Research On Intelligent Test Method Of Multiple Crop Varieties Based On AI Cloud Computing

Corn,rice,soybean and wheat are essential grain crops globally and crucial for human nutrition.Traditional crop phenotyping relies on manual measurement,resulting in subjective and imprecise results.Existing automated methods have low accuracy due to low-resolution imaging and traditional image processing algorithms,limiting their applicability to multiple crop varieties.Thus,this study proposes an AI cloud-based intelligent phenotyping method for multiple crop varieties.It aims to achieve accurate extraction of yield-related parameters such as grain length,width,area and perimeter quickly and at high throughput.The main research contents are as follows:(1)The creation of a multi-class crop grain deep learning dataset based on corn,rice,soybean and wheat.Using an Android high-definition camera to capture 500 high-definition images of corn,rice,soybean and wheat grains on A4 white paper backgrounds and using the HJ-Z60 high-speed camera with a white LED backlight board to capture 500high-definition images of corn,rice,soybean and wheat grains on the same background.An AI semi-automatic labeling method was designed and used to label the 4,000 high-definition crop grain images collected with the minimum outer bounding rectangle of the grains as the labeling box.Finally,the images and labeling files were prepared into YOLO and COCO format datasets for algorithm training and testing.(2)The project developed a crop multi-variety seed detection algorithm based on an improved YOLOv7 algorithm.The algorithm includes several optimization modules such as small target detection layer,SPD-Conv,GAM-Attention and multi-class NMS.Through parameter ablation experiments,the optimal algorithm optimization scheme was verified.The final optimized crop multi-variety seed detection algorithm achieved a precision of0.998,0.998,0.998,0.996,recall rates of 0.993,0.999,0.963,0.993,m AP@0.5 of 0.991,0.998,0.965,0.995 on test sets of corn,rice,soybean and wheat,respectively.Compared to Cascade R-CNN,Faster R-CNN,Retina Net,SSD and YOLOX models,the proposed algorithm has the lowest parameter volume of 36.91MB and its floating point operations per second(Flops)is 135.2G,both of which are the lowest.The processing time for a single image is 24.9ms,the detection speed is the fastest.The algorithm also performed excellently in the detection and classification of multi-target crop seed data sets.(3)The project developed a crop multi-variety intelligent breeding software based on QT interface and DAHENG camera.The software is designed using the Pyside2 framework and can collect high-definition images of multi-variety crop seeds,detect and calculate seed granular parameters,save and return real-time breeding results parameters.The software achieved high measurement accuracy and efficiency,with R~2 values of 0.952,1.000,0.990,0.991,MAPE of 2.894%,1.874%,0.892%,1.318%,RMSE of 2.489,1.055,0.989,2.315for corn,rice,soybean,wheat,respectively.It can accurately measure around 300 seeds per cycle.In the average grain length and grain width measurements of 100 groups,the R~2 of the systematic and manual measurements of corn,rice,soybean and wheat were 0.990,0.971,0.985 and 0.959,MAPE of 1.118%,2.231%,1.585%,4.548%,RMSE of 0.167,0.175,0.172,0.286,the R~2 of the particle width system measurement and the manual measurement were 0.988,0.972,0.978,0.946 and the MAPE were 0.971%,3.811%,2.338%,11.917%,RMSE of 0.084,0.227,0.164,0.294,the comprehensive results show that the software system also has extremely high accuracy for grain length and grain width measurement;(4)Based on AI cloud computing,this study presents the design of a web-based and Android-based intelligent crop breeding system.The study deployed crop seed detection and seed granule characterization algorithms on Alibaba Cloud GPU servers.A Flask-based web interface and functionality interface were designed for the web-based system,while an Android Studio-based user interface and functionality interface were designed for the mobile app.The resulting system integrates online batch uploading of multiple crop seed images,crop seed detection,seed granule characterization,and real-time storage and return of breeding results.The system also includes an intelligent cloud-based method for crop breeding and an intelligent mobile app for seed detection,seed granule characterization,and online editing of breeding results.In conclusion,this study presents an intelligent crop breeding system based on YOLOv7 algorithm and cloud computing technology,which has significant practical and promotional value.The system can improve crop production efficiency and quality,contribute to agricultural production and food security,and provide effective technical support for crop yield trait improvement and breeding.