Research On Crop Disease Diagnosis Technology Based On Image Recognition

In recent years,the outbreak of crop diseases has shown a trend of aggravation and scope expansion,causing significant losses to agricultural safety production and social property.With the development of science and technology,only relying on manual diagnosis can no longer meet the needs of modern agricultural production.Combining image recognition technology to diagnose crop diseases has the advantages of accuracy and efficiency,and has become an important way to solve this problem.Therefore,in-depth research on its related technologies has important practical significance and application value.From the perspective of taking into account the accuracy and speed of disease diagnosis algorithms,this paper focuses on the research of disease image preprocessing algorithms,segmentation algorithms,crops based on traditional machine learning and convolutional neural networks(CNN)in view of the shortcomings of current disease diagnosis methods in practical applications.Disease diagnosis algorithm,the main research contents are as follows:(1)Aiming at the problem that it is difficult for a single filtering algorithm to balance noise reduction performance and image quality,a hybrid superposition mean filtering algorithm using multiple filters is proposed.Aiming at the incomplete segmentation of disease spots caused by the two backgrounds of the environment and leaves in the disease image,combined with the characteristics of the disease image,a segmentation algorithm based on pixel difference is proposed.(2)Research and design crop disease diagnosis algorithms based on traditional machine learning.Aiming at the problem that the image information of crop diseases is complex and a single feature is difficult to accurately describe the disease,the color,texture and shape feature information of the image is extracted to represent the image.Aiming at the problem of slow color feature extraction,HSV(Hue Saturation Value)color space is introduced,and improvements are made through non-linear quantization to speed up calculations.Aiming at the problem of high texture feature dimension and poor stability,the rotation invariant equivalent pattern LBP(Local Binary Patterns)is introduced to replace the original operator,which reduces the feature dimension and improves the stability of the algorithm.Aiming at the problem of poor generalization ability of traditional shape features,Hu invariant moments are used as shape features to improve the applicability of the algorithm.Aiming at the problem of high dimensionality of the fused multi-feature information,Principal Component Analysis(PCA)is used to reduce the dimensionality of the data,And design a crop disease multi-classifier based on Support Vector Machine(SVM)for the classification of disease images.(3)Research and design crop disease diagnosis algorithms based on convolutional neural networks.Aiming at the problem of the single range of the receptive field when extracting features from VGG16,the spatial pyramid pooling layer is designed to replace the maximum pooling layer to increase the receptive field.Aiming at the problem of low discrimination between images of similar diseases,the method of adding auxiliary loss function to the original loss function is adopted to improve the diagnosis accuracy of the algorithm.Adjust the VGG16 network structure and design the training process of the model.Through experiments to integrate various modules and comprehensively evaluate the performance of the algorithm,based on the self-built data set in this article,we designed and conducted comprehensive experiments on disease diagnosis algorithms based on traditional machine learning and based on convolutional neural networks.The results show that both algorithms based on convolutional neural networks and traditional machine learning have excellent performance.Among them,the diagnostic accuracy rates were 92.1% and 90%,and the diagnostic time was 157.6ms and 327.5ms.