Research On The Detection Method Of Major Diseases And Pests In Cotton Leaves Based On Deep Learning

As the world’s largest cotton producer,China has been a world leader in cotton cultivation area and production.However,cotton pests and diseases are one of the main factors affecting the yield and quality of cotton.Pests and diseases are mostly found in the cotton leaves,and timely detection and control of cotton leaf pests and diseases are essential for cotton yield and quality.The traditional methods of cotton leaf pest detection have the problems of high human and material costs and low detection accuracy.In recent years,with the continuous development of image processing technology and the improvement of computer hardware performance,deep learning-based detection methods have been widely used in pest and disease detection.In this paper,we study the detection of major cotton leaf pests and diseases based on deep learning,and the main research contents and conclusions of the paper are as follows:(1)Construction of cotton leaf pest and disease dataset.To address the lack of cotton leaf pest and disease datasets and the problem of poor targeting,we collected images of cotton pests and diseases in the growing cycle of Alar city and surrounding regiments of the First Division of Xinjiang Production and Construction Corps,and constructed a cotton leaf pest and disease dataset for deep learning training and testing.The dataset covers four main types of cotton leaf pests and one type of healthy cotton leaf images,and a total of 4818 original cotton leaf pest images were collected,and different data enhancement methods were used to equalize different types of pest images,and the original images were expanded to 12,641,and Labelme was used to annotate these images for cotton leaf pest detection research.(2)Localization improvement of target detection model.Firstly,the attention mechanism CBAM is added to strengthen the features of cotton leaf pest and reduce its influence by noise and environment in the connection of Backbone and Head of YOLO v7 model.Second,4-fold downsampling is performed in the Head part of YOLO v7 model to change the three-scale detection into four-scale detection,which helps to improve the multi-scale target detection performance of the algorithm and solve the problem of small targets and difficult feature extraction of cotton leaf pests.Finally,the constructed cotton leaf pest data set was brought into the improved YOLO v7 detection model,and the experimental results showed that the m AP of the improved YOLO v7 model was 82.3%,which was 7.9% higher compared with YOLO v5 and 4.1% higher compared with YOLO v7.The improved YOLO v7 model has better detection performance and can detect cotton leaf pests and diseases more accurately.(3)Design and implementation of cotton leaf pest detection system.Based on the self-defined dataset and the improved YOLO v7 model,the system was developed using Python language,Py Qt and Pytorch framework.The system includes registration and login module,image upload module,cotton leaf pest detection module and detailed information module.The system can help cotton farmers quickly manage cotton leaf pests and diseases by providing accurate guidance on cotton pest and disease information,and effectively prevent the expansion of pest and disease events.Although this paper focuses on cotton,the model and method used in this system can be extended to other crops and has high application value.The system is important for the detection and control of cotton leaf pests and pests and cotton production.