20.6 Experiments for PSDP

This section conducts real image experiments to substantiate the utility of PSDP in a wide range of data exploitation. Since PSDP is preprocessing and very crucial to subsequent data analysis, its effectiveness can only be demonstrated through applications.

20.6.1 Endmember Extraction

A first immediate application of PSDP is endmember extraction that generally requires data dimensionality reduction. The data used for experiments for endmember extraction was the Cuprite data in Figure 1.12(a), where IN-FINDR was used as an endmember extraction algorithm to extract the five mineral signatures, A, B, C, K, and M, of major interest in the scene. Because of too many combinations of PI/PI that can be used for PSDP for an illustrative purpose, Figure 20.1(a) and (b) only plots results of extracting the mineral signatures by N-FINDR using PSDP with PI/PI = skewness/skewness and variance/kurtosis, respectively, where the x-axis is the number of prioritized dimensions and the y-axis is the endmembers extracted by N-FINDR. When the number of extracted endmembers was less than five, we need to know which endmembers were extracted in which case the extracted endmembers were also specified for reference. For example, in Figure 20.1(a) when PSDP was implemented by a pair of PIs, PI = skewness/PI = skewness for q = 4, 5, the number of extracted endmembers was 3 with extracted endmembers being identified as A, B, M. However, as q was increased to 6, the number of extracted ...

Get Hyperspectral Data Processing: Algorithm Design and Analysis now with O’Reilly online learning.

O’Reilly members experience live online training, plus books, videos, and digital content from 200+ publishers.