Research And Application Of Crop Leaf Diseases Detection Based On Deep Learning

Traditional tomato disease detection usually adopts manual detection,but it has the problems of low detection efficiency and high labor consumption.Nowadays,crop disease detection methods based on convolutional neural networks have received more and more attention from researchers.However,deep learning based detection can not be widespread because of its high hardware requirements.Therefore,in order to overcome the problem,this paper deeply researched the object detection algorithm of tomato disease leaves.The specific work contents and results are as follows:(1)We construct a tomato disease leaf data set for target detection algorithm: the 5types of tomato leaves required in this experiment are separated from 61 types of leaves,including healthy leaves and four kinds of diseased leaves.The image annotation tool labelImg is used to labelthe detected leaves.Then,the image data are made into Pascal VOC data set format for the experiment.We analyze the data of the detection result bound.(2)We construct a tomato disease detection model based on YOLO v5.Firstly,the Mosaic data augmentation method is adopted in the experiment to improve the convergence rate of the network and avoid overfitting.Then,we use DropBlock regularization technology to reduce the memory requirement of the model.Finally,We use PANet to remove noise from the feature information and locate the pixels of the feature network The mAP value of this model on the experimental data set is 96.52%,which is higher than 95.85% of the Faster RCNN.Experiments show that the YOLO v5 algorithm based on deep learning can complete the detection Faster and more accurately than Faster RCNN.(3)We propose the YOLO v5 s algorithm based on DIoU loss function.The loss function is optimized by reducing the confidence of the target in the background We use DIou loss as the loss function of the background target.Minimize the normalized distance between anchor Frame and Target frame.Experiment results show that compared to the original model,this model has faster convergence rate and can return to the target frame more accurately and faster when the target frame overlaps or even contains the detection frame.(4)Implement a tomato disease detection system based on YOLO v5 algorithm,the main function of this system is to diagnose tomato disease leaves through image detection.Using the PyQt framework,the system is divided into four layers: data processing layer is responsible for the identification data of tomato disease leaves,data loading layer is responsible for loading pictures and models,functional logic layer is responsible for identification and detection,and display layer is responsible for displaying detection results to users.The system calls the images from the background and displays the detection results to the pages after detection,which is a visual diagnostic tool for tomato diseases.