Mineral Mapping with Hyperspectral Image Based on an Improved K-Means Clustering Algorithm

Abstract

Mineral mapping with hyperspectral images has been demonstrated as an effective way for land resources survey. K-means, as a typical clustering algorithm, is commonly used to process the object identification of hyperspectral images. However, due to the influence of mixed pixel, the matching of data points and cluster centers of the traditional k-means clustering algorithm is very difficult. Therefore, this paper proposes an improved k-means clustering algorithm to identify the mineral types from the AVIRIS hyperspectral image of Cuprite mining area. This algorithm uses three methods to select the initial cluster centers and spectral information divergence instead of Euclidean distance for better measuring the similarity. Finally, by matching the clustering results with the mineral distribution map of this region and USGS mineral spectral library, it was found that the improved k-means clustering algorithm can get better clustering results and higher mineral mapping accuracy than the traditional algorithm.