Research on Hyperspectral Image Reconstruction Based on Spatial Attention and Channel Attention

Abstract

Hyperspectral images contain more information than RGB images and theoretically have a wider range of applications. However, they are limited by the high cost of hyperspectral imaging equipment, complex data processing, and other issues. Currently, they are mainly used in remote sensing, military applications, astronomy, and other specific fields. In recent years, scholars have proposed using mathematical methods to reconstruct hyperspectral images from RGB images, which can significantly expand the application scope of spectral imaging. Hyperspectral image reconstruction is enhanced by incorporating spatial attention and channel attention. This addresses the challenge of capturing global context features and addressing the lack of correlation information within channels. The addition of a spatial attention model and a channel attention model to the hollow convolution residuals results in the calculation of a shared attention weight vector for all spatial locations with minimal parameters, establishing dependency between different channels. Experimental results demonstrate that both the spatial attention model and the channel attention model effectively enhance feature extraction ability, leading to significant reduction in reconstruction error when used together.