Study On The Automatic Detection Of Tomato Diseases And Pests Based On Leaf Images

There are many species of tomato diseases and pests,and the pathology of which is complex.It is difficult and error-prone to simply rely on manual identification.For the ten most common tomato diseases and pests in China,this thesis collects a large number of leaf images of tomato diseases and pests,explores the detection algorithms based on Keras/TensorFlow deep learning framework,and implements the detection models of tomato diseases and pests by training.The main work of this thesis is as follows:1.Build an image dataset of tomato diseases and pests,which contains leaf images of ten common tomato diseases in China as well as some normal ones.The dataset has a total of 11 categories,7040 images,each category has 640 images.2.Build a multi-layer convolution neural network,which is composed of an input layer,four convolution layers,four pooling layers,two fully-connected layers and an output layer.The training process employs the technologies of data augmentation,dropout,SGD(Stochastic Gradient Descent),etc.The trained model achieves an average classification accuracy of 84%.3.Implement the convolution neural network classification model with transfer learning technology.(1)Employ the first fully-connected layer of VGG16 model as leaf image feature extractor,and SVM(Support Vector Machine)as the classifier,the trained model achieves an average classification accuracy of 88%.(2)Set the output nodes of VGG16 to 11,(each corresponding to one category of leaf images),use the fine-tuning technology to retrain the last fully-connected layer,the trained model achieves an average classification accuracy of 89%.4.Considering the portability,this thesis transplants the designed multi-layer convolution neural network to the Android operating system.Test results show that the system takes an average of 0.2s to detect a tomato leaf image,which can meet the real-time requirements.