A Fuzzy Clustering-Based Feature Extraction Approach for Hyperspectral Target Detection

The sufficient spectral information contained in hyperspectral image provides assistance for many applications. But on the other hand, it causes information redundancy and increases computational costs. To surmount these obstacles, an intuitive approach is to reduce dimension. In this paper, a fuzzy relative-weighted K-means (FRWKM) clustering algorithm is proposed to extract feature by integrating target spectral information. First, graph segmentation technique is utilized to determine the optimal number of clusters. Then, the fuzzy cluster centers as well as the membership degree of each band to each center can be obtained based on a weighted fuzzy clustering criterion. Concatenating these centers, the final low-dimensional feature cube is obtained. The target detection experiments on three real hyperspectral images show the preponderance of the proposed FRWKM over the compared dimensionality reduction methods.