Research On Grape Leaf Disease Image Data Enhancement And Detection Based On Deep Learning

Grapes are one of the important raw material of wine and daily popular fresh fruits in China.The planting scale of grapes is increasing year by year due to the growth of market demand.However,grapes are vulnerable to diseases in the growth process,resulting in damage to their quality and yield,which brings economic losses to growers.In the process of grape cultivation and management,growers mainly identify grape diseases through manual detection,which has problems of low efficiency and high error rate.With the development of computer vision technology,deep learning has been widely used in agriculture due to its advantages of high precision and high speed,especially in the field of crop disease recognition.Grape plants affected by disease often change in physiological structure and morphological structure.And the disease characteristics are usually obvious in the position of leaves.So the diagnosis of leaf disease is helpful to monitor the health status of the plant.In order to improve the effectiveness and accuracy of grape disease detection,this paper takes three common grape leaf diseases as the research object and uses deep learning method to study the related problems of grape leaf disease image data enhancement and disease detection.The main research contents and achievements are as follows:(1)Aiming at the problem that there are few samples of grape leaf disease image,an image enhancement method of grape leaf disease based on generative adversarial network is proposed.Firstly,combined with the advantages of Res Net and Dense Net,the residual-dense PRD module is proposed to make full use of the characteristics of the convolutional layer to enhance the information transmission between the convolutional layers.Then the residual module in the U-GAT-IT model generator is replaced with the six PRD module to improve the feature extraction ability of the model,so as to improve the problem of poor image quality generated by the original generator.Taking grape healthy leaves as the source domain and grape diseased leaves as the target domain,the disease spots of corresponding diseases are generated on healthy leaves.Through the quantitative and qualitative analysis of the images generated by different models,the experimental results show that the improved model can better extract image features and generate high-quality images.(2)Aiming at the problem of small grape leaf disease spot area and low detection accuracy,a grape leaf disease detection method of YOLOv4-Grape which is based on YOLOv4 model is proposed.Firstly,K-means++ clustering algorithm is used to regenerate anchor boxes on the dataset to improve the convergence speed of the model.Secondly,PSA module of attention mechanism is used to replace all the 3×3convolution in the feature extraction network to obtain richer image feature information.Finally,the maximum pooling layer in the SPP module is replaced by three dilated convolution with different dilated rates to enlarge the receptive field without losing spatial information.The improved U-GAT-IT model is used to expand the dataset and detect the disease on the dataset.The experimental results show that the map value of YOLOv4-Grape model is 3.66% higher than that of YOLOv4.