Abstract: An image classification algorithm based on Inception-v3 and migration learning is proposed to solve the problem that the current intelligent identification of tomato diseases is not accurate and time-consuming. Eight tomato disease leaves and healthy leaves were collected from the experimental field, the images were scanned into images using a 10 megapixel HD scanner, the images were classified into 9 folders, and the blade attributes were manually labeled. Finally, based on the Inception-v3 model combining migration learning algorithms to classify test healthy and diseased leaves, and compare them with traditional image classification algorithms （KNN、 SVM、 BP neural network） and non-migration learning algorithms. The experimental results show that, based on Inception-v3 model combined with migration learning algorithm can quickly and effectively identify grow healthy tomato and diseased tomato in tomato disease image classification, and can identify tomato disease types efficiently. Among them, the classification accuracy of health and disease is 0.976 0, and the average accuracy of disease types is 0.929 7, which provides a certain degree of support for tomato disease detection and prevention.

Key words: image classification, migration learning, tomato diseases detection, Inception-v3