Research On Classification Method Of Hyperspectral Image Based On Spatial Information

Hyperspectral image classification is an important research direction in remote sensing image processing.Hyperspectral images have two spatial dimensions and one spectral dimension.Traditional research on hyperspectral image classification usually only pays attention to the "spectral curve" of hyperspectral images.However,in actual features,the distribution of features often has a "spatial smoothing characteristic",that is,pixels that are closer together have a higher probability of belonging to the same category.If only the spectral characteristics of hyperspectral images are classified,and the spatial features in the image are ignored,it will be difficult to improve the classification accuracy in the case of small samples,over-fitting,"same spectrum but different objects" and "same objects different The reason of“spectrum” affects the classification effect and other issues.In addition,in hyperspectral image classification,the technology that only uses spectral information to classify has become mature,and there is little room for improvement on this basis.Therefore,the introduction of spatial information into hyperspectral image classification has increasingly become a study of hyperspectral image classification Hot spot.Based on the above content,this article focuses on how to effectively combine the spatial information of the image to assist image classification under the existing classification methods.The specific work of this article is as follows:1.Aiming at the problem of limited image classification accuracy in the case of insufficient label samples in hyperspectral images,this paper proposes a semi-supervised classification method for hyperspectral images that combines spatial neighborhood information.Under the premise of combining the texture features of the image,on the one hand,the neighborhood of the labeled sample is selected as the sample co-selection range in the sample co-selection stage,and the specific neighborhood size selected is determined according to the credibility of the neighborhood of the labeled sample;On the one hand,after the classification result is obtained,the classification result is corrected by combining spectral information and neighborhood information to reduce the "noise" in the classification result.Experiments show that the two aspects of the method combined with neighborhood information are effective in improving the classification accuracy.In the case of few initial labeled samples,the method in this paper is compared with the supervised classification method SVM and other semi-supervised classification methods with better classification results.The method in this paper is significantly higher than other comparison methods in classification accuracy.The experimental results show The method in this paper effectively combines the spatial information of the image when there are few initial labeled samples,and improves the accuracy of image classification.2.In view of the problem that only single information is often selected for image classification in hyperspectral image classification and the undifferentiated selection of neighboring samples in the joint sparse representation method may cause "misclassification",this paper proposes a multi-angle and joint sparse representation Classification method for hyperspectral images.In this method,on the one hand,the residual value obtained by the joint sparse representation classifier and the correlation coefficient between the pixels are used for decision-making fusion classification,to achieve complementary classification effects,and effectively combine the neighborhood information,spectral information and texture information of the image;on the other hand,in the joint sparse representation classification,this method considers The correlation and locality between neighboring pixels.The results of comparison experiments show that compared with other comparison algorithms,the method of fusing multi-angle information in this paper has achieved excellent classification results,which proves the effectiveness of this method.