Research And System Realization Of Tomato Leaf Disease Recognition Based On Deep Learning

Tomato is one of the most common vegetables in our daily life,which is widely planted in the north and south of China.In recent years,with the improvement of people’s living standards and the rapid development of agricultural products market,the planting area of tomato is increasing.However,due to natural disasters,diseases and insect pests,improper management and other factors,crop diseases and insect pests are also increasing,directly affecting the quality and yield of tomato.In this paper,the common diseases of tomato leaves,such as powdery mildew,early blight,late blight and red spider damage,are identified.Through the application of image preprocessing,image enhancement and deep neural network modeling technology,a deep learning model suitable for mobile terminal deployment and rapid recognition is constructed.Finally,the model is transplanted to Android terminal Carry out offline detection application.The main research contents are as follows:1.Based on the review of the research progress of crop diseases at home and abroad,this paper analyzes and puts forward the idea of building a fast off-line recognition system for Tomato Leaf Diseases suitable for the mobile end,and introduces the traditional machine learning algorithm and the pooling,downsampling,activation function and other technologies in convolutional neural network,which provides technical theoretical support for the subsequent construction of the model in this paper.2.Based on the Plant Village tomato data as the basic data source,and through the field collection of disease images for sample data supplement,image data was preprocessed and data enhancement technology,and finally 10 types of disease and healthy leaf sample training set,verification set and test set were selected and determined as 24196,3027 and 3027 pictures respectively.Based on this training data,training comparison analysis is carried out on Inception-v3,Resnet50 and improved Resnet18 model.Finally,the improved Resnet18 model is more suitable for this kind of small sample data recognition,and its best training accuracy is 0.9623.3.According to the final determined training model,through the transformation of TFlite framework model,and the introduction of the design and development of Android mobile terminal,through the final deployment and performance test of the detection system,the tomato disease recognition system designed in this paper can get 91% recognition accuracy in the offline state,which meets the needs of tomato farmers to identify and control ten common diseases It has certain application and promotion value.