Hyperspectral Image Super-Resolution Combined With Spatial-spectral Information

The hyperspectral image is a three-dimensional data structure that contains both spatial and spectral information.By recording scene information of different bands,it can describe the properties of objects more comprehensively.However,due to the limitations of imaging principles and hardware performance,the spatial resolution of hyperspectral images is usually low,which contributes to the classification of ground objects limited.Therefore,it is of great significance to study how to improve the spatial resolution of hyperspectral images.This article studies the super-resolution method for hyperspectral images,focusing on low spectral information utilization and spectral information distortion in reconstruction,and design corresponding solutions.Aiming at the problem of low utilization of spectral features in traditional networks,this thesis proposes a 3D channel-weighted residual network for image super-resolution which extracts spatial and spectral features with 3D convolution and combines the attention mechanism to perform weighting constraints on channels to improve the expression ability of network to spatial spectral features.To enhance the spatial and spectral details of the image,the residual feature aggregation framework is used in the network which highlights the residual characteristics of the spatial details and spectrum details and improves the effect of image detail reconstruction.A dual-channel image super-resolution network based on multi-feature fusion is proposed to facing the requirement of hyperspectral image and high spatial resolution multispectral image joint super-resolution reconstruction task.Aiming at the problem of weak learning ability of traditional network to shallow features and residual features,a multi-level feature fusion model is designed,and the feature layer weighting restriction mechanism is used to improve the expression ability of features from different images or different levels of features;Aiming at the problem of spectral information distortion,the space-spectrum joint constraint loss function is designed,which comprehensively considers the errors in the spatial domain and the spectral domain to improve the spectral information restoration ability of the network.The model proposed in the thesis was compared with the existing super-resolution algorithm on the public hyperspectral image dataset,and the results of the reconstruction of the two models were tested in the ground object classification experiment.The experiment showed that the reconstruction algorithm proposed by this thesis is superior to other methods in terms of visual effects and objective indicators,and it has positive significance for improving the accuracy of ground object classification.