SSRHCN: hybrid convolution-based spectral-spatial feature reconstruction for hyperspectral image classification

There is rich spectral-spatial information in hyperspectral images (HSIs), and how to extract useful spectral-spatial information from massive spectral data has always been a research hotspot in the field of HSI processing. With the development of deep-learning technology, Convolutional neural network (CNN) has become one of the most commonly used deep-learning HSI Classification methods. However, the high-dimensional nature of hyperspectral data and band correlations create redundancies that increase the computational effort and affect the extraction of spectral and spatial features, which are crucial for classification accuracy. To overcome these limitations, this paper proposes a new hybrid convolution model that utilizes spectral-spatial channel reconstruction. By using hybrid convolution to extract and reconstruct spectral-spatial features separately, redundant information is suppressed and the computational load of the model is reduced. In addition, the model integrates a global attention mechanism to enhance the model's receptive field and facilitate learning of global spectral-spatial features. The model classification experiments on the Pavia University and Houston 2013 datasets showed that the overall accuracy of the proposed model reached 99.49% and 99.07%, verifying the effectiveness of the hybrid convolutional model in HSI classification.