Comparative analysis of representation-based classification methods for detecting mucilage using PRISMA data

In this thesis, representation-based classification methods were investigated on hyperspectral satellite data to detect mucilage formations in the Sea of Marmara through remote sensing. For this purpose, hyperspectral images acquired by the PRISMA satellite, developed by the Italian Space Agency (ASI), from the coastal regions of Istanbul and Bursa were used. These high spectral resolution data enable spectral differentiation of mucilage from water surfaces. During the classification stage, representation-based, collaborative representation-based, and subspace-based classification methods that can use spectral, spatial, and spectral-spatial information together were analyzed comparatively. The performance of the methods was evaluated using metrics such as overall accuracy, average accuracy, and F-score, and the resulting classification maps were also analyzed visually. Experimental findings demonstrate that methods based on NSC and multi-scale super-pixel segmentation exhibit superior performance in terms of overall accuracy and class balance. The results obtained from real satellite data not only go beyond the standard data sets commonly used in the literature, but also strongly demonstrate the applicability of these methods in real-world problems.