Welcome to Smart Agriculture 中文

Archive By Volume

    Smart Agriculture 2022 Vol.4
    Please wait a minute...
    For Selected: Toggle Thumbnails
    Methods and New Research Progress of Remote Sensing Monitoring of Crop Disease and Pest Stress Using Unmanned Aerial Vehicle | Open Access
    YANG Guofeng, HE Yong, FENG Xuping, LI Xiyao, ZHANG Jinnuo, YU Zeyu
    Smart Agriculture    2022, 4 (1): 1-16.   doi:10.12133/j.smartag.SA202201008
    Accepted: 25 March 2022

    Abstract881)   HTML230)    PDF (937KB)(380)      

    Diseases and pests are main stresses to crop production. It is necessary to accurately and quickly monitor and control the stresses dynamically, so as to ensure the food security and the quality and safety of agricultural products, protect the ecological environment, and promote the sustainable development of agriculture. In recent years, with the rapid development of the unmanned aerial vehicle (UAV) industry, UAV agricultural remote sensing has played an important role in the application of crop diseases and pests monitoring due to its high image spatial resolution, strong data acquisition timeliness and low cost. The relevant background of UAV remote sensing monitoring of crop disease and pest stress was introduced, then the current methods commonly used in remote sensing monitoring of crop disease and pest stress by UAV was summarized. The data acquisition method and data processing method of UAV remote sensing monitoring of crop disease and pest stress were mainly discussed. Then, from the six aspects of visible light imaging remote sensing, multispectral imaging remote sensing, hyperspectral imaging remote sensing, thermal infrared imaging remote sensing, LiDAR imaging remote sensing and multiple remote sensing fusion and comparison, the research progress of remote sensing monitoring of crop diseases and pests by UAV worldwide was reviewed. Finally, the unresolved key technical problems and future development directions in the research and application of UAV remote sensing monitoring of crop disease and pest stress were proposed. Such as, the performance of the UAV flight platform needs to be optimized and upgraded, as well as the development of low-cost, lightweight, modular, and more adaptable airborne sensors. Convenient and automated remote sensing monitoring tasks need to be designed and implemented, and more remote sensing monitoring information can be obtained. Data processing algorithms or software should be designed and developed with greater applicability and wider applicability, and data processing time should be shortened by using 5G-based communication networks and edge computing devices. The applicability of the algorithm or model for UAV remote sensing monitoring of crop disease and pest stress needs to be stronger, so as to build a corresponding method library. We hope that this paper can help Chinese UAV remote sensing monitoring of crop diseases and pests to achieve more standardization, informatization, precision and intelligence.

    Table and Figures | Reference | Related Articles | Metrics
    Monitoring Wheat Powdery Mildew (Blumeria graminis f. sp. tritici) Using Multisource and Multitemporal Satellite Images and Support Vector Machine Classifier | Open Access
    ZHAO Jinling, DU Shizhou, HUANG Linsheng
    Smart Agriculture    2022, 4 (1): 17-28.   doi:10.12133/j.smartag.SA202202009
    Accepted: 11 March 2022
    Online available: 15 March 2022

    Abstract246)   HTML77)    PDF (7036KB)(106)      

    Since powdery mildew (Blumeria graminis f. sp. tritici) mainly infects the foliar of wheat, satellite remote sensing technology can be used to monitor and assess it on a large scale. In this study, multisource and multitemporal satellite images were used to monitor the disease and improve the classification accuracy. Specifically, four Landsat-8 thermal infrared sensor (TIRS) and twenty MODerate-resolution imaging spectroradiometer (MODIS) temperature product (MOD11A1) were used to retrieve the land surface temperature (LST), and four Chinese Gaofen-1 (GF-1) wide field of view (WFV) images was used to identify the wheat-growing areas and calculate the vegetation indices (VIs). ReliefF algorithm was first used to optimally select the vegetation index (VIs) sensitive to wheat powdery mildew, spatial-temporal fusion between Landsat-8 LST and MOD11A1 data was performed using the spatial and temporal adaptive reflectance fusion model (STARFM). The Z-score standardization method was then used to unify the VIs and LST data. Four monitoring models were then constructed through a single Landsat-8 LST, multitemporal Landsat-8 LSTs (SLST), cumulative MODIS LST (MLST) and the combination of cumulative Landsat-8 and MODIS LST (SMLST) using the Support Vector Machine (SVM) classifier, that were LST-SVM, SLST-SVM, MLST-SVM and SMLST-SVM. Four assessment indicators including user accuracy, producer accuracy, overall accuracy and Kappa coefficient were used to compare the four models. The results showed that, the proposed SMLST-SVM obtained the best identification accuracies. The overall accuracy and Kappa coefficient of the SMLST-SVM model had the highest values of 81.2% and 0.67, respectively, while they were respectively 76.8% and 0.59 for the SLST-SVM model. Consequently, multisource and multitemporal LSTs can considerably improve the differentiation accuracies of wheat powdery mildew.

    Table and Figures | Reference | Related Articles | Metrics
    Research Progress of Deep Learning in Detection and Recognition of Plant Leaf Diseases | Open Access
    SHAO Mingyue, ZHANG Jianhua, FENG Quan, CHAI Xiujuan, ZHANG Ning, ZHANG Wenrong
    Smart Agriculture    2022, 4 (1): 29-46.   doi:10.12133/j.smartag.SA202202005
    Online available: 18 February 2022

    Abstract780)   HTML189)    PDF (1061KB)(402)      

    Accurate detection and recognition of plant diseases is the key technology to early diagnosis and intelligent monitoring of plant diseases, and is the core of accurate control and information management of plant diseases and insect pests. Deep learning can overcome the disadvantages of traditional diagnosis methods and greatly improve the accuracy of diseases detection and recognition, and has attracted a lot of attention of researchers. This paper collected the main public plant diseases image data sets all over the world, and briefly introduced the basic information of each data set and their websites, which is convenient to download and use. And then, the application of deep learning in plant disease detection and recognition in recent years was systematically reviewed. Plant disease target detection is the premise of accurate classification and recognition of plant disease and evaluation of disease hazard level. It is also the key to accurately locate plant disease area and guide spray device of plant protection equipment to spray drug on target. Plant disease recognition refers to the processing, analysis and understanding of disease images to identify different kinds of disease objects, which is the main basis for the timely and effective prevention and control of plant diseases. The research progress in early disease detection and recognition algorithm was expounded based on depth of learning research, as well as the advantages and existing problems of various algorithms were described. It can be seen from this review that the detection and recognition algorithm based on deep learning is superior to the traditional detection and recognition algorithm in all aspects. Based on the investigation of research results, it was pointed out that the illumination, sheltering, complex background, different disorders with similar symptoms, different changes of disease symptoms in different periods, and overlapping coexistence of multiple diseases were the main challenges for the detection and recognition of plant diseases. At the same time, the establishment of a large-scale and more complex data set that meets the specific research needs is also a difficulty that need to face together. And at further, we point out that the combination of the better performance of the neural network, large-scale data set and agriculture theoretical basis is a major trend of the development of the future. It is also pointed out that multimodal data can be used to identify early plant diseases, which is also one of the future development direction.

