Hyperspectral Data Processing: Algorithm Design and Analysis

VII

Hyperspectral Signal Characterization

The hyperspectral signal coding in Part VI is designed to produce credible discrete versions of hyperspectral signals as fingerprints so that these fingerprints provide sufficient information of their own identities. But such signal coding does not necessarily tell you what a real signal looks like and how it behaves. In other words, a signal identity and its fingerprint is one-to-one correspondence relationship such as one-to-one identification between a person and his unique nickname where the nickname does not have to describe the person in detail. Using a more specific example for illustration we consider a set of signals, to be used for data transmission. When one of these signals is selected for transmission, it is its subscript instead of the signal itself being transmitted. According to information theory, if a fixed length coding is used, only bits required to derive a set of p code words corresponding to fingerprints of the p signals, for signal transmission without actually transmitting these signals themselves. This is because a signal can be identified by its subscript through its code word used as its fingerprint. By means of ...