.1.202101-SA001
    Asbtract ( 507 )   HTML ( 25)   PDF (2632KB) ( 285 )  
    Figures and Tables | References | Related Articles | Metrics

    To satisfy the communication needs of multiple robots working in orchards, an improved Ad Hoc on-demand distance vector routing protocol based on signal strength threshold and priority nodes (AODV-SP), and the prediction model of Wi-Fi signal reception in peach orchards, was proposed in this study. Different from the traditional AODV protocol, AODV-SP utilizes the idea of priority nodes and strength thresholds to construct a discovery routing algorithm and a selection routing algorithm by seeking priority nodes and calculating the maximum strength threshold between nodes, respectively. The discovery routing message and selection routing message of the AODV-SP protocol were designed according to the discovery routing and selection routing algorithms. To verify the performance of the AODV-SP protocol, the performance of the protocol with different maximum movement speeds of nodes was analyzed by using NS2 simulation software and the performance was compared with the traditional AODV protocol. The simulation results showed that the average end-to-end delay, route initiation frequency, and route overhead of AODV-SP protocol with the introduction of priority node and path signal strength thresholds were smaller than those of the traditional AODV protocol, and the packet delivery rate improved significantly compared with that of AODV protocol. Among them, when the maximum node movement speed was 5 m/s, the route initiation frequency and route overhead of AODV-SP protocol reduced by 3.65% and 7.09%, respectively, compared with AODV protocol. When the maximum node movement speed was 8 m/s, the packet delivery rate of AODV-SP protocol improved by 0.59% and the average end-to-end delay reduced by 13.09%. To further verify the simulation results of AODV-SP making AODV-SP protocol applicable to a multi-robot wireless communication system and ensure the normal operation of multi-robot wireless communication in orchards, a physical platform for multi-robot wireless communication was built in a laboratory environment, and software was designed to enable the physical platform to communicate properly under the AODV-SP protocol. And the physical platform for multi-robot wireless communication using the AODV-SP protocol was tested under static and dynamic conditions, respectively. The experiment results showed that, under static condition, when distance between nodes was less than or equal to 25 m, the packet loss rate of the robot was 0; when distance between nodes was 100 m, tthe packet loss rate of the robot was 21.01%, and the following robots could maintain the chain topology with the leader robot in dynamic conditions. Simulation and physical platform experiments results showed that the AODV-SP protocol could be used for the construction of multi-robot communication systems in orchard.

    Distilled-MobileNet Model of Convolutional Neural Network Simplified Structure for Plant Disease Recognition | Open Access
    QIU Wenjie, YE Jin, HU Liangqing, YANG Juan, LI Qili, MO Jianyou, YI Wanmao
    2021, 3(1):  109-117.  doi:10.12133/j.smartag.2021.3.1.202009-SA004
    Asbtract ( 1012 )   HTML ( 72)   PDF (1643KB) ( 558 )  
    Figures and Tables | References | Related Articles | Metrics

    The development of convolutional neural networks(CNN) has brought a large number of network parameters and huge model volumes, which greatly limites the application on devices with small computing resources, such as single-chip microcomputers and mobile devices. In order to solve the problem, a structured model compression method was studied in this research. Its core idea was using knowledge distillation to transfer the knowledge from the complex integrated model to a lightweight small-scale neural network. Firstly, VGG16 was used to train a teacher model with a higher recognition rate, whose volume was much larger than the student model. Then the knowledge in the model was transfered to MobileNet by using distillation. The parameters number of the VGG16 model was greatly reduced. The knowledge-distilled model was named Distilled-MobileNet, and was applied to the classification task of 38 common diseases (powdery mildew, Huanglong disease, etc.) of 14 crops (soybean, cucumber, tomato, etc.). The performance test of knowledge distillation on four different network structures of VGG16, AlexNet, GoogleNet, and ResNet showed that when VGG16 was used as a teacher model, the accuracy of the model was improved to 97.54%. Using single disease recognition rate, average accuracy rate, model memory and average recognition time as 4 indicators to evaluate the accuracy of the trained Distilled-MobileNet model in a real environment, the results showed that, the average accuracy of the model reached 97.62%, and the average recognition time was shortened to 0.218 s, only accounts for 13.20% of the VGG16 model, and the model size was reduced to only 19.83 MB, which was 93.60% smaller than VGG16. Compared with traditional neural networks, distilled-mobile model has a significant improvement in reducing size and shorting recognition time, and can provide a new idea for disease recognition on devices with limited memory and computing resources.

