Classification Method of Rape Root Swelling Disease Based on Convolution Neural Network

Abstract

Both the seedling stage and the adult plant stage of rape can be infected with root edema, and the damaged roots swell to form tumors of different sizes and shapes. The incidence of rape root swelling at the seedling stage reached 17%, and the average incidence at the adult plant stage was 15%, resulting in a 10.2% reduction in rape production. The average plant height, number of siliques, number of kernels per horn, 1000-seed weight and yield per plant of healthy plants were significantly higher than those of diseased plants. Grading root lesions can help trace the root causes of root lesions. However, the method of grading is often performed manually by professionals at present, which has the problems of low speed and low efficiency. In order to solve this problem, a method for grading rape root swelling based on deep convolutional neural network is proposed in this paper. Firstly, a rape root swelling model based on convolutional neural network and regional candidate network was established, and then implement it on the deep learning Tensorflow framework Model, and finally compare and analyze the results. The rape root swelling model uses the VGG16 network to extract the characteristics of the rape root swelling image. The regional candidate network generates the preliminary position candidate frame of the rape root swelling, and Fast-RCNN realizes the classification and positioning of the candidate frame. The results show that this method can achieve rapid and accurate detection of healthy, first-level tumors, second-level tumors, and third-level tumors of four-level rape root swelling, with an average accuracy rate of 84.12%. The experimental results show that the accuracy rate can reach more than 90%.