    Reference | Related Articles | Metrics
    Identification of Tomato Leaf Diseases Based on Improved Lightweight Convolutional Neural Networks MobileNetV3 | Open Access
    ZHOU Qiaoli, MA Li, CAO Liying, YU Helong
    Smart Agriculture    2022, 4 (1): 47-56.   doi:10.12133/j.smartag.SA202202003
    Abstract414)   HTML13)    PDF (1623KB)(230)      

    Timely detection and treatment of tomato diseases can effectively improve the quality and yield of tomato. In order to realize the real-time and non-destructive detection of tomato diseases, a tomato leaf disease classification and recognition method based on improved MobileNetV3 was proposed in this study. Firstly, the lightweight convolutional neural network MobileNetV3 was used for transfer learning on the image net data set. The network was initialized according to the weight of the pre training model, so as to realize the transfer and fine adjustment of large-scale shared parameters of the model. The training method of transfer learning could effectively alleviate the problem of model over fitting caused by insufficient data, realized the accurate classification of tomato leaf diseases in a small number of samples, and saved the time cost of network training. Under the same experimental conditions, compared with the three standard deep convolution network models of VGG16, ResNet50 and Inception-V3, the results showed that the overall performance of MobileNetV3 was the best. Next, the impact of the change of loss function and the change of data amplification mode on the identification of tomato leaf diseases were observed by using MobileNetV3 convolution network. For the test of loss value, focal loss and cross entropy function were used for comparison, and for the test of data enhancement, conventional data amplification and mixup hybrid enhancement were used for comparison. After testing, using Mixup enhancement method under focal loss function could improve the recognition accuracy of the model, and the average test recognition accuracy of 10 types of tomato diseases under Mixup hybrid enhancement and focal loss function was 94.68%. On the basis of transfer learning, continue to improve the performance of MobileNetV3 model, the dilated convolution convolution with expansion rate of 2 and 4 was introduced into convolution layer, 1×1 full connection layer after deep convolution of 5×5 was connected to form a perceptron structure in convolution layer, and GLU gating mechanism activation function was used to train the best tomato disease recognition model. The average test recognition accuracy was as high as 98.25%, the data scale of the model was 43.57 MB, and the average detection time of a single tomato disease image was only 0.27s, after ten fold cross validation, the recognition accuracy of the model was 98.25%, and the test results were stable and reliable. The experiment showed that this study could significantly improve the detection efficiency of tomato diseases and reduce the time cost of disease image detection.

    Table and Figures | Reference | Related Articles | Metrics
    Application Scenarios and Research Progress of Remote Sensing Technology in Plant Income Insurance | Open Access
    CHEN Ailian, ZHAO Sijian, ZHU Yuxia, SUN Wei, ZHANG jing, ZHANG Qiao
    Smart Agriculture    2022, 4 (1): 57-70.   doi:10.12133/j.smartag.SA202201011
    Abstract259)   HTML34)    PDF (820KB)(202)      

    Plant income insurance has become an important part of agricultural insurance in China. It has been recommended to pilot since 2016 by Chinese government in several counties, and is now (2022) required to be implemented in all major grain producing counties in the 13 major grain producing provinces. The measurement of yield for plant income insurance in such huge volume urgently needs the support of remote sensing technology. Therefore, the development history and application status of remote sensing technology in the whole agricultural insurance industry was reviewed to help understanding the whole context circumstances of plant income insurance firstly. Then, the application scenarios of remote sensing technology were analyzed, and the key remote sensing technologies involved were introduced. The technologies involved include crop field plot extraction, crop classification, crop disaster estimation, and crop yield estimation. Research progress of these technologies were reviewed and summarized,and the satellite data sources that most commonly used in plant income insurance were summarized as well. It was found that to obtain a better support for a development of plant income insurance as well as all crop insurance from remote sensing communities, issues existed not only in the involved remote sensing technologies, but also in the remote sensing industry as well as the insurance industry. The most two important technical problems in the current application scenario of planting income insurance are that: the plot extraction and crop classification are not automated enough; the yield estimation mechanism is not strong, and the accuracy is not high. At the industry level, the first issue is the limitation of the remote sensing technology itself in that the remote sensing is not almighty, suffering from limited data source, either from satellite or from other platform, laborious data preprocessing, and pricey data fees for most of the data, and the second is the compatibility between the current business of the insurance industry and the combination of remote sensing. In this regard, this paper proposed in total five specific suggestions, which are: 1st, to establish a data distribution platform to solve the problems of difficult data acquisition and processing and standardization of initial data; 2nd, to improve the sample database to promote the automation of plot extraction and crop classification; 3rd, to achieve faster, more accurate and more scientific yields through multidisciplinary research; 4th, to standardize remote sensing technology application in agricultural insurance, and 5th, to write remote sensing applications in crop insurance contract. With these improvements, the application mode of plant income insurance and probably the whole agriculture insurance would run in a way with easily available data, more automated and intelligent technology, standards to follow, and contract endorsements.

    Table and Figures | Reference | Related Articles | Metrics
    Wheat Biomass Estimation in Different Growth Stages Based on Color and Texture Features of UAV Images | Open Access
    DAI Mian, YANG Tianle, YAO Zhaosheng, LIU Tao, SUN Chengming
    Smart Agriculture    2022, 4 (1): 71-83.   doi:10.12133/j.smartag.SA202202004
    Online available: 18 February 2022

    Abstract254)   HTML64)    PDF (1004KB)(162)      

    In order to realize the rapid and non-destructive monitoring of wheat biomass, field wheat trials were conducted based on different densities, nitrogen fertilizers and varieties, and unmanned aerial vehicle (UAV) was used to obtain RGB images in the pre-wintering stage, jointing stage, booting stage and flowering stage of wheat. The color and texture feature indices of wheat were obtained using image processing, and wheat biomass was obtained by manual field sampling in the same period. Then the relationship between different color and texture feature indices and wheat biomass was analyzed to select the suitable feature index for wheat biomass estimation. The results showed that there was a high correlation between image color index and wheat biomass in different stages, the values of r were between 0.463 and 0.911 (P<0.05). However, the correlation between image texture feature index and wheat biomass was poor, only 5 index values reached significant or extremely significant correlation level. Based on the above results, the color indices with the highest correlation to wheat biomass or the combining indices of color and texture features in different growth stages were used to construct estimation model of wheat biomass. The models were validated using independently measured biomass data, and the correlation between simulated and measured values reached the extremely significant level (P<0.01), and root mean square error (RMSE) was smaller. The R2 of color index model in the four stages were 0.538, 0.631, 0.708 and 0.464, and RMSE were 27.88, 516.99, 868.26 and 1539.81 kg/ha, respectively. The R2 of the model combined with color and texture index were 0.571, 0.658, 0.753 and 0.515, and RMSE were 25.49, 443.20, 816.25 and 1396.97 kg/ha, respectively. This indicated that the estimated results using the models were reliable and accurate. It also showed that the estimation models of wheat biomass combined with color and texture feature indices of UAV images were better than the single color index models.

    Table and Figures | Reference | Related Articles | Metrics
    Monitoring Specified Depth Soil Moisture in Field Scale with Ground Penetrating Radar | Open Access
    ZHANG Wenhan, DU Keming, SUN Yankun, LIU Buchun, SUN Zhongfu, MA Juncheng, ZHENG Feixiang
    Smart Agriculture    2022, 4 (1): 84-96.   doi:10.12133/j.smartag.SA202202010
    Accepted: 11 March 2022
    Online available: 15 March 2022

    Abstract239)   HTML26)    PDF (2258KB)(278)      

    Ground-penetrating radar (GPR) is one of the emerging technologies for soil moisture measurement. However, the measurement accuracy is difficult to determine due to some influence factors including radar wave frequency, soil texture type, etc. The GPR equipment with 1000 MHz center frequency and the measurement method of common midpoint (CMP) were adopted in the research to collect radar wave raw data in the selected field area under arid soil and moist soil conditions. The transmitter and receiver antennas of the GPR equipment were moved 0.01 m respectively in opposite directions on each radar wave raw data collection. Therefore, a CMP radar image consisted of 100 pieces of radar wave raw data by increasing the antenna distance from 0 m to 2 m. Each radar wave raw data indicated that the radar waves were reflected in the reflective layer with different dielectric constant under the same antenna distance. And the reflected and refracted radar waves were acquired by the receiving antenna at different two-way travel time respectively, and recorded in the computer. The collection of CMP soundings aimed to determine the inversion accuracy, optimum inversion depth, effective inversion depth and optimal inversion model of soil moisture content at different depth ranges and adjacent reflective layers by GPR at field scale. The reflected and refracted radar wave data were extracted from the raw data. The velocities of the surface waves and reflected waves were obtained respectively from the line slope of the surface wave data and the hyperbolic curves fitting of the reflected wave data. In addition, the relative dielectric constant of the soil at specified depth were deduced according to the soil dielectric constant and its reflected wave velocity. Moreover, 4 different models including Topp, Roth, Herkelrath and Ferre were used to figure out the soil volumetric water content inversion. Meanwhile, the measured data of soil volumetric moisture content obtained by oven drying method were used to verify the accuracy of the inversion results. The results showed that the effective inversion depth of 1000 MHz GPR ranged from 0 to 50 cm. The best inversion depth was 50 cm in arid soil and 40 cm in moist soil. The Roth model had the best correlation and stability with the highest R2 was 0.750, the Root Mean Square Error (RMSE) was 0.0114 m3/m3 and the lowest Relative Error (RE) was 3.0%. The GPR could possess the capacity of quick, precise and non-destructive measurement of specified depth soil moisture in field scale. The inversion model of soil moisture content needs to be calibrated according to different soil conditions.