    Agricultural Named Entity Recognition Based on Semantic Aggregation and Model Distillation | Open Access
    LI Liangde, WANG Xiujuan, KANG Mengzhen, HUA Jing, FAN Menghan
    2021, 3(1):  118-128.  doi:10.12133/j.smartag.2021.3.1.202012-SA001
    Asbtract ( 778 )   HTML ( 33)   PDF (1473KB) ( 326 )  
    Figures and Tables | References | Related Articles | Metrics

    With the development of smart agriculture, automatic question and answer (Q&A) of agricultural knowledge is needed to improve the efficiency of agricultural information acquisition. Agriculture named entity recognition plays a key role in automatic Q&A system, which helps obtaining information, understanding agriculture questions and providing answer from the knowledge graph. Due to the scarcity of labeled ANE data, some existing open agricultural entity recognition models rely on manual features, can reduce the accuracy of entity recognition. In this work, an approach of model distillation was proposed to recognize agricultural named entity data. Firstly, massive agriculture data were leveraged from Internet, an agriculture knowledge graph (AgriKG) was constructed. To overcome the scarcity of labeled named agricultural entity data, weakly named entity recognition label on agricultural texts crawled from the Internet was built with the help of AgriKG. The approach was derived from distant supervision, which was used to solve the scarcity of labeled relation extraction data. Considering the lack of labeled data, pretraining language model was introduced, which is fine tuned with existing labeled data. Secondly, large scale pretraining language model, BERT was used for agriculture named entity recognition and provided a pretty well initial parameters containing a lot of basic language knowledge. Considering that the task of agriculture named entity recognition relied heavily on low-end semantic features but slightly on high-end semantic features, an Attention-based Layer Aggregation mechanism for BERT(BERT-ALA) was designed in this research. The aim of BERT-ALA was to adaptively aggregate the output of multiple hidden layers of BERT. Based on BERT-ALA model, Bidirectional LSTM (BiLSTM) and conditional random field (CRF) were coupled to further improve the recognition precision, giving a BERT-ALA+BiLSTM+CRF model. Bi-LSTM improved BERT's insufficient learning ability of the relative position feature, while conditional random field models the dependencies of entity recognition label. Thirdly, since BERT-ALA+BiLSTM+CRF model was difficult to serve online because of the extremely high time and space complexity, BiLSTM+CRF model was used as student model to distill BERT-ALA+BiLSTM+CRF model. It fitted the BERT-ALA+BiLSTM+CRF model's output of BiLSTM layer and CRF layer. The experiment on the database constructed in the research, as well as two open datasets showed that (1) the macro-F1 of the BERT-ALA + BiLSTM + CRF model was improved by 1% compared to the baseline model BERT + BiLSTM + CRF, and (2) compared with the model trained on the original data, the macro-F1 of the distilled student model BiLSTM + CRF was increased by an average of 3.3%, the prediction time was reduced by 33%, and the storage space was reduced by 98%. The experimental results verify the effectiveness of the BERT-ALA and knowledge distillation in agricultural entity recognition.

    Intelligent Equipment and Systems
    Design and Prospect for Anti-theft and Anti-destruction of Nodes in Solar Insecticidal Lamps Internet of Things | Open Access
    HUANG Kai, SHU Lei, LI Kailiang, YANG Xing, ZHU Yan, WANG Xiaochan, SU Qin
    2021, 3(1):  129-143.  doi:10.12133/j.smartag.2021.3.1.202102-SA034
    Asbtract ( 855 )   HTML ( 60)   PDF (2413KB) ( 444 )  
    Figures and Tables | References | Related Articles | Metrics

    Solar insecticidal lamps (SILs) are widely used in agriculture for the purpose of effectively controlling pests and reducing pesticide dosage. With the increasing deployment of SILs, there are more and more reports about theft and destruction of SILs, seriously affecting the pest control effect and leading to great economic losses. Unfortunately, many efforts remain unsuccessful, since people can destruct the components of SIL in part but not steal the whole SIL, which cannot be detected by GPRS module or can only be labeled as a fault of component. To realize the broader effect of anti-theft and anti-destruction in the scenario of Solar Insecticidal Lamps Internet of Things (SIL-IoTs), there were two types of designs which would enable substantial improvements. On one hand, SIL was reformed and designed to obtain more information from different kinds of sensors and increase the difficulty of theft and destruction of SILs. Four modules were equipped including gated switch, voltage and current module, emergency power module, acceleration sensor module. Gated switch was used to judge whether the gate of power was open or closed. Voltage and current module of battery, solar panel, lamp, and metal mesh were used to judge whether the components were stolen or destructed. Emergency power module was used for communication module after the battery being stolen. Acceleration sensor module was used to judge whether the SIL was shaking by stealer. On the other hand, the auxiliary equipment of SIL, i.e., unmanned aerial vehicle insecticidal lamp (UAV-IL), was put forward for emergency applications after theft and destruction of SIL, e.g., deployment, tracking, patrol inspection, and so on. Through the above-mentioned hardware design and application of UAV-IL, more information from different kinds of sensors could be obtained to make judgements about the situation of theft and destruction. However, considering the short occurrence time of theft and destruction, the design was not enough to realize fast and accurate judgments. Therefore, six key research issues in the design of internal hardware, software algorithm and appearance structure design level were discussed, including 1) optimal design of anti-theft and anti-destruction of SILs; 2) establishment of anti-theft and anti-destruction judgment rules of SILs; 3) fast and accurate judgments of theft and destruction of SILs; 4) emergency measures after theft and destruction of SILs; and 5) prediction and prevention of theft and destruction of SILs; 6) optimal calculation to reduce the load of network data transmission. The anti-theft and anti-destruction have crucial roles in equipment safety, which can be extended to various agricultural applications.