One immediate application of 3D ROC analysis is hyperspectral imaging where the strength/concentration of a detected signal is specified by the abundance fraction of a particular target which is actually a real value and plays the central role in data analysis. By varying the value of threshold τ in (3.25) the resulting different abundance fractions can be generated and different performances can be produced.
3.6.1 Hyperspextral Imaging
Two examples, hyperspectral target detection and linear hyperspectral mixture analysis, are presented in this section for illustration.
126.96.36.199 Hyperspectral Target Detection
In this section, the CEM specified by (2.33) in Section 2.2.3 of Chapter 2 will be used for target detection for the HYDICE scene in Figure 1.15 and panel signatures in Figure 1.16 will be used for desired target signature, t. Since we are only interested in panel detection, pixels other than panel pixels will be considered as background pixels. In this case, there are five panel classes identified as target classes, of which one class for each of five panel signatures and one background class which accounts for all nontarget pixels. Figure 3.7 shows the detection results for all the five panel signatures with each of the five panel signatures in Figure 1.16 used as the desired target signature t for detection. As shown in Figure 3.7, all the panels in five rows are detected very well and effectively.
The results in ...