    Table and Figures | Reference | Related Articles | Metrics
    Estimating the Differences of Light Capture Between Rows Based on Functional-Structural Plant Model in Simultaneous Maize-Soybean Strip Intercropping | Open Access
    LI Shuangwei, ZHU Junqi, EVERS Jochem B., VAN DER WERF Wopke, GUO Yan, LI Baoguo, MA Yuntao
    Smart Agriculture    2022, 4 (1): 97-109.   doi:10.12133/j.smartag.SA202202002
    Accepted: 15 March 2022
    Online available: 15 March 2022

    Abstract206)   HTML20)    PDF (1459KB)(121)      

    Intercropping creates a heterogeneous canopy and triggers plastic responses in plant growth and structural development. In order to quantify the effect of planting pattern, strip width and row position on the structural development and light capture of maize and soybean in simultaneous intercropping, both experimental and modelling approaches were used. Field experiments were conducted in 2017-2018 with two sole crops (maize and soybean) and two intercrops: Two rows of maize alternating with two rows of soybeans (2:2 MS) and three rows of maize alternating with six rows of soybean (3:6 MS). The morphological traits of maize and soybean e.g., leaf length and width, internode length and diameter, leaf and petiole declination angle in different rows and different planting patterns, and photosynthetically active radiation (PAR) above and below the canopy of 2:2 MS were measured throughout the growing season. A functional-structural plant model of maize-soybean intercropping was developed in the GroIMP platform. The model was parameterized based on the morphological data set of 2017, and was validated with the leaf area index (LAI), plant height and PAR data set of 2018. The model simulated the morphological development of individual organs based on growing degree days (thermal time) and calculated the light capture at leaf level. The model well reproduced the observed dynamics of leaf area index and plant height (RMSE: 0.24-0.70 m2/m2 for LAI and 0.06-0.17 m for plant height), and the fraction of light capture in the 2:2 MS intercropping (RMSE: 0.06-0.10). Maize internode diameter in intercrops increased, but the internode length did not change. Soybean internodes in intercrops became longer and thinner compared to sole soybean probably caused by the shading imposed by maize, and the 2:2 MS had longer internodes than the 3:6 MS, indicating the effects of strip width. Simulated light capture of maize in 2:2 MS intercropping was 35.6% higher than sole maize. For maize in 3:6 MS intercropping, the light capture of the border rows and inner row were 27.8% and 20.3% higher than sole maize, respectively. Compared to sole soybean, the simulated light capture of soybean in border rows was 36.0% lower in 2:2 MS intercropping, and was 28.8% lower in 3:6 MS intercropping. For 3:6 MS intercropping, light capture of soybean in inner rows I and inner rows II were 4.1% and 1.8% lower than sole soybean, respectively. In the future, the model could be further developed and used to explore and optimize the planting patterns of maize soybean intercropping under different environmental conditions using light capture as an indicator.

    Table and Figures | Reference | Related Articles | Metrics
    Construction and Application of A Novel Abscisic Acid Electrochemical Immunosensor Based on Carboxylated Graphene-Sodium Alginate Nanocomposite | Open Access
    DONG Hongtu, ZHOU Simeng, WANG Qingtao, WANG Cheng, LUO Bin, LI Aixue
    Smart Agriculture    2022, 4 (1): 110-120.   doi:10.12133/j.smartag.SA202202007
    Accepted: 21 February 2022
    Online available: 10 March 2022

    Abstract174)   HTML15)    PDF (1757KB)(73)      

    Abscisic acid (ABA) is an important plant hormone, which can control seed and bud dormancy, organ size control, senescence and death, and participate in both biological and abiotic stress, inhibit plant growth, and participate in plant disease resistance. In order to determine the content of ABA in plants quickly and accurately, a new type of ABA immunosensor was developed. To improve the detection performance of the sensor, the detection performance of the sensor was increased by modifying GR-COOH and SA on the electrode surface. The concentration of GR-COOH, SA, and ABA-Antibody were optimized, the optimal conditions for the three materials were 1.5 mg/ml, 1.25 mg/ml and 0.5 mg/ml. The immunosensor was constructed based on the electrode impedance changes (△Z )due to the binding reaction of ABA antibody and antigen. It was found that the sensor showed linear relationship with ABA in the response range of 10 pmol/L~1 μmol/L, R2 was 0.99927, and the detection limit was about 10 pmol/L. The sensor also had good selectivity and stability. Using the electrochemical immunosensor, the content of ABA in navel orange leaf that have been successfully inoculated with citrus Huanglongbing by PCR was determined, and healthy plants were used as control. The test results showed that the impedance changes(△Z ) of healthy leaves and diseased leaves were 72 and 823, respectively, which indicated that the level of ABA in the infected plants increased significantly. The sensor provides a tool for the detection of plant hormone levels under disease stress. The results showed that the content of ABA increased in the leaves of navel orange infected by citrus Huanglongbing, which indicated that ABA played an important role in plant disease resistance. Furthermore, the changes of gene expression of key enzymes CitZEP in ABA synthesis pathway were studied, The results showed that the expression of CitZEP increased in plants infected with Huanglongbing disease, and the results were consistent with the detection results of the sensor, which indicated that the sensor had good practicability.

    Reference | Related Articles | Metrics
    Quantitative Determination of Plant Hormone Abscisic Acid Using Surface Enhanced Raman Spectroscopy | Open Access
    ZHANG Yanyan, LI Can, SU Rui, LI Linze, WEI Wentao, LI Baolei, HU Jiandong
    Smart Agriculture    2022, 4 (1): 121-129.   doi:10.12133/j.smartag.SA202202001
    Accepted: 01 March 2022
    Online available: 18 February 2022

    Abstract210)   HTML16)    PDF (1359KB)(106)      

    Plant hormone Abscisic Acid (ABA) plays an important role in regulating plant growth. However, the content of ABA in plant tissues is very low, and rapid and sensitive detection methods are urgently needed. In this study, a rapid and quantitative ABA detection method was established based on aptamer recognition and surface-enhanced Raman spectroscop (SERS). The gold nanoparticles modified by ABA aptamer had the characteristics of SERS signal enhancement and selective recognition, realizing the rapid and sensitive detection of trace ABA in complex plant sample matrix. When ABA molecules appeared in detect system, the aptamer would specifically bind with ABA molecules, and the aptamer folded into G-tetrad structure at same time, which wrapped ABA molecules in the tetrad structure, shortened the distance between ABA molecules and gold nanoparticles, and the enhanced and stable ABA molecules SERS signal were obtained. Under the condition of optimized aptamer concentration at 0.12 μmol/L, different concentrations of ABA solutions in the detection system were detected. Within the concentration range of 0.1-100 μmol/L, the SERS intensity of ABA presented a good linear relationship with the concentration. The detection limit of this method was 0.1 μmol/L and the linear correlation coefficient R2 was 0.9855. The repeatability test of 20 points randomly on SERS substrate showed that the relative standard deviation (RSD) was 6.71%, indicating the stability of SERS substrate was well. Furthermore, the substrate of gold nanoparticles modified by the ABA aptamer terminal with sulfhydryl group (SH-Apt) could be stored in the refrigerator for more than half a year, indicating that the substrate has good stability. Once the preparation of the synthesized SH-Apt modified gold nanoparticles was completed. It could be used on demand without the need to prepare SERS substrate for every detection. In this sense, the constructed aptamer SERS biosensor could realize the rapid and quantitative detection of ABA. The method was used for the determination of ABA in wheat leaves, and the result was in good agreement with the Enzyme Linked Immunosorbent Assay (ELISA) (The max relative error was 9.13%). This biosensor is an exploratory study on the detection of plant hormones by SERS, and the results of the study will have important reference value for the subsequent quantitative and on-site detection of ABA, as well as the detection of other plant hormones.

