Correlation matrix feature extraction based on spectral clustering for hyperspectral image segmentation

Recently, the correlation matrix is used for dimension reduction by combining the greedy modular eigenspace and the positive Boolean function. However, it is hard to determine the threshold values for the greedy modular eigenspace. In addition, spectral clustering based on a similarity matrix, an affinity matrix, or a kernel matrix has become a popular clustering algorithm. Therefore, in this study, the spectral clustering is applied to the correlation matrix of bands, and the corresponding membership values determine the transformation matrix. Experimental results show that the proposed method achieves good segmentation performance on the Indian Pine site dataset, and the proposed feature extraction outperforms principal component analysis and independent component analysis.