Detection Of Maize Leaf Diseases Based On Improved YOLOv5 Model

As an important food crop in China,maize is closely linked with the national economy.During the growth and development of maize,leaf diseases seriously interfere with the normal growth of maize.If not treated in time,the yield and quality will decline.Therefore,it is of great significance to find and diagnose the disease as early as possible and take corresponding control measures to improve the yield and quality of maize.The traditional deep learning method lacks in recognition accuracy and recognition efficiency.This thesis effectively explores the corn leaf disease based on YOLOv5 target detection algorithm,improves the Backbone end and Neck end respectively,and develops a corn leaf disease detection system,which has achieved remarkable results in improving the detection effect of corn leaf disease.The main research work of this thesis is summarized as follows:(1)Data preprocessing of maize leaf disease image.The data set of PlantVillage and the original data set collected by myself were analyzed,and finally the data set containing four types of data,namely 478 pieces of healthy leaves of corn,503 pieces of big leaf spot of corn,515 pieces of small leaf spot of corn,and 445 pieces of rust of corn,was selected.LabelImg software is used to classify and label the collected leaf image data,and five data enhancement techniques are used to expand the data set to alleviate the network over fitting problem.(2)Improve the Backbone end and Neck end of YOLOv5 model.To solve the problem of interference caused by complex background and other factors,ECA attention mechanism is introduced into the Backbone end of YOLOv5 model,which enables model learning to ignore information irrelevant to the detection task in the input image and enhance the ability to extract key features.As the model with attention mechanism may ignore some other useful information while selectively strengthening input features,BiFPN bidirectional weighted feature pyramid network is introduced at the Neck end of the model to achieve more comprehensive feature fusion;The experimental results show that the EBYOLOv5 model,which integrates ECA attention mechanism and BiFPN bidirectional weighted feature pyramid network,makes the input image more reliable,the target detection effect better,and achieves the goal of accurate disease detection.(3)Design and implementation of corn leaf disease detection system based on EBYOLOv5 model.Using the Django framework and the improved EB-YOLOv5 detection model in this thesis,the detection system of maize leaf disease is developed.Through local image detection and photo uploading,we can quickly obtain the type information of the disease and get the knowledge of related diseases.In practical application,several common kinds of maize diseased plants and healthy maize leaves were detected,and good detection results were obtained.