    Table and Figures | Reference | Related Articles | Metrics
    Underwater Fish Species Identification Model and Real-Time Identification System | Open Access
    LI Shaobo, YANG Ling, YU Huihui, CHEN Yingyi
    Smart Agriculture    2022, 4 (1): 130-139.   doi:10.12133/j.smartag.SA202202006
    Accepted: 21 February 2022
    Online available: 10 March 2022

    Abstract482)   HTML54)    PDF (1329KB)(253)      

    Convolutional neural network models have different advantages and disadvantages, it is becoming more and more difficult to select an appropriate convolutional neural network model in an actual fish identification project. The identification of underwater fish is a challenge task due to varies in illumination, low contrast, high noise, low resolution and sample imbalance between each type of image from the real underwater environment. In addition, deploying models to mobile devices directly will reduce the accuracy of the model sharply. In order to solve the above problems, Fish Recognition Ground-Truth dataset was used to training model in this study, which is provided by Fish4Knowledge project from University of Edinburgh. It contains 27,370 images with 23 fish species, and has been labeled manually by marine biologists. AlexNet, GoogLeNet, ResNet and DenseNet models were selected initially according to the characteristics of real-time underwater fish identification task, then a comparative experiment was designed to explore the best network model. Random image flipping, rotation and color dithering were used to enhance data based on ground-truth fish dataset in response to the limited number of underwater fish images. Considering that there was a serious imbalance in the number of samples in each category, the label smoothing technology was used to alleviate model overfitting. The Ranger optimizer and Cosine learning rate attenuation strategy were used to further improve the training effect of the models. The accuracy and recall rate information of each model were recorded and counted. The results showed that, the accuracy and recall rate of the fish recognition model based on DenseNet reached 99.21% and 96.77% in train set and validation set respectively, its F1 value reached 0.9742, which was the best model obtained in the experiment. Finally, a remote fish identification system was designed based on Python language, in this system the model was deployed to linux server and the Android APP was responsible for uploading fish images via http to request server to identify the fishes and displaying the identification information returned by server, such as fish species, profiles, habits, distribution, etc. A set of recognition tests were performed on real Android phone and the results showed that in the same local area net the APP could show fish information rapidly and exactly within 1 s.

    Table and Figures | Reference | Related Articles | Metrics
    Scale Adaptive Small Objects Detection Method in Complex Agricultural Environment: Taking Bees as Research Object | Open Access
    GUO Xiuming, ZHU Yeping, LI Shijuan, ZHANG Jie, LYU Chunyang, LIU Shengping
    Smart Agriculture    2022, 4 (1): 140-149.   doi:10.12133/j.smartag.SA202203003
    Online available: 15 March 2022

    Abstract249)   HTML28)    PDF (1997KB)(178)      

    Objects in farmlands often have characteristic of small volume and high density with variable light and complex background, and the available object detection models could not get satisfactory recognition results. Taking bees as research objects, a method that could overcome the influence from the complex backgrounds, the difficulty in small object feature extraction was proposed, and a detection algorithm was created for small objects irrelevant to image size. Firstly, the original image was split into some smaller sub-images to increase the object scale, and the marked objects were assigned to the sub-images to produce a new dataset. Then, the model was trained again using transfer learning to get a new object detection model. A certain overlap rate was set between two adjacent sub-images in order to restore the objects. The objects from each sub-image was collected and then non-maximum suppression (NMS) was performed to delete the redundant detection boxes caused by the network, an improved NMS named intersection over small NMS (IOS-NMS) was then proposed to delete the redundant boxes caused by the overlap between adjacent sub-images. Validation tests were performed when sub-image size was set was 300×300, 500×500 and 700×700, the overlap rate was set as 0.2 and 0.05 respectively, and the results showed that when using single shot multibox detector (SSD) as the object detection model, the recall rate and precision was generally higher than that of SSD with the maximum difference 3.8% and 2.6%, respectively. In order to further verify the algorithm in small target recognition with complex background, three bee images with different scales and different scenarios were obtained from internet and test experiments were conducted using the new proposed algorithm and SSD. The results showed that the proposed algorithm could improve the performance of target detection and had strong scale adaptability and generalization. Besides, the new algorithm required multiple forward reasoning for a single image, so it was not time-efficient and was not suitable for edge calculation.

    Table and Figures | Reference | Related Articles | Metrics
    Advances and Challenges in Physiological Parameters Monitoring and Diseases Diagnosing of Dairy Cows Based on Computer Vision
    KANG Xi, LIU Gang, CHU Mengyuan, LI Qian, WANG Yanchao
    Smart Agriculture    2022, 4 (2): 1-18.   doi:10.12133/j.smartag.SA202204005
    Online available: 13 July 2022

    Abstract122)   HTML27)    PDF (1082KB)(61)      

    Realizing the construction of intelligent farming by using advanced information technology, thus improving the living welfare of dairy cows and the economic benefits of dairy farms has become an important goal and task in dairy farming research field. Computer vision technology has the advantages of non-contact, stress-free, low cost and high throughput, and has a broad application prospect in animal production. On the basis of describing the importance of computer vision technology in the development of intelligent farming industry, this paper introduced the cutting-edge technology of cow physiological parameters and disease diagnosis based on computer vision, including cow temperature monitoring, body size monitoring, weight measurement, mastitis detection and lameness detection. The introduction coverd the development process of these studies, the current mainstream techniques, and discussed the problems and challenges in the research and application of related technology, aiming at the problem that the current computer vision-based detection methods are susceptible to individual difference and environmental changes. Combined with the development status of farming industry, suggestions on how to improve the universality of computer vision technology in intelligent farming industry, how to improve the accuracy of monitoring cows' physiological parameters and disease diagnosis, and how to reduce the influence of environment on the system were put forward. Future research work should focus on research and developmentof algorithm, make full use of computer vision technology continuous detection and the advantage of large amount of data, to ensure the accuracy of the detection, and improve the function of the system integration and data utilization, expand the computer vision system function. Under the premise that does not affect the ability of the system, to improve the study on the number of function integration and system function and reduce equipment costs.

    Table and Figures | Reference | Related Articles | Metrics
    Pig Sound Analysis: A Measure of Welfare | Open Access
    JI Nan, YIN Yanling, SHEN Weizheng, KOU Shengli, DAI Baisheng, WANG Guowei
    Smart Agriculture    2022, 4 (2): 19-35.   doi:10.12133/j.smartag.SA202204004
    Accepted: 15 July 2022
    Online available: 01 August 2022

    Abstract74)   HTML18)    PDF (689KB)(45)      

    Pig welfare is closely related to the economical production of pig farms. With regard to pig welfare assessment, pig sounds are significant indicators, which can reflect the quality of the barn environment, the physical condition and the health of pigs. Therefore, pig sound analysis is of high priority and necessary. In this review, the relationship between pig sound and welfare was analyzed. Three kinds of pig sounds are closely related to pig welfare, including coughs, screams, and grunts. Subsequently, both wearable and non-contact sensors were briefly described in two aspects of advantages and disadvantages. Based on the advantages and feasibility of microphone sensors in contactless way, the existing techniques for processing pig sounds were elaborated and evaluated for further in-depth research from three aspects: sound recording and labeling, feature extraction, and sound classification. Finally, the challenges and opportunities of pig sound research were discussed for the ultimate purpose of precision livestock farming (PLF) in four ways: concerning sound monitoring technologies, individual pig welfare monitoring, commercial applications and pig farmers. In summary, it was found that most of the current researches on pig sound recognition tasks focused on the selection of classifiers and algorithm improvement, while fewer research was conducted on sound labeling and feature extraction. Meanwhile, pig sound recognition faces some challenging problems, involving the difficulty in obtaining the audio data from different pig growth stages and verifying the developed algorithms in a variety of pig farms. Overall, it is suggested that technologies involved in the automatic identification process should be explored in depth. In the future, strengthen cooperation among cross-disciplinary experts to promote the development and application of PLF is also nessary.

