Hyperspectral Data Processing: Algorithm Design and Analysis

2.3 Mixed Sample Analysis

In analogy with subsample analysis presented in Section 2.2, mixed sample analysis can be performed similarly with one key difference. In subsample analysis the background knowledge about the b considered in (2.13) is assumed to be unknown or can be obtained a posteriori directly from the data or a secondary data set, whereas in mixed sample analysis the target knowledge present in the r must be completely specified a priori. One fundamental task of mixed sample analysis is to perform spectral unmixing via linear spectral mixture analysis (LSMA). More specifically, LSMA assumes that a data sample vector r can be specified by a set of p known signals, as a linear mixture in terms of their respective mixing abundance fractions specified by . When LSMA is used for classification, it classifies the r into one of these p signals, . Using the five-fruit mixed juice described in the introduction as an example, the classification of mixed sample analysis is performed by assuming that the five fruits (apple, banana, lemon, orange, and strawberry) represent five known signals and the mixed juice r is an admixture of these five fruits with different concentrations