Pixel Unmixing via Information of Neighboring Pixels

Abstract

Pixel unmixing is a highly undetermined procedure. Due to sensor measurement noise, environmental changes, and intra-class variance, it is very difficult to make a high-quality classification. It is even more unlikely to decompose the mixed pixel with a high degree of confidence. Conventional algorithms for linear pixel unmixing are only concerned with a single pixel or a small set of pixels with the same mixing proportions. In this paper, a novel method for pixel unmixing based on the classification information of neighboring pixels is presented. In addition to the widely used residual root mean square error, four performance metrics are proposed for comparing quantitatively the classification accuracy. While conventional methods aim to minimize residual errors, the proposed method tries to achieve the best possible correct end-member combination. A case study shows that, in terms of both RMSE and new performance metrics, the proposed method achieves significant improvement over conventional algorithms.