Study On Feature Extraction And Recognition Medthod For Wheat Disease Image

Wheat is one of the world's four major food crops,occupying an important position in the staple consumption structure of Chinese residents.The identification of wheat leaf disease based on image analysis is of great significance for the prevention and control of wheat diseases,the quantitative spraying of pesticides and the improvement of crop quality.Based on the research results at home and abroad,this paper studied the identify methods and techniques of segmentation,feature extraction and classification of wheat disease images which were infected by powdery mildew,stripe rust,leaf rust and stem rust.The main contents of the paper are as follows:Study on image preprocessing and segmentation algorithm for leaf disease of wheat.In order to find the appropriate algorithm to effectively eliminate nosie with slightly blur the images,median filter algorithm and wavelet transform methods were used to eliminate nosie of images.Study on image segmentation of wheat leaf disease based on k-means method.Extraction of characteristic parameters of wheat leaf disease.According to the color characteristics of disease images of wheat leaves,combined with predecessors' methods of the color feature extraction,in this paper,we used the color components of RGB color space to construct nine color characteristic parameters for disease.Texture features parameters were extracted by Gabor wavelet transform algorithm and gray level co-occurrence algorithm.Shape features parameters were extracted based on Hu invariant moments and the geometric shape parameters such as circularness,rectangle and elongation of the lesion,finally the feature space is 69.Characteristic parameter selection of wheat leaf disease.In order to reduce the dimension of the feature space,it is difficult to filter the features by using the traditional statistical analysis method when the sample value is crossed on a certain feature.A method of Variance algorithm combined with heuristic search was proposed to reduce dimension.Firstly,this method eliminated the features with little difference,then,for the remaining features,the method used SFFS algorithm to search the excellent feature subset,and eleven features were selected by this method.In order to illustrate the proposed method is effective,the PCA and LDA methods were used to reduce the dimension of the feature,and the dimensions of feature space were respectively thirty and three.Construction of wheat leaf disease identification model.The model was used to classify the four diseases of wheat by "one-to-one" decision-making method,and the samples were tested in the feature space obtained by different dimension reduction methods.The experiments were conducted based on 5-fold cross-validation,and results showed that the accurancy of feature sapces which were obtained by variance-SFFS proposed in this paper,LDA method and PCA method were 78.2%,92.98%and 93.89% respectively.It is showed that the variance-SFFS method designed in this paper couldeffectively select the useful characteristics,remove the redundant features and improve the classification accuracy effectively.