Abstract: In order to solve the problem of poor accuracy of single feature classification caused by the "same spectrum foreign body, same body foreign spectrum" of high-resolution remote sensing image, a high-resolution remote sensing image classification method based on multi-feature manifold discrimination embedding is proposed in this paper. Firstly, the spectral features and LBP texture features of high-resolution image data are extracted. Then, through the joint spectrum of sample data, the spatial distance of texture features and the corresponding category information, the inter class and intra class graphs of image objects are constructed to learn the discriminative manifold structure on high-resolution images, so as to ensure that the features of different features in the embedded space are separated as far as possible, the same ground features are closely clustered to ensure the similarity of spectral and texture features of the same ground features, and complete the effective extraction of spectral and texture identification features, so as to fully mine image features and effectively improve the classification accuracy of images. Experimental results on GF-2 remote sensing data set show that the proposed algorithm can effectively fuse multiple features, and the classification accuracy can reach 93.41%, which is better than the traditional methods.

Key words: high-resolution remote sensing image, spectral feature, texture feature, manifold learning