Hyperspectral Data Processing: Algorithm Design and Analysis

8.1 Introduction

In order to find endmembers, two general approaches have been used in the past. One approach extracts all required endmembers simultaneously, referred to as simultaneous endmember extraction algorithm (SM-EEA) that has already been studied in Chapter 7. The other approach extracts endmembers one at a time sequentially, referred to as sequential EEA (SQ-EEA) which will be discussed in this chapter. In general, extraction of endmembers must be performed by finding all endmembers at once. Hence, an optimal EEA should be an SM-EEA. Unfortunately, this also requires an SM-EEA to conduct an exhaustive search for an optimal set of endmembers. Computationally speaking, this may not be a good option because the process will be extremely slow, especially when the number of endmembers grows. On the other hand, an SQ-EEA can become an acceptable alternative even if it may not be as optimal as an SM-EEA. As will be demonstrated by experiments, in most cases, an SQ-EEA is comparable to an SM-EEA with regard to performance of endmember extraction. Most importantly, two benefits gained by an SQ-EEA can really remedy two major drawbacks suffered from an SM-EEA. One benefit is significant reduction in computing time resulting from SQ-EEA that only needs to find one endmember at a time sequentially without finding all endmembers simultaneously as required by SM-EEA. The second benefit is that SQ-EEA retains previously generated endmembers while continuing to add new endmembers, an ...