Spectral super-resolution adaptive weighted attention mechanism deep network data processing method

Abstract

The invention belongs to the technical field of hyperspectral image processing, and discloses a spectral super-resolution adaptive weighted attention mechanism deep network data processing method, which comprises the steps of selecting two spectral reconstruction challenge data sets for verification; evaluation index selection: using a root mean square error RMSE and an average relative absolute value error MRAE as evaluation indexes, and then calculating the MRAE and the RMSE; constructing an adaptive weighted channel attention mechanism module; constructing a second-order non-local module based on partitioning; and constructing an adaptive weighted attention mechanism network and training. According to the invention, a brand-new adaptive weighted attention mechanism network is designed for spectral super-resolution, and a backbone network is formed by stacking a plurality of double-residual modules which realize double-residual learning through long and short connection. And a self-adaptive weighted channel attention module is embedded in the double-residual module, so that the channel characteristic response is recalibrated, and the characteristic expression capability of the network is enhanced.

Description

technical field [0001] The invention belongs to the technical field of hyperspectral image processing, and in particular relates to a deep network data processing method of spectral super-resolution adaptive weighted attention mechanism. Background technique [0002] Currently, hyperspectral imaging records the reflectance or transmittance of objects, and the acquired hyperspectral images usually have a variety of spectral bands from the infrared spectrum to the ultraviolet spectrum. Rich spectral features have been widely explored for various tasks, such as face recognition, image classification, and anomaly detection. However, due to the limitations of imaging techniques, the process of capturing these hyperspectral images with high spatiotemporal resolution and rich spectral information is time-consuming, thus hindering the application range of hyperspectral imaging. In recent years, one of the approaches to solve this problem is to develop scanless or snapshot hyperspec...

AI Technical Summary

Problems solved by technology

Most CNN-based super-resolution methods focus on designing deeper or wider network architectures to obtain higher-level feature representations, lacking in extracting rich contextual information and exploring correlations between intermediate features, thus limiting CNN-based The performance of the super-resolution method

In addition, the existing CNN-based super-resolution models always complete a complex RGB-to-hyperspectral mapping function, and seldom consider integrating the camera response curve a priori into spectral super-resolution for more accurate reconstruction. Currently, most of them are based on CNN's super-resolution algorithms all ignore the prior information of the camera response curve, thus limiting the reconstruction performance of the spectral super-resolution algorithm

[0003] Through the above analysis, the existing problems and defects of existing technologies are: most CNN-based super-resolution methods are devoted to designing deeper or wider network architectures to obtain higher-level feature expressions, and lack the ability to extract rich contextual information and explore intermediate The correlation between features, while ignoring the prior information of the camera response curve, limits the reconstruction performance of super-resolution algorithms

Images