An endmember optimization approach for linear spectral unmixing of fine-scale urban imagery

Spectral unmixing of high spatial resolution imagery has attracted growing interest for interpreting urban surface material characteristics. This study proposes an endmember optimization method based on endmember spatial distribution (i.e. solid angle and tetrahedron volume) to select the optimal endmember combination for urban spectral unmixing. Specifically, a linear spectral unmixing model (SESMA) is implemented in a suitable 3-D spectral space structured by the green, red and near infrared bands of the imagery, and endmember spatial distribution is measured with solid angle and tetrahedron volume. Both the solid angle and tetrahedron volume are found to have a strong linear or logarithmic relationship with valid and correct unmixed proportions, whereas the latter measure also takes the photometric shade into account as an endmember. The spectral unmixing results based on the proposed endmember optimization method are compared with those from a common multiple endmember spectral mixture analysis (MESMA) model. Towards different classes, each model has its own advantages over the other.