Tomato Disease Detection and Classification by Deep Learning
Huiqun Hong, Jin‐Fa Lin, Fenghua Huang
Abstract
Tomato is widely cultivated in China as a kind of fruit or vegetable. In the whole growth cycle of tomatoes, there are many types of tomato diseases and pests, therefore, the detection and diagnosis of these diseases are very necessary. Many Deep Learning architectures have been implemented for plant disease detection. In this research, transfer learning is used to reduce the size of the training data, the time and the computational costs when building deep learning. And 9 types of disease leaves including healthy tomato leaves are classified. Five deep network structures of Resnet50, Xception, MobileNet, ShuffleNet, Densenet121_Xception were applied to perform the feature extraction. Those network structures with different learning rates were compared in experiment. Adjust the appropriate training parameters and test those networks. Compared the five convolutional neural network, the parameters and the average accuracy are different. The best recognition accuracy of Densenet_Xception is 97.10%, but the parameters of Densenet_Xception are at most, the recognition accuracy of ShuffleNet is 83.68%, and the paramenters are small. Through comparison in this paper which provides model support for the further development of intelligent tomato disease diagnosis system based on smart phones and other mobile terminals, and is of great significance for assisting tomato pest control decision-making.