    Table and Figures | Reference | Related Articles | Metrics
    Research Progress and Technology Trend of Intelligent Morning of Dairy Cow Motion Behavior | Open Access
    WANG Zheng, SONG Huaibo, WANG Yunfei, HUA Zhixin, LI Rong, XU Xingshi
    Smart Agriculture    2022, 4 (2): 36-52.   doi:10.12133/j.smartag.SA202203011
    Online available: 02 August 2022

    Abstract76)   HTML14)    PDF (1141KB)(62)      

    The motion behavior of dairy cows contains much of health information. The application of information and intelligent technology will help farms grasp the health status of dairy cows in time and improve breeding efficiency. In this paper, the development trend of intelligent morning technology of cow's motion behavior was mainly analyzed. Firstly, on the basis of expounding the significance of monitoring the basic motion (lying, walking, standing), oestrus, breathing, rumination and limping of dairy cows, the necessity of behavior monitoring of dairy cows was introduced. Secondly, the current research status was summarized from contact monitoring methods and non-contact monitoring methods in chronological order. The principle and achievements of related research were introduced in detail and classified. It is found that the current contact monitoring methods mainly rely on acceleration sensors, pedometers and pressure sensors, while the non-contact monitoring methods mainly rely on video images, including traditional video image analysis and video image analysis based on deep learning. Then, the development status of cow behavior monitoring industry was analyzed, and the main businesses and mainstream products of representative livestock farm automation equipment suppliers were listed. Industry giants, such as Afimilk and DeLaval, as well as their products such as intelligent collar (AfiCollar), pedometer (AfiActll Tag) and automatic milking equipment (VMS? V300) were introduced. After that, the problems and challenges of current contact and non-contact monitoring methods of dairy cow motion behavior were put forward. The current intelligent monitoring methods of dairy cows' motion behavior are mainly wearable devices, but they have some disadvantages, such as bring stress to dairy cows and are difficult to install and maintain. Although the non-contact monitoring methods based on video image analysis technology does not bring stress to dairy cows and is low cost, the relevant research is still in its infancy, and there is still a certain distance from commercial use. Finally, the future development directions of relevant key technologies were prospected, including miniaturization and integration of wearable monitoring equipment, improving the robustness of computer vision technology, multi-target monitoring with limited equipment and promoting technology industrialization.

    Table and Figures | Reference | Related Articles | Metrics
    Gait Phase Recognition of Dairy Cows based on Gaussian Mixture Model and Hidden Markov Model | Open Access
    ZHANG Kai, HAN Shuqing, CHENG Guodong, WU Saisai, LIU Jifang
    Smart Agriculture    2022, 4 (2): 53-63.   doi:10.12133/j.smartag.SA202204003
    Accepted: 14 July 2022
    Online available: 01 August 2022

    Abstract60)   HTML5)    PDF (1418KB)(24)      

    The gait phase of dairy cows is an important indicator to reflect the severity of lameness. IThe accuracy of available gait segmentation methods was not enough for lameness detection. In this study, a gait phase recognition method based on Gaussian mixture model (GMM) and hidden Markov model (HMM) was proposed and tested. Firstly, wearable inertial sensors LPMS-B2 were used to collect the acceleration and angular velocity signals of cow hind limbs. In order to remove the noise of the system and restore the real dynamic data, Kalman filter was used for data preprocessing. The first-order difference of the angular velocity of the coronal axis was selected as the eigenvalue. Secondly, to analyze the long-term continuous recorded gait sequences of dairy cows, the processed data was clustered by GMM in the unsupervised way. The clustering results were taken as the input of the HMM, and the gait phase recognition of dairy cows was realized by decoding the observed data. Finally, the cow gait was segmented into 3 phases, including the stationary phase, standing phase and swing phase. At the same time, gait segmentation was achieved according to the standing phase and swing phase. The accuracy, recall rate and F1 of the stationary phase were 89.28%, 90.95% and 90.91%, respectively. The accuracy, recall rate and F1 of the standing phase recognition in continuous gait were 91.55%, 86.71% and 89.06%, respectively. The accuracy, recall rate and F1 of the swing phase recognition in continuous gait were 86.67%, 91.51% and 89.03%, respectively. The accuracy of cow gait segmentation was 91.67%, which was 4.23% and 1.1 % higher than that of the event-based peak detection method and dynamic time warping algorithm, respectively. The experimental results showed that the proposed method could overcome the influence of the cow's walking speed on gait phase recognition results, and recognize the gait phase accurately. This experiment provides a new method for the adaptive recognition of the cow gait phase in unconstrained environments. The degree of lameness of dairy cows can be judged by the gait features.

    Table and Figures | Reference | Related Articles | Metrics
    Automatic Acquisition and Target Extraction of Beef Cattle 3D Point Cloud from Complex Environment | Open Access
    LI Jiawei, MA Weihong, LI Qifeng, XUE Xianglong, WANG Zhiquan
    Smart Agriculture    2022, 4 (2): 64-76.   doi:10.12133/j.smartag.SA202206003
    Online available: 13 July 2022

    Abstract82)   HTML14)    PDF (2798KB)(24)      

    Non-contact measurement based on the point cloud acquisition technology is able to alleviate the stress responses among beef cattle while collecting core body dimension data, but the current 3D data collection for beef cattle is usually time-consuming and easily influenced by the environment, which is in fact inapplicable to the actual breeding environment. In order to overcome the difficulty in obtaining the complete beef cattle point clouds, a non-contact phenotype data acquisition equipment was developed with a 3D reconstruction function, which can provide a large amount of standardized 3D quantitative phenotype data for beef cattle breeding and fattening process. The system is made up of a Kinect DK depth camera, an infrared grating trigger, and an Radio Frequency Identification (RFID) trigger, which enables the multi-angle instantaneous acquisition of beef cattle point clouds when the beef cattle pass through the walkway. The point cloud processing algorithm was developed based on the C++ platform and Point Cloud Library (PCL), and 3D reconstruction of beef cattle point clouds was achieved through spatial and outlier point filtering, Random Sample Consensus (RANSAC) shape fitting, point cloud thinning, and perceptual box filtering based on the dimensionality reduction density clustering to effectively filter out the interference, such as noises from the railings close to the beef cattle, without destroying the integrity of the point clouds. In the present work, a total of 124 sets of point clouds were successfully collected from 20 beef cattles on the actual farm using this system, and the target extraction experiments were completed. Notably, the beef cattle passed through the walkway in a natural state without any intervention during the whole data collection process. The experimental results showed that the acquisition success rate of this device was 91.89%. The coordinate system of the collected point cloud was consistent with the real situation and the body dimension reconstruction error was 0.6%. This device can realize the automatic acquisition and 3D reconstruction of beef cattle point cloud data from multiple angles without human intervention, and can automatically extract the target beef cattle point clouds from a complex environment. The point cloud data collected by this system help to restore the body size and shape of beef cattle, thereby provide solid support for the measurement of core parameters such as body height, body width, body oblique length, chest circumference, abdominal circumference, and body weight.

