LAUSR

Convolutional neural networks (CNNs) have been developed as an effective strategy for hyperspectral image (HSI) classification. However, the lack of feature extraction by CNN networks is due to the network failing to effectively extract global features and poor capability in distinguishing between different feature categories that are similar. In order to solve these problems, this paper proposes a novel approach to hyperspectral image classification using a multidimensional spectral transformer with channel-wise correlation. The proposed method consists of two key components: an input mask and a channel correlation block. The input mask is used to extract relevant spectral information from hyperspectral images and discard irrelevant information, reducing the dimensionality of the input data and improving classification accuracy. The channel correlation block captures the correlations between different spectral channels and is integrated into the transformer network to improve the model’s discrimination power. The experimental results demonstrate that the proposed method achieves great performance with several benchmark hyperspectral image datasets. The input mask and channel correlation block effectively improve classification accuracy and reduce computational complexity.