Research Of Maize Disease Identification Based On Feature Fusion Of Canonical Correlation Analysis

The image processing technology can realize the detection of the maize growth status,quickly and accurately identify the maize disease,which can effectively reduce the impact of the disease on the maize growth process and promote the yield of the maize crop.Specifically,maize has a wide range of cultivation,but it is susceptible to diseases during the growth process.The specific research is divided into four parts:1)Image preprocessing of maize diseases.First,the type of maize disease was determined,and the disease image was filtered through median filtering.Secondly,the Kmeans clustering algorithm was improved to avoid the problem that the K-means clustering algorithm was easy to fall into the local optimal solution and enhanced the segmentation effect of edge information.2)Image feature extraction of maize diseases.Moments of each color channel in HSV color space were extracted by MATLAB as color features;the uniform local binary patterns(LBP)was used to extract texture features;the parameters of the area,circumference,circularity,elongation and rectangularity were extracted as shape features.3)Feature fusion of maize disease images.The extracted image features of maize diseases were fused by canonical correlation analysis(CCA).To effectively fuse the features,and consider the differences between sample features and category information,a multi set weighted local discriminant canonical correlation analysis(MWLDCCA)was proposed.The improved fusion method can not only ensure the intrinsic correlation between features,but also can describe the important features and secondary features.4)Construction of maize disease image recognition model.Two pattern recognition methods of BP neural network and support vector machine(SVM)were compared.SVM model was selected and the recognition rate was 96.0%.SVM parameters was optimized by particle swarm algorithm,and the recognition rate was increased to 98.0%.Finally,the visual interface was designed through VS2019 to realize human-computer interaction.Figure 34;Table 15;Reference 60...