Convolutional Neural Network-Based Approach for Citrus Diseases Recognition

For the development of agricultural modernization, an automatic recognition network for crop diseases with the advantages of high efficiency, non-destructiveness and continuity is indispensable. In this paper, we construct a 13-layers convolutional neural network for common citrus diseases recognition, which improving the efficiency and reducing overfitting by using a stack of small kernel and dropout, etc. Finally, we conduct comparative experiments with other networks. The experiment results show that the performances of our network are better than CNN and AlexNet. It indicates that the network we constructed is an effective citrus diseases recognition method, which can provide technical support for the identification and prevention of citrus diseases.