    Table and Figures | Reference | Related Articles | Metrics
    Multiscale Feature Fusion Yak Face Recognition Algorithm Based on Transfer Learning | Open Access
    CHEN Zhanqi, ZHANG Yu'an, WANG Wenzhi, LI Dan, HE Jie, SONG Rende
    Smart Agriculture    2022, 4 (2): 77-85.   doi:10.12133/j.smartag.SA202201001
    Online available: 02 August 2022

    Abstract89)   HTML13)    PDF (1831KB)(46)      

    Identifying of yak is indispensable for individual documentation, behavior monitoring, precise feeding, disease prevention and control, food traceability, and individualized breeding. Aiming at the application requirements of animal individual identification technology in intelligent informatization animal breeding platforms, a yak face recognition algorithm based on transfer learning and multiscale feature fusion, i.e., transfer learning-multiscale feature fusion-VGG(T-M-VGG) was proposed. The sample data set of yak facial images was produced by a camera named GoPro HERO8 BLACK. Then, a part of dataset was increased by the data enhancement ways that involved rotating, adjusting the brightness and adding noise to improve the robustness and accuracy of model. T-M-VGG, a kind of convolutional neural network based on pre-trained visual geometry group network and transfer learning was input with normalized dataset samples. The feature map of Block3, Block4 and Block5 were considered as F3, F4 and F5, respectively. What's more, F3 and F5 were taken by the structure that composed of three parallel dilated convolutions, the dilation rate were one, two and three, respectively, to dilate the receptive filed which was the map size of feature map. Further, the multiscale feature maps were fused by the improved feature pyramid which was in the shape of stacked hourglass structure. Finally, the fully connected layer was replaced by the global average pooling to classify and reduce a large number of parameters. To verify the effectiveness of the proposed model, a comparative experiment was conducted. The experimental results showed that recognition accuracy rate in 38,800 data sets of 194 yaks reached 96.01%, but the storage size was 70.75 MB. Twelve images representing different yak categories from dataset were chosen randomly for occlusion test. The origin images were masked with different shape of occlusions. The accuracy of identifying yak individuals was 83.33% in the occlusion test, which showed that the model had mainly learned facial features. The proposed algorithm could provide a reference for research of yak face recognition and would be the foundation for the establishment of smart management platform.

    Table and Figures | Reference | Related Articles | Metrics
    Research Progress and Outlook of Livestock Feeding Robot
    YANG Liang, XIONG Benhai, WANG Hui, CHEN Ruipeng, ZHAO Yiguang
    Smart Agriculture    2022, 4 (2): 86-98.   doi:10.12133/j.smartag.SA202204001
    Accepted: 17 June 2022
    Online available: 01 August 2022

    Abstract85)   HTML15)    PDF (1901KB)(46)      

    The production mode of livestock breeding has changed from extensive to intensive, and the production level is improved. However, low labor productivity and labor shortage have seriously restricted the rapid development of China's livestock breeding industry. As a new intelligent agricultural machinery equipment, agricultural robot integrates advanced technologies, such as intelligent monitoring, automatic control, image recognition technology, environmental modeling algorithm, sensors, flexible execution, etc. Using modern information and artificial intelligence technology, developing livestock feeding and pushing robots, realizing digital and intelligent livestock breeding, improving livestock breeding productivity are the main ways to solve the above contradiction. In order to deeply analyze the research status of robot technology in livestock breeding, products and literature were collected worldwide. This paper mainly introduced the research progress of livestock feeding robot from three aspects: Rail feeding robot, self-propelled feeding robot and pushing robot, and analyzed the technical characteristics and practical application of feeding robot.The rail feeding robot runs automatically along the fixed track, identifies the target animal, positions itself, and accurately completes feed delivery through preset programs to achieve accurate feeding of livestock. The self-propelled feeding robot can walk freely in the farm and has automatic navigation and positioning functions. The system takes single chip microcomputer as the control core and realizes automatic walking by sensor and communication module. Compared with the rail feeding robot, the feeding process is more flexible, convenient and technical, which is more conducive to the promotion and application of livestock farms. The pushing robot will automatically push the feed to the feeding area, promote the increase of feed intake of livestock, and effectively reduce the labor demand of the farm. By comparing the feeding robots of developed countries and China from two aspects of technology and application, it is found that China has achieved some innovation in technology, while developted countries do better in product application. The development of livestock robot was prospected. In terms of strategic planning, it is necessary to keep up with the international situation and the trend of technological development, and formulate the agricultural robot development strategic planning in line with China's national conditions. In terms of the development of core technologies, more attention should be paid to the integration of information perception, intelligent sensors and deep learning algorithms to realize human-computer interaction. In terms of future development trends, it is urgent to strengthen innovation, improve the friendliness and intelligence of the robot, and improve the learning ability of the robot. To sum up, intelligent livestock feeding and pushing machine operation has become a cutting-edge technology in the field of intelligent agriculture, which will surely lead to a new round of agricultural production technology reform, promote the transformation and upgrading of China's livestock industry. .

    Table and Figures | Reference | Related Articles | Metrics
    Design and Key Technologies of Big Data Platform for Commercial Beef Cattle Breeding | Open Access
    MA Weihong, LI Jiawei, WANG Zhiquan, GAO Ronghua, DING Luyu, YU Qinyang, YU Ligen, LAI Chengrong, LI Qifeng
    Smart Agriculture    2022, 4 (2): 99-109.   doi:10.12133/j.smartag.SA202203005
    Accepted: 17 May 2022
    Online available: 20 May 2022

    Abstract147)   HTML21)    PDF (1983KB)(37)      

    Focusing on the low level of management and informatization and intelligence of the beef cattle industry in China, a big data platform for commercial beef cattle breeding, referring to the experience of international advanced beef cattle breeding countries, was proposed in this research. The functions of the platform includes integrating germplasm resources of beef cattle, automatic collecting of key beef cattle breeding traits, full-service support for the beef cattle breeding process, formation of big data analysis and decision-making system for beef cattle germplasm resources, and joint breeding innovation model. Aiming at the backward storage and sharing methods of beef cattle breeding data and incomplete information records in China, an information resource integration platform and an information database for beef cattle germplasm were established. Due to the vagueness and subjectivity of the breeding performance evaluation standard, a scientific online evaluation technology of beef cattle breeding traits and a non-contact automatic acquisition and intelligent calculation method were proposed. Considering the lack of scientific and systematic breeding planning and guidance for farmers in China, a full-service support system for beef cattle breeding and nanny-style breeding guidance during beef cattle breeding was developed. And an interactive progressive decision-making method for beef cattle breeding to solve the lack of data accumulation of beef cattle germplasm was proposed. The main body of breeding and farming enterprises was not closely integrated, according to that, the innovative breeding model of regional integration was explored. The idea of commercial beef cattle breeding big data software platform and the technological and model innovation content were also introduced. The technology innovations included the deep mining of germplasm resources data and improved breed management pedigree, the automatic acquisition and evaluation technology of non-contact breeding traits, the fusion of multi-source heterogeneous information to provide intelligent decision support. The future goal is to form a sustainable information solution for China's beef cattle breeding industry and improve the overall level of China's beef cattle breeding industry.

    Table and Figures | Reference | Related Articles | Metrics
    Development of China Feed Nutrition Big Data Analysis Platform | Open Access
    XIONG Benhai, ZHAO Yiguang, LUO Qingyao, ZHENG Shanshan, GAO Huajie
    Smart Agriculture    2022, 4 (2): 110-120.   doi:10.12133/j.smartag.SA202205003
    Online available: 13 July 2022

    Abstract74)   HTML10)    PDF (1581KB)(44)      

    The shortage of feed grain is continually worsening in China, which leads to the transformation of feed grain security into national food security. Therefore, comprehensively integrating the basic data resources of feed nutrition and improving the nutritional value of all available feed resources will be one of the key technical strategies to ensure national food security in China. In this study, based on the description specification and attribute data standard of 16 categories of Chinese feed raw materials, more than 500,000 pieces of data on the types, spatial distribution, chemical composition and nutritional value characteristics of existing feed resources, which were accumulated through previous projects from the sixth Five-Year Plan to the thirteenth Five-Year Plan period, were digitally collected, recorded, categorized and comprehensively analyzed. By using MySQL relational database technology and PHP program, a new generation of feed nutrition big data online platform (http://www.chinafeeddata.org.cn/) was developed and web data sharing service was provided as well. First of all, the online platform provided visual analysis of all warehousing data, which could realize the visual comparison of a single or multiple feed nutrients in various graphic forms such as scatter diagram, histogram, curve line and column chart. By using two-dimensional code technology, all feed nutrition attribute data and feed entity sample traceability data could be shared and downloaded remotely in real-time on mobile phones. Secondly, the online platform also incorporated various regression models for prediction of feed effective nutrient values using readily available feed chemical composition in the datasets, providing dynamic analysis for feed raw material nutrient variation. Finally, based on Geographic Information System technology, the online platform integrated the data of feed chemical composition and major mineral element concentrations with their geographical location information, which was able to provide the distribution query and comparative analysis of the geographic information map of the feed raw material nutrition data at both provincial and national level. Meanwhile, the online platform can also provide a download service of the various datasets, which brought convenience to the comprehensive application of existing feed nutrition data. This research also showed that expanding feed resource data and providing prediction and analysis models of feed effective nutrients could maximize the utilization of the existing feed nutrition data. After embedding online calculation modules of various feed formulation software, this platform would be able to provide a one-stop service and optimize the utilization of the feed nutrition data.

