The Research Method Of Extraction And Classification On Crater Appearance By Blending Features

With the rapid development of the implementation of the domestic and international star exploration project and related technologies, especially to analyze and deal with the massive moon data by Chang-e I and Chang-e II shooting. On account of multimode lunar crater image in the complex illumination and geographical environment, the existing feature extraction and classification methology still can not recognize the craters exactly and rapidly,the researchers have extreme interesting in the new blending feature and classfication mechanism.This thesis mainly researches three parts which are lunar crater image preprocessing,feature extraction and classification methods based on digital image processing, computer vision, pattern recognition and artificial intelligence. Firstly, it researches the crater image anisotropic diffusion and Gauss filter in the preprocessing part, it also proposes a new weighted multi direction Sobel operator edge processing image detail enhancement method,and researchs the attribute morphology processing methods for refining crater’s feature by area, gray contrast and shape attributes threshhold. Then, it introduces the crater area Haar-like feature and PHOG(gradient direction histogram) feature in the features selection and extraction part, and proposes that using the covariance matrix efficiently fuse multiple features on crater image to express with multi modal crater’s blend feature.It studys the Bayes classifier, AdaBoost classifier and support vector machine classifier in the crater area classification and recognition. The crater image experiments of the combinations of different preprocessing, different features and different classification algorithms are carried out. It realizes the crater image classification and gives the analysis of the experiment.Experimental results show that reasonable the fused feature combination constitute the covariance matrix descriptor has good and quick lunar crater region resolution.In this study course, we designed a structure framework of lunar crater appearance by fusing feature, and verified these methods and algorithms by this software package.