Quantitative Analysis of Mixed Pixels in Hyperspectral Image Using Fractal Dimension Technique

Abstract

In this paper, a novel approach for identification of the components of a mixed pixel along with the quantitative analysis of the determined components of hyperspectral data has been proposed. The overall method is divided into three phases. First, the spectral sensitivity curves (SSCs) of the identified endmembers soil, water and vegetation using N-FINDR algorithm are generated. Next, the percentage of the mixture of the endmembers is computed for calculating the fractal dimension (FD). Afterward, the FD of SSCs is utilized for determining the mathematical model of the obtained curve. In order to verify that the adjacent pixels belong to the class identified using N-FINDR algorithm, two algorithms, i.e., spectral angle mapper and spectral information divergence, are employed for determining the spectral similarity of the mixed pixel with the reference spectrum. Lastly, the quantitative analysis of the SSC of the mixed pixel is performed by using the corresponding calculated FD in the derived mathematical model. The performance of the mathematical model is evaluated on the airborne visible infrared imaging spectrometer image data, and the outcome shows that the proposed framework achieves very promising results with a small number of mixed pixel samples.