Wheat Leaf Disease Recognition Based On Image Analysis

The research of image analysis-based for wheat leaf disease is very important for the effective control of wheat diseases, increasing wheat yield, reducing pesticides on wheat products and environmental pollution. On the basis of adequate summary about the advanced achievements at home and aboard, aiming at improving wheat disease recognition accuracy and efficiency, in this paper systematical research has been made on the key technology of wheat leaf disease image segmentation, feature extraction and disease diagnosis recognition.The main work and innovations are as follows:(1)In order to improve the segmentation accuracy of K-means clustering used on wheat lesion images, a segmentation algorithm of wheat disease image based on multi-channels by LDA method’s mapping and K-means’ clustering is proposed. Firstly, six color channels from RGB model and HSV model are obtained, and six channels of all pixels are layed out to six columns. Then one of these channels is regarded as label and the others are regarded as sample features, where these datas are grouped to carry out linear discrimination analysis,and obtain the eigen vector space according to the first three bigger eigen values,so the mapping value that the other five channels is exerted on the eigen vector space.Secondly,the mapping value is used as the input data for K-means segmentation method and the minimum and maximum pixels are used as the initial cluster center, to overcome the randomness for selecting initial cluster center in K-means. And the segmented pixels are changed into background and foreground,so that the K-means segmentation is the clustering of two classes for background and foreground. Finally, experimental result shows that the segmentation effect of the proposed LDA+K-means method is better than that of K-means method and Ex R method.(2)In order to improve the segmentation accuracy and reduce the segmentation running time of Gaussian mixture model used on wheat lesion images, a segmentation method based on PCA and Gaussian mixture model is proposed in this paper. Firstly, in full view of the color information of an image, three primary color channels of the image are obtained through the principal component analysis(PCA) method from R, G, B or H, S, V color channels of this image. Secondly, the image is divided into many blocks, which are then sorted according to their mean pixel values. After sorting, those blocks lying on forward and rearward are selected to comprise a new pixel set by the Gaussian mixture model and further the corresponding Gaussian model parameters are obtained. Finally, the proposed method travels all pixels in the image and classifies each pixel into the corresponding Gaussian model category. Experiments show that, the segmentation error rates of the proposed method are lower than that of the Gaussian mixture model, K-means segmentation methods 5.33 and 17.23 percentage, respectively.(3)According to the low classification rates of the color moments,a feature extraction method of color moments based on multi-channel selected by PCA method is proposed.At firstly,the method gets three main channels of six channels from RGB model and HSV model,and obtain the first-order,second-order,third-order of these three channels.Secondly, color moments feature extraction method based on multi-channel mapped by LDA is proposed,by which three eigen vector space mapping values through LDA from six channels and the first-order,second-order,third-order of the best eigen vector space are obtained.The method makes full use of color information, selects good color space and raises the utilization rate of color features.(4)Through analyzing and extracting such color, texture and shape as thirteen features from wheat disease images, and selecting ten useful features by PCA method, these ten selected features are regarded as the inputs of classification models of wheat leaf diseases.(5)Through selecting ten useful features as input data, the recognition models including the 10-11-3 three layer BP neural net and SVM for wheat disease recognition are built to classify wheat leaf rust, powdery mildew and stripe rust disease images,respectively. Experiments show that the highest recognition correct ratios of the BP-based and SVM-based recognition models are 87.62% and 89.96%, respectively.(6)A wheat disease recognition algorithm based on random forest is proposed,with bagging method constructing forests,randomly bootstrapping to samples,growing trees with the maximum information gain and boosting result. The random forest-based algorithm is established to classify three kinds of wheat leaf disease images. Experiment shows that the highest recognition correct rate of the proposed method is 92.74%, and the random forest-based recognition method has higher recognition correct ratio than that of BP and that of SVM.