Research On Soybean Disease Recognition Based On Deep Learning

There may be many soybean disease problems in the process of soybean production and planting,and how to accurately identify soybean diseases and adopt corresponding solutions to reduce the spread of soybean diseases has become the focus of current research.Among them,soybean leaves are a high incidence area of soybean diseases,and their disease identification will have certain practical significance.Therefore,based on the key image processing technologies such as image background segmentation,generative adversarial network,convolutional neural network and transfer learning,and referring to the practical application of deep learning technology and transfer learning technology in the existing research results,this paper proposes a soybean leaf disease identification and analysis method,in order to provide a certain theoretical reference for the subsequent soybean disease identification and treatment.Specifically,the main research content of this paper includes the following parts.First,using eight soybean leaf diseases and one healthy image as research image samples,a generative adversarial network image data enhancement processing method is proposed,and the image background segmentation and leaf image translation experiments are used to realize the matching and comparison of the generative adversarial network image data enhancement processing method with the traditional image data enhancement processing method,and verify the effectiveness of the generative adversarial network image data enhancement processing method.Second,after analyzing and processing nine research image samples by using the traditional data enhancement processing method and the generative adversarial network image data augmentation processing method,two datasets of training set A and training set B are formed,and the transfer learning technology based on Bayesian optimization is used to carry out the experimental analysis of the improved optimization performance of the volume integral class model,which confirms that compared with the traditional fine-tuning method,the Auto Tun classifier formed by the transfer learning technology based on Bayesian optimization is more conducive to improving the recognition accuracy of convolutional neural network.In addition,through seven evaluations,the Auto Tun classifier using the data images of the two training sets was comprehensively evaluated,and finally it was confirmed that the Dense Net121 pre-training model based on training set B could obtain high performance in seven indicators,and compared with the convolutional neural network model based on training set A,the accuracy of the convolutional neural network model based on training set B was relatively high,which further proved the application value of the image data enhancement processing method of generative adversarial network.Thirdly,an experimental system for image recognition of soybean leaf diseases is constructed based on the image data enhancement processing method of generative adversarial network,Auto Tun classifier and convolutional neural network pre-training model.This system mainly integrates three functions: image background segmentation,leaf disease translation,and leaf disease identification,which can realize the rapid identification and visual interpretation of 9 soybean leaf diseases,and meet the relevant functional requirements of soybean leaf disease identification and treatment.