Improving sparse coding based hyperspectral image classification via tensor decomposition and oversegmentation

Hyperspectral imaging, with its detailed spectral and spatial information, can be applied in various fields such as re-mote sensing, biomedical engineering, and quality control. However, the high dimensionality and large data volume of HS images pose significant challenges for efficient processing and classification. A well-known hyperspectral classification approach is sparse coding. This study proposes an efficient method to enhance the sparse coding-based classification of HS images by integrating two methods, namely the High Dimensional Model Representation and Simple Linear Iterative Clustering. Initially, HDMR was applied to decompose the 3-D HS tensor into manageable components, effectively reducing noise and correlations. Subsequently, SLIC oversegmentation was employed to generate superpixels, facilitating localized feature extraction. The average spectral signal of each superpixel is classified using a sparse coding classifier. Experimental results on public HS datasets - Indian Pines, Pavia University, and Salinas - demonstrate that the proposed HDMR and SLIC integration significantly improves classification accu-racy and reduces computational complexity compared to conventional methods. This approach leveraged the strengths of tensor decomposition and superpixel segmentation to offer a robust and efficient solution for HS image classification.