Research On Identification And Classification Of Wheat Scab Disease Based On Convolutional Neural Network

Wheat has a wide range of planting areas all over the world,and the large-scale reduction of grain yields caused by wheat scab disease has occurred from time to time.In order to deal with the outbreak of wheat scab,it is usually used to spray pesticides in advance to prevent wheat scab.However,the problem of drug abuse often occurs in the process of pesticide spraying,which not only pollutes the environment,but also increases the cost of planting.In order to better solve the problem of pesticide pollution and reduce the abuse of pesticides,accurate and rapid identification of wheat scab disease and identification of grades have a guiding role in production.With the development of convolutional neural network,it provides a theoretical basis for nondestructive detection of wheat scab.In order to identify and evaluate the grade of wheat scab,this thesis will carry out research on the identification and classification of wheat scab disease images based on convolutional neural network theory and image segmentation algorithm.The main work of this thesis is as follows:(1)This thesis improves the convolutional neural network based on VGG-Net,and compares the recognition effects of various convolutional neural network models with different depths on wheat scab.The wheat images were collected in the experimental field environment,and the wheat images with different disease degrees were segmented.The dataset was expanded according to the image enhancement theory,and the wheat scab disease dataset was made.Neural network models based on Alex-Net,VGG-16 and VGG-19 were constructed,and the wheat scab identification experiment was carried out on the self-made dataset,and the results of the above three types of models were compared and analyzed.This thesis reduces the depth of the fully connected layer on the convolutional neural network model,reduces the number of neurons in the fully connected layer,and increases the depth of the convolutional layer.The improved VGG＿16 model test set recognition accuracy rate reached 98.0 %.The results show that the improved VGG＿16convolutional neural network structure has higher recognition accuracy.(2)This thesis studies the classification method of wheat scab based on fully convolutional neural network.Based on VGG＿16,a deconvolution module DGVGG＿16 for segmenting the wheat scab disease spots was constructed;upsampling is achieved through nearest neighbor interpolation and convolution,and the image is reconstructed with skip connections,and disease spot are segmented for the identified disease images.After training,the final test set pixel accuracy,average pixel accuracy and average intersection ratio are80.3%,69.8% and 60.2%,respectively.The disease grade was obtained by the ratio of the number of pixels of diseased wheat ears to the number of pixels of whole wheat ears.