Study On Ensemble Learning Classification Of Multi-spectral Remote Sensing Imagery

With the guarantee quality and quantity of satellite sensors, in view of the multispectral remote sensing image classification research theoretical arguments continue to increase, compared with traditional remote sensing image, multispectral remote sensing image spectrum information rich, spatial information obvious enrichment.The traditional classification method is not very well reflect the characteristics of multi-spectral remote sensing images precise terrain, can not improve the classification of the terrain feature. methods cannot reflect well the terrain characteristics of multispectral remote sensing image accuracy and cannot complete the extraction of the terrain characteristics. In order to overcome the disadvantages of traditional classification methods, In this paper, classification methods are studied based on integrated study of multi-spectral remote sensing image classification. Some key problems are studied which include the texture characteristics of multi-spectral remote sensing images and classification of characteristics of the simplified dimension reduction. The specific contents and conclusions of the research are as follows:In the aspect of integrated learning algorithm, two parts of content are studied. In the first part, after multiple classification algorithm is shaped based on the combination of multiple classification ECOC framework and binary classification algorithm Logit AdaBoost algorithm and is carried on the classification on the ALOS images, the superiority of the algorithm is proved compared to traditional AdaBoost algorithm and Logit AdaBoost algorithm.In the second part,in order to solve problems of unbalanced remote sensing image classification, the RUSBoost algorithm is turned into multiple classification algorithm by changing the base classifier output form and is applied in the presence of category unbalanced Pleiades high-resolution images. Under the fact that the overall accuracy is not affected a few categories of classification accuracy can be increased by using the proposed algorithm.In terms of texture features extraction, in this paper, LBPV method is applied on the extraction of multi-spectral remote sensing images to extract the image the LBP texture features and VAR contrast features. This method describes the spatial characteristics of the image better compared with the traditional method to extract texture. After spectralfeatures and texture features are extracted, classification feature space was optimized and the effect of classification accuracy is improved by using the ICA transform principal component analysis to extract independent image characteristics of higher order statistics.Finally, this article summarizes the implement process of multispectral remote sensing image classification based on the integrated study. Its effects are analyzed in the experiment combined with the proposed classification algorithm and feature extraction.Results show that the proposed method can effectively solve the problem of multispectral remote sensing image classification. And these effects are higher than the AdaBoost and the SVM classification on both the effect of information extraction and classification accuracy.