Classification Percentage Enhancement of Segmentation Indexed Image based on Clustering Algorithm

This paper indicate that polarization channel based feature grouping followed by a multi-classifier decision fusion framework performs markedly better than traditional SLDA and LDA based dimensionality reduction, followed by a single-classifier. Results illustrates the classification maps generated using the proposed MCDF framework as well as conventional dimensionality reduction and single-classifier approaches. The benefit of using decision fusion is apparent from the reduced salt-and-pepper misclassifications with the MCDF approach. For many applications, such as the Levee health monitoring problem used in this letter, collecting ground-reference data is very expensive. This robustness to a very small training sample size is hence highly desired.