Research And Application Of Corn Leaf Disease Detection Method Based On Improved YOLOv5s Model

Corn is one of the important food crops in China,and its planting area and production have become the first among food crops.However,corn often suffers from various diseases during growth,especially the leaf parts are often affected.These diseases are harmful to production and affect the yield and quality of corn.Traditional corn leaf disease detection methods mainly rely on expert judgment,which is subjective and cannot meet the demand of real-time.Deep learning-based crop disease detection technology has the advantages of high accuracy and high speed.Therefore,it is important to construct a corn leaf disease detection model to improve corn yield and quality and reduce farmers’ economic losses.The main work accomplished in this thesis is as follows:(1)A corn leaf disease dataset was constructed.In this study,images of corn leaf diseases were collected by manual photography and web crawlers,including five diseases,including Northern corn leaf blight,Southern corn leaf blight,Gray leaf spot,Rust,and Phaeosphaeria leaf spot.A total of 3712 images of corn leaf diseases in 5 categories were obtained.In order to ensure the balance of the dataset and avoid the occurrence of overfitting,the original images were subjected to data enhancement.At the same time,the dataset was divided into training and test sets in the ratio of 8:2,and annotation tools were used to annotate the corn leaf disease regions in order to produce a corn leaf disease dataset that meets the requirements of model training and test data formats.(2)A corn leaf disease detection method based on improved YOLOv5 s is proposed.Three target detection models,Faster R-CNN,SSD,and YOLOv5 s,are compared in the experimental dataset,and it is concluded that YOLOv5 s performs the best.For the actual detection scenario,the problem of low model detection accuracy due to many background disturbances and similar spot characteristics of corn leaf diseases exists.YOLOv5 s was selected as the benchmark model,and the CBAM attention mechanism was added to its neck structure to enhance the anti-interference ability of the model and improve the detection accuracy;the Rev FP network was used in the neck structure to enhance the information transfer between different network layers and improve the feature extraction ability for multi-scale disease spots.The improved YOLOv5 s model has an P of 84.38% and an R of 81.08%,a m AP@0.5of 84.89%,a m AP@0.5:0.95 of 60.69%,with a 3.48% improvement in P and a 4.75% improvement in R compared with the original YOLOv5 s,m AP@0.5 was elevated by 5.74%,m AP@0.5:0.95 was elevated by 8%.The experimental results showed that the constructed corn leaf disease detection model was superior and had a good detection effect,which could better achieve the disease detection of corn leaves.(3)A corn disease detection system was designed and developed.The system was developed with the YOLOv5s-based corn leaf disease detection model as the core,using a frontand back-end separation model,and deployed on a mobile platform using the Flask framework.The system has three main functions,such as disease information profile,disease detection and disease control strategy.Users can upload corn leaf disease images,get the disease type and disease spot location detected in the images,and obtain disease related pathological knowledge and control measures.After testing,the system can detect corn leaf diseases more accurately,providing strong support to promote the development of agricultural intelligence and precision.