Hyperspectral Data Processing: Algorithm Design and Analysis

18.6 Conclusions

Many algorithms have been developed for various applications in hyperspectral data exploitation. Of particular interest is, “What pixel information do these algorithms really extract to achieve the goal that they are designed for?” This chapter investigates this issue via three categories of algorithms, EEAs, UTDAs, and an RX-anomaly detection algorithm, which have received considerable interest in the past. Since these algorithms are unsupervised and performed without prior knowledge, it is important to examine the utility of these algorithms. This chapter explores the insights into these algorithms. In particular, we address an important issue of whether these algorithms really do what they are designed to do. While doing so, four types of pixels, pure pixel, mixed pixel, homogeneous pixel, and anomalous pixel, are introduced for pixel information analysis. In addition, a set of custom-designed experiments are conducted, where the four types of pixels described above are used to evaluate the performance of these algorithms in terms of pixel information extraction. Interestingly, experimental results provide many intriguing findings in these algorithms that may help image analysts in selection of algorithms for specific applications.