Hyperspectral imaging method based on unsupervised network dedispersion blurring

Abstract

The invention discloses a hyperspectral imaging method based on unsupervised network dedispersive blurring. The method comprises the following steps: S1, acquiring an RGB image with fuzzy dispersion; S2, creating an unsupervised learning convolutional neural network, wherein the input of the network is a dispersion RGB image, and the output of the network is a reconstructed hyperspectral image; S3, inputting the single dispersion RGB image acquired in the step S1 into the convolutional neural network in the step S2, and reconstructing spectral information of the image by using an online training method; S4, driving parameter optimization of the convolutional neural network according to the physical relationship of the dispersion image generated by the imaging system, and training the ability of the convolutional neural network to reconstruct hyperspectral data from the dispersion RGB image by using a back propagation algorithm; and S5, repeatedly iterating for multiple times to obtain a reconstruction result gradually approaching the real hyperspectral image. According to the hyperspectral imaging method, the unsupervised network is used for dedispersive ambiguity, model driving is used for replacing data driving, the reconstruction precision is guaranteed, meanwhile, the system is simpler, and the cost is reduced.

Description

technical field [0001] The invention belongs to the field of spectral imaging, in particular to a hyperspectral imaging method based on an unsupervised network to resolve dispersion blur. Background technique [0002] Hyperspectral imaging technology has important applications in various fields. Compared with RGB images, hyperspectral images retain more color information. However, traditional spectrometers and imaging devices have disadvantages such as high cost, large volume, and complex systems. With the development of computer science, software algorithms have been introduced into spectral imaging technology, making spectral imaging instruments cheaper and more compact, mainly including compressed sensing and computational reconstruction algorithms. In compressive sensing reconstruction, high-dimensional spectral data can be solved by optimization problems with sparse priors, but artificially designed priors cannot find effective and robust priors, resulting in low accur...

AI Technical Summary

Problems solved by technology

However, traditional deep learning-based computational imaging methods mostly use supervised learning, which requires pre-obtaining a large amount of labeled data or real data to train the network. The quality and quantity of the data set can directly affect the effect of network training, and data-driven obtained The generalization ability of the model is limited, and it is impossible to achieve good reconstruction accuracy for all data

Although there are some unsupervised learning methods, they still use data-driven acquisition models, which take a long time in model training, and the prediction accuracy of the obtained models cannot reach the effect of supervised learning

Images