Comparison of a new classifier, the Mix-Unmix Classifier, with conventional hard and soft classifiers

摘要

'The number of bands must be more than the number of end-members...' is perhaps the most ubiquitous statement in linear spectral unmixing. The Mix-Unmix Classifier overcomes this limitation. Further, the classifier creates a processing environment that allows any pixel to be unmixed without any sort of restrictions (e.g. minimum determinable fraction), impracticalities (e.g. negative fractions), or trade-offs (e.g. either positivity or unity sum). The classifier gives not only the most probable fractions of end-members, but also their most probable contributory DNs. The contributory DNs directly define the qualities, (e.g. the phenological stages) of the end-members. The classifier is applied as a dual classification method and compared with popular conventional hard and soft classifiers in production of two to eight spectral classes/end-members from Landsat 7 ETM+ data. The classifiers considered are Spectral Angle Mapper, Binary Encoding Classifier, and Maximum Likelihood Classifier for hard classification; and IDRISI Kilimanjaro Probability Guided Option linear unmixing technique for soft classification. The Mix-Unmix Classifier performs better than the others.