    Table and Figures | Reference | Related Articles | Metrics
    Typical Raman Spectroscopy Ttechnology and Research Progress in Agriculture Detection | Open Access
    GAO Zhen, ZHAO Chunjiang, YANG Guiyan, DONG Daming
    Smart Agriculture    2022, 4 (2): 121-134.   doi:10.12133/j.smartag.SA202201013
    Abstract75)   HTML8)    PDF (807KB)(38)      

    Raman spectroscopy is a type of scattering spectroscopy with features such as rapid, less susceptible to moisture interference, no sample pre-treatment and in vivo detection. As a powerful characterization tool for analyzing and testing the molecular composition and structure of substances, Raman spectroscopy is also playing an extremely important role in the detection of plant and animal phenotypes, food safety, soil and water quality in the agricultural field with the continuous improvement of Raman spectroscopy technology. In this paper, the detection principles of Raman spectroscopy are introduced, and the new progresses of eight Raman spectroscopy technology are summarized, including confocal microscopy Raman spectroscopy, Fourier transform Raman spectroscopy, surface-enhanced Raman spectroscopy, tip-enhanced Raman spectroscopy, resonance Raman spectroscopy, spatially shifted Raman spectroscopy, frequency-shifted excitation Raman difference spectroscopy and Raman spectroscopy based on nonlinear optics, etc. And their advantages and disadvantages and application scenarios are prerented, respectively. The applications of Raman spectroscopy in plant detection, soil detection, water quality detection, food detection, etc. are summarized. It can be specifically subdivided into plant phenotype, plant stress, soil pesticide residue detection, soil colony detection, soil nutrient detection, food pesticide detection, food quality detection, food adulteration detection, and water quality detection. In future agricultural applications, the elimination of fluorescence background due to complex living organisms in Raman spectroscopy is the next research direction. The study of stable enhanced substrates is an important direction in the application of Surface Enhanced Raman Spectroscopy (SERS). In order to meet the measurement of different scenarios, portable and telemetric Raman spectrometers will also play an important role in the future. Raman spectroscopy needs to be further explored for a wide variety of research objects in agriculture, especially for applications in animal science, for which there is still a paucity of relevant studies up to now. In the existing field of agricultural research, it is necessary to pursue the characterization of more specific substances by Raman spectroscopy, which can prompt the application of Raman spectroscopy for a wider range of uses in agriculture. Further, the pursuit of lower detection limits and higher stability for practical applications is also the direction of development of Raman spectroscopy in the field of agriculture. Finally, the challenges that need to be solved and the future development directions of Raman spectroscopy are proposed in the field of agriculture in order to bring more inspiration to future agricultural production and research.

    Table and Figures | Reference | Related Articles | Metrics
    Research Progress and Enlightenment of Japanese Harvesting Robot in Facility Agriculture | Open Access
    HUANG Zichen, SUGIYAMA Saki
    Smart Agriculture    2022, 4 (2): 135-149.   doi:10.12133/j.smartag.SA202202008
    Online available: 02 August 2022

    Abstract125)   HTML30)    PDF (1768KB)(75)      

    Intelligent equipment is necessary to ensure stable, high-quality, and efficient production of facility agriculture. Among them, intelligent harvesting equipment needs to be designed and developed according to the characteristics of fruits and vegetables, so there is little large-scale mechanization. The intelligent harvesting equipment in Japan has nearly 40 years of research and development history since the 1980s, and the review of its research and development products has specific inspiration and reference significance. First, the preferential policies that can be used for harvesting robots in the support policies of the government and banks to promote the development of facility agriculture were introduced. Then, the development of agricultural robots in Japan was reviewed. The top ten fruits and vegetables in the greenhouse were selected, and the harvesting research of tomato, eggplant, green pepper, cucumber, melon, asparagus, and strawberry harvesting robots based on the combination of agricultural machinery and agronomy was analyzed. Next, the commercialized solutions for tomato, green pepper, and strawberry harvesting system were detailed and reviewed. Among them, taking the green pepper harvesting robot developed by the start-up company AGRIST Ltd. in recent years as an example, the harvesting robot developed by the company based on the Internet of Things technology and artificial intelligence algorithms was explained. This harvesting robot can work 24 h a day and can control the robot's operation through the network. Then, the typical strawberry harvesting robot that had undergone four generations of prototype development were reviewed. The fourth-generation system was a systematic solution developed by the company and researchers. It consisted of high-density movable seedbeds and a harvesting robot with the advantages of high space utilization, all-day work, and intelligent quality grading. The strengths, weaknesses, challenges, and future trends of prototype and industrialized solutions developed by universities were also summarized. Finally, suggestions for accelerating the development of intelligent, smart, and industrialized harvesting robots in China's facility agriculture were provided.

    Table and Figures | Reference | Related Articles | Metrics
    Phenotypic Traits Extraction of Wheat Plants Using 3D Digitization
    ZHENG Chenxi, WEN Weiliang, LU Xianju, GUO Xinyu, ZHAO Chunjiang
    Smart Agriculture    2022, 4 (2): 150-162.   doi:10.12133/j.smartag.SA202203009
    Online available: 10 June 2022

    Abstract97)   HTML19)    PDF (1791KB)(43)      

    Aiming at the difficulty of accurately extract the phenotypic traits of plants and organs from images or point clouds caused by the multiple tillers and serious cross-occlusion among organs of wheat plants, to meet the needs of accurate phenotypic analysis of wheat plants, three-dimensional (3D) digitization was used to extract phenotypic parameters of wheat plants. Firstly, digital representation method of wheat organs was given and a 3D digital data acquisition standard suitable for the whole growth period of wheat was formulated. According to this standard, data acquisition was carried out using a 3D digitizer. Based on the definition of phenotypic parameters and semantic coordinates information contained in the 3D digitizing data, eleven conventional measurable phenotypic parameters in three categories were quantitative extracted, including lengths, thicknesses, and angles of wheat plants and organs. Furthermore, two types of new parameters for shoot architecture and 3D leaf shape were defined. Plant girth was defined to quantitatively describe the looseness or compactness by fitting 3D discrete coordinates based on the least square method. For leaf shape, wheat leaf curling and twisting were defined and quantified according to the direction change of leaf surface normal vector. Three wheat cultivars including FK13, XN979, and JM44 at three stages (rising stage, jointing stage, and heading stage) were used for method validation. The Open3D library was used to process and visualize wheat plant data. Visualization results showed that the acquired 3D digitization data of maize plants were realistic, and the data acquisition approach was capable to present morphological differences among different cultivars and growth stages. Validation results showed that the errors of stem length, leaf length, stem thickness, stem and leaf angle were relatively small. The R2 were 0.93, 0.98, 0.93, and 0.85, respectively. The error of the leaf width and leaf inclination angle were also satisfactory, the R2 were 0.75 and 0.73. Because wheat leaves are narrow and easy to curl, and some of the leaves have a large degree of bending, the error of leaf width and leaf angle were relatively larger than other parameters. The data acquisition procedure was rather time-consuming, while the data processing was quite efficient. It took around 133 ms to extract all mentioned parameters for a wheat plant containing 7 tillers and total 27 leaves. The proposed method could achieve convenient and accurate extraction of wheat phenotypes at individual plant and organ levels, and provide technical support for wheat shoot architecture related research.

