Single-pixel Hyperspectral Imaging Via An Untrained Deep Neural Network

Correlation imaging,also known as two-photon imaging or ghost imaging(GI),is a method of obtaining object images through the correlation of two beams of light: the object beam and the reference beam.The object beam interacts with the imaging object and is eventually detected by a single-pixel detector without any spatial resolution.In contrast,the reference beam is directly recorded by a detector with spatial resolution capabilities,such as a charge-coupled device(CCD)that does not carry object information.GI was first demonstrated in 1995 using entangled photon pairs,which was considered a quantum phenomenon,but it was soon implemented with classical light sources as well.According to the principle of GI,the reference beam can be removed when it can be numerically evaluated.GI can be achieved in a single-beam configuration and object images can be reconstructed using correlation calculations.This method is known as computational ghost imaging(CGI)or single-pixel imaging(SPI).In general,the main research content of this paper is as follows:1.Image enhancement techniques have been applied in many fields so far,but their wider practical applications are still limited by imaging speed and quality.Various GI algorithms have been proposed to improve imaging efficiency and quality,such as differential ghost imaging,normalized ghost imaging,and compressive sensing ghost imaging(CSGI).In recent years,deep learning has been widely used to solve various inverse problems in computational imaging.In these deep learning cases,thousands or even tens of thousands of labeled data are usually required to train neural networks.In the imaging process,this would require a significant amount of time to acquire sufficient labels.This paper integrates the closely related physical processes of single-pixel hyperspectral imaging(HSI)experiments into the classic U-Net neural network.Based on untrained deep neural networks(DNN),single-pixel hyperspectral imaging experiments were conducted at ultra-low sampling rates,and high-quality inverted images of the target objects were obtained for each visible light wavelength from 432 nm to 680 nm.Specifically,the imaging physical model of single-pixel HSI was integrated into a randomly initialized DNN,allowing image reconstruction relying solely on the interaction between the imaging physical process and the neural network,without the need for pre-training the neural network.2.This paper designs a compressive sensing-based spectral super-resolution single-pixel imaging method.The method is based on a single-pixel imaging system illuminated by a broadband light source and is capable of achieving super-resolution imaging of spatial targets using spectral dimension information in conditions where the system’s spatial imaging resolution is insufficient.According to the position of the spectrometer in the system,i.e.,pre-imaging spectrometer and post-imaging spectrometer,the obtained single-wavelength reconstructed images are fused through a wavelet multiresolution decomposition algorithm to achieve super-resolution fusion reconstruction of color targets.When combined with other spatial super-resolution imaging methods,this approach can find wide applications in high-resolution imaging fields such as microscopy and remote sensing.3.In recent years,several optical system architectures have been proposed for information security,storage,and decryption,with a particular focus on optical processing systems using random phase encoding schemes.Optical encryption has the ability to encrypt data in multiple dimensions,such as phase,wavelength,spatial frequency,or polarization,making it difficult for unauthorized individuals to access protected information.This paper introduces the discrete wavelet transform(DWT)and double-random-phase-encoding(DRPE)encryption techniques and presents an optical encryption method based on double-random phase encoding.Users with the key can effectively recover the target image,and this method exhibits high encryption performance,passing tests for external noise and screenshot attacks.The design of dual keys greatly enhances the reliability of this method.With the gradual upgrade of optoelectronic devices in the future.