Hyperspectral Data Processing: Algorithm Design and Analysis

14.6 Conclusions

Abundance-constrained linear spectral mixture analysis (AC-LSMA) using LSE as a criterion has been studied extensively in the literature. It is, in general, referred to as least-squares AC-LSMA. However, including a weighting matrix in the least-squares AC-LSMA to account for significance of individual bands has not been explored in the past few years until Chang and Ji, (2006a). This chapter investigates weighted AC-LSMA in terms of LSE and further develops three approaches to weighted AC-LSMA, each of which can be obtained by the commonly used criteria, Mahalanobis distance or Gaussian maximum likelihood estimation, Fisher's ratio, and OSP. In particular, the least-squares AC-LSMA can be considered an unweighted AC-LSMA. The experimental results demonstrate that weighted AC-LSMA generally performs better than unweighted AC-LSMA.

Finally, we summarize the advantages and disadvantages of the unweighted AC-LSMA (i.e., FCLS), along with all the four weighted AC-LSMA methods considered in Table 14.6 in this chapter.

Table 14.6 Summary of unweighted AC-LSMA (FCLS) and four weighted AC-LSMA methods.