Spatial-Spectral Analysis of Hyperspectral Imagery with Multilevel Thresholding and Multi-OTSU Segmentation

This study examines the use of Multilevel Thresholding and Multi-OTSU segmentation algorithms on a variety of Hyperspectral datasets, including the benchmarks Salinas, Pavia University, and Indian Pines, using accuracy measures like Mean Squared Error (MSE), Root Mean Squared Error (RMSE), and Peak Signal-to-Noise Ratio (PSNR). We compare the effectiveness of these fast yet extremely efficient thresholding-based segmentation techniques for hyperspectral imagery across several datasets and present improvements to both techniques. Indian Pines has the highest average PSNR values, 74.82 and 74.16 for Multi-OTSU and Multilevel Thresholding, respectively, while Pavia University shows the lowest average MSE values, 3.36 for Multi-OTSU and 19.28 for Multilevel. Furthermore, an analysis of segmented bands shows that, for all three datasets, Multilevel Thresholding outperforms Multi-OTSU segmentation in terms of speed and segmentation quality. These results provide insights into algorithmic performance differences under various dataset situations and further the field of hyperspectral image segmentation approaches.