Underwater hyperspectral imaging-based classification of manganese nodules using a multi-scale spatial-spectral residual network

Accurate classification of manganese nodules is critical for deep-sea mineral exploration and environmental assessments. This study presents a novel multi-scale spectral-spatial residual network (MSSRN) combined with underwater hyperspectral imaging (UHI) for the precise differentiation of deep-sea manganese nodules. The proposed MSSRN leverages joint spectral-spatial information extraction at multiple scales through specially designed spectral-spatial residual blocks. Experiments conducted on a custom-built UHI dataset demonstrate that MSSRN achieves outstanding classification performance, with an overall accuracy (OA) of 99.7%, average accuracy (AA) of 99.7%, and a Kappa coefficient of 99.41%, significantly outperforming established benchmark models such as 2D-CNN, 3D-CNN, ResNet, and HybridSN. These findings underscore the effectiveness of multi-scale spectral-spatial feature extraction for accurately differentiating manganese nodule types using UHI data. By providing a highly accurate method for manganese nodule classification, this research offers promising implications for advancing the deep-sea mineral resource assessment and environmental impact studies.