Research And Development Of Grape Leaf Disease Diagnosis System Based On Deep Learning

China is a big producer of grapes.Disease control is an important part of grape production.Accurate and timely diagnosis is a prerequisite for effective disease prevention.There are many types of grape diseases.Their leaf symptoms are similar and difficult to identify but easy to confuse.Traditional grape disease diagnosis relies on artificial vision.Compared with artificial vision,machine vision based on deep learning is more suitable for grape disease diagnosis tasks.In recent years,the rapid popularity and performance evolution of smart phones have provided a hardware foundation for the mobile deployment of deep learning.This research focuses on the systematic application of deep learning technology in the diagnosis of grape leaf diseases.The specific content is as follows:1.The diagnostic information that computer vision tasks such as recognition,detection,and segmentation can provide in the diagnosis of grape leaf diseases is analyzed.Among them,image recognition and semantic segmentation are selected to form composite diagnostic information.The feasibility of constructing a multi-branch deep learning model to output compound diagnostic information is discussed.Then the program implementation method of the deep learning model and its mobile terminal deployment plan are determined.2.The image samples of grape diseased leaves were obtained by two methods:field collection and citing the Plant Village data set.With the help of the API provided by the computer vision library Open CV,the semantic segmentation and annotation of image samples is simplified.A dedicated semantic segmentation and annotation tool has been developed.A dynamic data enhancement program including geometric transformation,brightness adjustment and background replacement has been developed.3.A number of lightweight deep learning models are compared horizontally.Among them,Mobile Net V3 Large was selected as the feature extractor.Then additional image classification branches and semantic segmentation branche are added to form the basic model.Then its performance was tested.The structure of the basic model is optimized,and the amount of parameters is reduced by about 4.19 M without reducing the classification performance of the model,and the semantic segmentation performance of the model is improved at the cost of a small amount of calculation.The optimized model is tested on the grape diseased leaves dataset,the disease recognition accuracy rate reached 99.1%,the background judgment accuracy rate reached 99.3%,and the semantic segmentation m IOU reached 0.983.All three are improved compared with the basic model.The mobile terminal deployment of the model was realized through model conversion,and the inference speed of 11 FPS was reached in the actual test.4.Based on the above work,using software engineering methods and software development framework,a grape leaf disease diagnosis system was developed.Functions such as image sample collection and labeling,training data set organization and management,deep learning model training and conversion,and mobile deep learning disease diagnosis have all been implemented.The system provides a fully functional and easy-to-use user interface.The systematic application of deep learning in the diagnosis of grape leaf diseases is realized.