Hyperspectral image reconstruction method based on neural network

Abstract

The invention discloses a hyperspectral image reconstruction method based on a neural network. Problems of low stability and precision of hyperspectral images reconstructed in the prior art are mainly solved. The method is characterized in that training and fitting are performed on a nonlinear mapping relation between a color image and a hyperspectral image by use of the neural network. The method comprises steps of: 1) acquiring and taking a hyperspectral image set and a color image set as training samples of the neural network; 2) constructing a neural network model and using the training samples to train parameters of the neural network; and 3) using any given new color image as a test sample, and inputting the test sample into a trained neural network model, wherein the output result of the neural network is the reconstructed hyperspectral image. According to the invention, computing complexity of the reconstruction of the hyperspectral image is reduced; stability and precision of the reconstructed hyperspectral image are improved; and the method is applicable to spectrum detection, geological exploration, environment monitoring and agricultural remote sensing.

Description

technical field [0001] The invention relates to the technical field of image processing, in particular to a hyperspectral image reconstruction method, which can be used for spectral detection, geological survey, environmental monitoring and agricultural remote sensing. Background technique [0002] Hyperspectral images not only contain the spatial information of the observed target, but also each pixel in the image has dozens or even hundreds of narrow bands of rich spectral information, which has the property of "integration of graphs and spectra". Since hyperspectral images can maintain the spectral characteristics reflecting the properties of matter and the image information presenting the geometric space information of matter, it has greatly improved the ability of human beings to recognize the objective world. It has been proved to have great application value in many fields. [0003] Using hyperspectral imaging equipment to obtain precise hyperspectral images is very ...

AI Technical Summary

Problems solved by technology

The spectral reconstruction problem is an ill-conditioned inverse problem, which uses the sparsity of the spectrum and combines the spatial information of the image to reconstruct the hyperspectral image, but does not make full use of the strong local and non-local spatial structure similarity in the spectral image , making the stability and accuracy of the reconstructed hyperspectral image difficult to meet the requirements of scientific research and large-scale practical applications

Images