Automatic Recognition and Classification of Tomato Leaf Diseases Using Transfer Learning Model

Timely diagnosis of plant disease is important to get better crop yields. Infected plants can cause significant financial losses to farmers by lowering crop yields. It is extremely desirable to detect early signs and symptoms of plant diseases in a nation like India, where agriculture supports the majority of the population. More accurate and faster plant disease detection might assist in lowering the damage. With tremendous improvements and advancements in deep learning, the effectiveness and precision of plant disease detection and identification systems may be improved. The goal of this study is to discover leaf illnesses found in tomato crops and reduce the financial losses caused by the diseases. We have implemented transfer learning using a pre-trained Squeeze Net Model to detect and classify tomato leaf diseases. Our model can automatically detect 9 types of deadly diseases that are very common in tomato crops. We have acquired 10 classes (one healthy leaf class and 9 diseased leaf classes) consisting of 16,012 tomato leaf samples from a benchmarked Plant Village dataset to train and validate the suggested method. On the public dataset, the class-wise classification precision rate varies from 77.9% to 99.6%, and the overall classification accuracy of the suggested model is observed as 93.1% which is a significant enhancement in performance over previous tomato disease detection techniques. 

Keywords: CNN, Deep learning, Plant disease, Tomato, Transfer learning, SqueezeNet.