23.5 Endmember Extraction

The reflectance Cuprite data scene in Figure 1.12 was used for experiments by PBS with BP/ID-BD to investigate an application in endmember extraction where a threshold value ε for ID-BD was empirically set to 0.01 to remove correlated bands. After BP/ID-BD, only 95 spectral bands were retained. Table 23.1 tabulates the first 50 spectral bands after BP/ID-BD using 7 BP criteria: variance, SNR, skewness, kurtosis, entropy, ID, and negentropy. Interestingly, the interband correlation for this scene is clearly characterized by second-order statistics rather than by high-order statistics (HOS), where only a few bands are removed by ID-BD using three specific HOS criteria: kurtosis, ID, and negentropy.

Table 23.1 Seven BP criteria to produce the first 50 bands after BP/ID-BD with bands removed by ID-BD for Cuprite data by setting ε = 0.01.

Figure 23.1(a)–(g) shows the results of IN-FINDR-extracted mineral signatures among A, B, C, K, and M using BP/ID-BD-generated spectral bands given in Table 23.1, where the figures on the right and the left were obtained by PBS with/without using ID-BD, respectively. Also in Figure 23.1, a vertical line in each figure shows a cutoff threshold for determined and set by DDA = 26 that was the largest value obtained by Huffman coding among five mineral signatures.

Comparing Figure 23.1 with Figure 21.8, the results produced by BP/ID-BD-generated bands are different from the results the generated by using BP without BD. However, the improvement of endmember extraction was not really significant.