    Table and Figures | Reference | Related Articles | Metrics
    Identification and Counting of Silkworms in Factory Farm Using Improved Mask R-CNN Model
    HE Ruimin, ZHENG Kefeng, WEI Qinyang, ZHANG Xiaobin, ZHANG Jun, ZHU Yihang, ZHAO Yiying, GU Qing
    Smart Agriculture    2022, 4 (2): 163-173.   doi:10.12133/j.smartag.SA202201012
    Accepted: 20 May 2022
    Online available: 01 August 2022

    Abstract49)   HTML9)    PDF (2347KB)(23)      

    Factory-like rearing of silkworm (Bombyx mori) using artificial diet for all instars is a brand-new rearing mode of silkworm. Accurate feeding is one of the core technologies to save cost and increase efficiency in factory silkworm rearing. Automatic identification and counting of silkworm play a key role to realize accurate feeding. In this study, a machine vision system was used to obtain digital images of silkworms during main instars, and an improved Mask R-CNN model was proposed to detect the silkworms and residual artificial diet. The original Mask R-CNN was improved using the noise data of annotations by adding a pixel reweighting strategy and a bounding box fine-tuning strategy to the model frame. A more robust model was trained to improve the detection and segmentation abilities of silkworm and residual feed. Three different data augmentation methods were used to expand the training dataset. The influences of silkworm instars, data augmentation, and the overlap between silkworms on the model performance were evaluated. Then the improved Mask R-CNN was used to detect silkworms and residual feed. The AP50 (Average Precision at IoU=0.5) of the model for silkworm detection and segmentation were 0.790 and 0.795, respectively, and the detection accuracy was 96.83%. The detection and segmentation AP50 of residual feed were 0.641 and 0.653, respectively, and the detection accuracy was 87.71%. The model was deployed on the NVIDIA Jetson AGX Xavier development board with an average detection time of 1.32 s and a maximum detection time of 2.05 s for a image. The computational speed of the improved Mask R-CNN can meet the requirement of real-time detection of the moving unit of the silkworm box on the production line. The model trained by the fifth instar data showed a better performance on test data than the fourth instar model. The brightness enhancement method had the greatest contribution to the model performance as compared to the other data augmentation methods. The overlap between silkworms also negatively affected the performance of the model. This study can provide a core algorithm for the research and development of the accurate feeding information system and feeding device for factory silkworm rearing, which can improve the utilization rate of artificial diet and improve the production and management level of factory silkworm rearing.

    Table and Figures | Reference | Related Articles | Metrics
    Supply and Demand Forecasting Model of Multi-Agricultural Products Based on Deep Learning
    ZHUANG Jiayu, XU Shiwei, LI Yang, XIONG Lu, LIU Kebao, ZHONG Zhiping
    Smart Agriculture    2022, 4 (2): 174-182.   doi:10.12133/j.smartag.SA202203013
    Online available: 13 July 2022

    Abstract93)   HTML15)    PDF (1055KB)(38)      

    To further improve the simulation and estimation accuracy of the supply and demand process of agricultural products, a large number of agricultural data at the national and provincial levels since 1980 were used as the basic research sample, including production, planted area, food consumption, industrial consumption, feed consumption, seed consumption, import, export, price, GDP, population, urban population, rural population, weather and so on, by fully considering the impact factors of agricultural products such as varieties, time, income and economic development, a multi-agricultural products supply and demand forecasting model based on long short-term memory neural network (LSTM) was constructed in this study. The general thought of supply and demand forecasting model is packaging deep neural network training model as an I/O-opening modular model, reserving control interface for input of outside data, and realizing the indicators forecasting of supply and demand and matrixing of balance sheet. The input of model included forecasting balance sheet data of agricultural products, annual price data, general economic data, and international currency data since 2000. The output of model was balance sheet data of next decade since forecasting time. Under the premise of fully considering the mechanical constraints, the model used the advantages of deep learning algorithms in nonlinear model analysis and prediction to analyze and predict supply and demand of 9 main types of agricultural products, including rice, wheat, corn, soybean, pork, poultry, beef, mutton, and aquatic products. The production forecast results of 2019-2021 based on this model were compared and verified with the data published by the National Bureau of Statistics, and the mean absolute percentage error was 3.02%, which meant the average forecast accuracy rate of 2019-2021 was 96.98%. The average forecast accuracy rate was 96.10% in 2019, 98.26% in 2020, and 96.58% in 2021, which shows that with the increase of sample size, the prediction effect of intelligent learning model would gradually get better. The forecasting results indicate that the multi-agricultural supply and demand prediction model based on LSTM constructed in this study can effectively reflect the impact of changes in hidden indicators on the prediction results, avoiding the uncontrollable error introduced by manual experience intervention. The model can provide data production and technical support such as market warning, policy evaluation, resource management and public opinion analysis for agricultural production and management and macroeconomic regulation, and can provide intelligent technical support for multi-regional and inter-temporal agricultural outlook work by monitoring agricultural operation data in a timely manner.

    Table and Figures | Reference | Related Articles | Metrics
    Comparative Study of the Regulation Effects of Artificial Intelligence-Assisted Planting Strategies on Strawberry Production in Greenhouse
    GENG Wenxuan, ZHAO Junye, RUAN Jiwei, HOU Yuehui
    Smart Agriculture    2022, 4 (2): 183-193.   doi:10.12133/j.smartag.SA202203006
    Online available: 10 June 2022

    Abstract90)   HTML27)    PDF (858KB)(36)      

    Artificial intelligence (AI) assisted planting can improve in the precise management of protected horticultural crops while also alleviating the increasingly prevalent problem of labor shortage. As a typical representative of labor-intensive industries, the strawberry industry has a growing need for intelligent technology. To assess the regulatory effects of various AI strategies and key technologies on strawberry production in greenhouse, as well as provide valuable references for the innovation and industrial application of AI in horticultural crops, four AI planting strategies were evaluated. Four 96 m2 modern greenhouses were used for planting strawberry plants. Each greenhouse was equipped with standard sensors and actuators, and growers used artificial intelligence algorithms to remotely control the greenhouse climate and crop growth. The regulatory effects of four different AI planting strategies on strawberry growth, fruit yield and qualitywere compared and analyzed. And human-operated cultivation was taken as a reference to analyze the characteristics, existing problems and shortages. Each AI planting strategy simulated and forecast the greenhouse environment and crop growth by constructing models. AI-1 implemented greenhouse management decisions primarily through the knowledge graph method, whereas AI-2 transferred the intelligent planting model of Dutch greenhouse tomato planting to strawberry planting. AI-3 and AI-4 created growth and development models for strawberries based on World Food Studies (WOFOST) and Product of Thermal Effectiveness and Photosynthesis Active Radiation (TEP), respectively. The results showed that all AI supported strategy outperformed a human-operated greenhouse that served as reference. In comparison to the human-operated cultivation group, the average yield and output value of the AI planting strategy group increased 1.66 and 1.82 times, respectively, while the highest Return on Investment increased 1.27 times. AI can effectively improve the accuracy of strawberry planting management and regulation, reduce water, fertilizer, labor input, and obtain higher returns under greenhouse production conditions equipped with relatively complete intelligent equipment and control components, all with the goal of high yield and quality. Key technologies such as knowledge graphs, deep learning, visual recognition, crop models, and crop growth simulators all played a unique role in strawberry AI planting. The average yield and Return on Investment (ROI) of the AI groups were greater than those of the human-operated cultivation group. More specifically, the regulation of AI-1 on crop development and production was relatively stable, integrating expert experience, crop data, and environmental data with knowledge graphs to create a standardized strawberry planting knowledge structure as well as intelligent planting decision-making approach. In this study, AI-1 achieved the highest yield, the heaviest average fruit weight, and the highest ROI. This group's AI-assisted strategy optimized the regulatory effect of growth, development, and yield formation of strawberry crops in consideration of high yield and quality. However, there are still issues to be resolved, such as the difficulty of simulating the disturbance caused by manual management and collecting crop ontology data.

    Table and Figures | Reference | Related Articles | Metrics