Hyperspectral Data Processing: Algorithm Design and Analysis

3.7 Conclusions

This chapter presents a 3D ROC analysis for multiple-signal detection and classification. Its idea arises from the fact that the performance of a detector is generally measured by a likelihood ratio test which is indeed a real-valued function, both detection power and false alarm rate are actually determined by a value that thresholds a real-valued detector. This problem is not only a pattern classification, where classification accuracy is also determined by a threshold used for clustering data. Therefore, including a threshold as a parameter to account for performance analysis seems more realistic and effective. To substantiate its utility in versatile applications, four examples representing a wide range of applications are presented for demonstration, which are linear spectral unmixing and target detection for hyperspectral data, medical diagnosis in tumor detection, and tissue characterization for magnetic resonance images, chemical/biological agent detection for water monitoring, and evaluation of biometric recognition systems.