The improved N-FINDR endmember extraction algorithm and Its application in the oil analysis

Endmember extraction is the main step in the linear mixing spectrum unmixing. Its results determine the accuracy of the subsequent data processing. As an unsupervised endmember extraction algorithm, N-FINDR can automatically unmix spectrum. However, the extraction results could be instable, because it selects the initial candidate endmembers randomly. On the other hand, searching for the largest simplex value also involves determinant calculation. In this paper, the correlation between endmembers is introduced into the procedure of the N-FINDR, so as to improve the quality of the initial candidate endmembers. The improved N-FINDR algorithm is also applied to analyze oil thickness in the airborne hyperspectral image, so as to estimate the thickness more precise than the pattern matching methods. The experimental results show that the improved N-FINDR algorithm could not only obtain a more accurate and stable endmembers, but also reduce the computation to certain extent.