The Spatial Spectral K-Nearest Neighbor Minimax Label Projection Semi-supervised Classification For Hyperspectral Remote Sensing Image

Abstract

Abstract Aiming at the issues of classification performance degradation of the algorithm due to too much redundant information of hyperspectral remote sensing images and incomplete classification of MMLP algorithm due to sparse graph, the Spatial Spectral K-Nearest Neighbor Minimax Label Projection (SSKMMLP) semi-supervised hyperspectral image classification algorithm was proposed, which is a semi-supervised hyperspectral image classification algorithm. SSKMMLP algorithm firstly uses Minimum noise fraction (MNF) and Principal component analysis (PCA) algorithm to reduce the dimension of image data. Secondly, MMLP algorithm is used to preliminarily classify the data after dimensionality reduction, and a new MMLP second algorithm is proposed to reclassify the unlabeled samples, to solve the problem that MMLP algorithm may not be able to completely classify. Finally, combined with spatial neighborhood information and classification results, all sample categories are judged again, so as to improve the classification accuracy further. Experimental results on data sets of Pavia University, Salinas, WHU-Hi-Longkou, WHU-Hi-Hanchuan, WHU-Hi-Honghu show that compared with the comparison method used, SSKMMLP algorithm only needs less class label samples to obtain higher classification accuracy.