Tomato leaf diseases classification with convolutional neural network

Abstract

In recent years, convolutional neural network (CNN) has developed rapidly and been widely used in plant disease detection. Tomato leaf disease is an important diseases of plant diseases. Therefore, it is necessary to improve the classification results of tomato leaf diseases. In this paper, a new CNN framework is proposed to classify tomato leaf diseases. The proposed framework includes several key deep-learning-based techniques: dilated convolution, batch normalization (BN), and the dropout after each convolutional block. Specifically, we apply dilated convolution to tomato leaf disease classification, which enlarges the receptive field without the reduction of image size. In addition, BN and the dropout are used to accelerate model convergence and reduce the redundant features generated by each convolutional block. Experiments show that the proposed model achieved the best results in both classification accuracy and the number of parameters compared with different CNN models.