A Method of Corn Disease Identification Based on Convolutional Neural Network

Abstract

In order to accurately identify common diseases of corn, a method of corn disease identification based on convolutional neural network was proposed. This method firstly on the basis of maize disease images, wavelet threshold method, histogram threshold method for image background, diagnosis and disease spot segmentation preprocessing form sample library, then using convolution neural network to build a structure with five layers deep learning model samples to study, and use the mean square deviation, gradient method and control the learning process, the process combined with the feature of migration method learning image of maize disease finally to obtain the feature set is used to identify the disease of corn pictures, and forms an on-line identification system. The experimental results of three samples in Changchun show that the method can effectively identify the common diseases of maize, such as big spot, head smut, rust, smut, crown rot and sheath blight, and the comprehensive identification rate can reach 96.8%, which can be applied to practical production management.