Multispectral single-pixel imaging deep learning image reconstruction method

Abstract

The invention provides a multispectral single-pixel imaging deep learning image reconstruction method, which comprises a measurement process and a reconstruction process, and is characterized in thatin the measurement process is characterized by using a coding pattern for coding a target scene, and then using a multispectral detector for recording the light intensities corresponding to differentwavelengths; after the multispectral single-pixel detection is realized in a physical mode, based on an image reconstruction method of a deep neural network, realizing a process of reconstructing anoriginal signal X from all detection signals Yc, wherein the deep neural network is composed of a linear mapping network and a convolutional neural network; splicing C measurement vectors together according to columns to form a new matrix Y ', wherein the linear mapping network uses Y' as the input data to preliminarily perform linear processing on the data, then subjecting the linear processing result to the information fusion processing between channels through the convolutional neural network, and finally obtaining a to-be-observed image X through reconstruction. According to the technicalscheme, the problems that in the prior art, the algorithm complexity is high, the reconstruction time is longer, and the requirement for the sampling rate is higher are solved.

Description

technical field [0001] The present invention relates to the fields of single-pixel imaging technology and deep learning technology, in particular, to a deep learning image reconstruction method for multi-spectral single-pixel imaging. Background technique [0002] Multi-spectral imaging technology is a new generation of photoelectric detection technology, which emerged in the 1980s, formed a research and development boom after the 1990s, and is still developing rapidly. Multispectral imaging technology is different from traditional single broadband imaging technology, but combines imaging technology and spectral measurement technology. The information obtained includes not only two-dimensional spatial information, but also spectral radiation information distributed with wavelength. The rich spectral information of the target combined with the space image of the target greatly improves the accuracy of target detection and expands the functions of traditional detection technol...

AI Technical Summary

Problems solved by technology

Although a lot of research has been done on compressed sensing reconstruction algorithms, the main problems of image reconstruction based on compressed sensing are the high complexity of the algorithm, which requires a large number of iterative operations, the reconstruction time is long, and the sampling rate is high.

[0005] For multi-spectral single-pixel imaging, the project team proposed a joint multi-dimensional observation vector reconstruction algorithm in the "Color single-pixel imaging based on multiple measurement vectors model" published in the Journal of Optical Engineering in 2016. band structure characteristics to improve the imaging quality, but it is essentially an image reconstruction algorithm based on compressed sensing, and the reconstruction quality still needs to be improved

Images