A Lightweight Spatiotemporal Separable Convolution Network for HSI Classification

HSI classification is crucial but challenged by high dimensionality. 3D-CNNs perform well but face high computational cost, limited generalization, and overfitting. This study proposes an STSC framework that decomposes 3D convolutions into simpler operations and reduce computational overhead related to classification. Experiments on public HSI datasets demonstrate that STSC achieves competitive or superior accuracy compared to state-of-the-art methods.