CN-121994743-A - Multi-wavelength single-pixel imaging method and system based on graphene detector

CN121994743ACN 121994743 ACN121994743 ACN 121994743ACN-121994743-A

Abstract

The invention relates to the technical field of optical imaging, in particular to a multi-wavelength single-pixel imaging method and a system based on a graphene detector, wherein the method comprises the steps of constructing a binary mask pattern; the method comprises the steps of obtaining a single-pixel measured value of image coding by utilizing an mth binary mask pattern, selecting a training set, coding the image of the training set to obtain a K pair of mark data pairs, constructing a loss function, enabling a neural network to learn from statistical information of the K pair of mark data pairs to map the single-pixel measured value lower than a set sampling rate to the image in the training set, obtaining an optimized neural network mapping function model and the binary mask pattern through minimizing the loss function, and enabling an imaging system to comprise a visible light Shan Xiangsu imaging light path and a terahertz single-pixel imaging light path. The invention can be applied to single-pixel imaging of visible light and terahertz wave bands, and can improve the imaging quality and the imaging speed of single-pixel imaging.

Inventors

CHEN HUANJUN
HU JING
YU TIANXIANG
WEI QUANFENG
DENG SHAOZHI

Assignees

中山大学

Dates

Publication Date: 20260508
Application Date: 20260112

Claims (10)

1. The multi-wavelength single-pixel imaging method based on the graphene detector is characterized by comprising the following steps of: S1, constructing M binary mask patterns, wherein P= { P m |m=1, 2,3,. M } represents M binary mask patterns formed by elements 0 and 1; S2, coding an image O with pixel resolution of N multiplied by N by using an mth binary mask pattern P m to obtain a single-pixel measured value for coding the image O; s3, selecting an image data set as a training set for training a neural network, wherein the training set with the number of images being K is defined as { I k=1, 2,3,..k }, wherein, Representing an image with the resolution of NxN of the kth pixel in the training set; S4, obtaining K pairs of mark data pairs omega T = {(I k of the training set according to the coding process of the step S2 by the images of the training set in the step S3, ) | k = 1, 2, 3, ..., K}; S5, constructing a loss function through root mean square error of an output image and an input image of a neural network, wherein the neural network can learn from statistical information of K pairs of mark data pairs omega T to map single-pixel measured values lower than a set sampling rate to And obtaining an optimized neural network mapping function model and an optimized binary mask pattern by minimizing the loss function.
2. The graphene detector-based multi-wavelength single-pixel imaging method according to claim 1, wherein in step S1: in the formula, An ith row and jth column element representing an mth binary mask pattern P m ; Trainable weights representing ith row and jth column of mth binary mask pattern, N representing row element or column element, M representing number of binary masks, defining sampling rate SR=M- 。
3. The graphene detector-based multi-wavelength single-pixel imaging method according to claim 2, wherein in step S2: in the formula, I m represents a single-pixel measurement value encoding the image O, thus, for encoding the image O using M binary mask patterns p= { P m |m=1, 2,3,..m } resulting in a single-pixel measurement value i= { I m |m=1, 2,3,..m } of length M; In step S4: wherein I k represents the kth image Single pixel measurement encoded by M binary mask patterns P.
4. The graphene-detector-based multi-wavelength single-pixel imaging method according to claim 3, wherein in step S5, the optimized neural network mapping function model and the optimized binary mask pattern obtained by minimizing the loss function are expressed as: in the formula, Representation is made of a set of weights and bias parameters With reference to the defined neural network mapping function model, Representing the optimized neural network mapping function model, Representing the optimized binary mask pattern.
5. An imaging system for application in the graphene detector-based multi-wavelength single-pixel imaging method of any one of claims 1 to 4, comprising a visible light Shan Xiangsu imaging optical path, a terahertz single-pixel imaging optical path, wherein: The visible light Shan Xiangsu imaging light path comprises a visible light source (100), a beam expander (200), a digital micro-mirror device (300), a projection lens group, a terahertz modulator (510), an off-axis parabolic lens group and a graphene detector (700) which are sequentially arranged, and further comprises a single-pixel imaging device which is in communication connection with the graphene detector (700) and the digital micro-mirror device (300), wherein an optimized binary mask pattern is used for being input into the digital micro-mirror device (300), and an optimized neural network mapping function model is used for being called by the single-pixel imaging device; In the visible light Shan Xiangsu imaging light path, the visible light source (100) emits a light beam, the light beam propagates to the digital micro-mirror device (300) after being expanded by the beam expander (200) to form structural visible light, the structural visible light is projected to the surface of the terahertz modulator (510) by the projection lens group and then reflected to the off-axis parabolic lens group, then the structural visible light is subjected to the off-axis parabolic lens group and a to-be-imaged object in the off-axis parabolic lens group to obtain visible light carrying object information and converged to the graphene detector (700), the visible light is transmitted to the single-pixel imaging device by the graphene detector (700), and the visible light Shan Xiangsu is imaged at the single-pixel imaging device by the optimized neural network mapping function model; The terahertz single-pixel imaging light path can be used for terahertz single-pixel imaging at the single-pixel imaging device through an optimized neural network mapping function model.
6. The imaging system of claim 5, wherein the terahertz single-pixel imaging light path comprises a terahertz source (800) arranged at the terahertz modulator (510), the terahertz single-pixel imaging light path further comprises an off-axis parabolic mirror group, a graphene detector (700) and a single-pixel imaging device, wherein the terahertz source (800) emits a light beam, the light beam passes through the terahertz modulator (510) and then propagates to the off-axis parabolic mirror group, and then the light beam passes through the off-axis parabolic mirror group and a terahertz beam which is positioned in the off-axis parabolic mirror group and is to be imaged into an object to be imaged, obtains the terahertz beam carrying object information, converges to the graphene detector (700), and is transmitted to the single-pixel imaging device through the graphene detector (700), and the terahertz single-pixel imaging is performed at the single-pixel imaging device through an optimized neural network mapping function model.
7. The imaging system of claim 5, wherein the set of projection lenses comprises a first projection lens (410), a mirror (420), and a second projection lens (430) arranged in sequence, wherein the set of off-axis parabolic lenses comprises a first off-axis parabolic mirror (610) and a second off-axis parabolic mirror (620), wherein the visible light Shan Xiangsu imaging path further comprises a first chopper (520) disposed between the visible light source (100) and the beam expander (200), and wherein the first chopper (520) is communicatively coupled to the single-pixel imaging device.
8. The imaging system of claim 5, wherein the single pixel imaging device comprises a computer (910), a data acquisition member (920), a lock-in amplifier (930), and a current amplifier (940) electrically connected in sequence, wherein the current amplifier (940) is electrically connected to the graphene detector (700), wherein the digital micromirror device (300) is electrically connected to the data acquisition member (920), and wherein the optimized neural network mapping function model is stored in the computer (910) and is capable of being invoked.
9. The imaging system of claim 5, wherein the method of preparing the graphene detector (700) comprises the steps of: Transferring the graphene film to a substrate layer (710); coating photoresist on a substrate sample with a graphene film, performing maskless patterning exposure and development; performing plasma etching on the developed sample, and removing the photoresist to obtain a patterned graphene layer (720); coating photoresist on a substrate sample with a patterned graphene layer (720), performing maskless alignment and development, and completing positioning photoetching patterning of an electrode area on the sample; Sequentially depositing a Ti film and an Au film in the electrode area subjected to positioning photoetching patterning, and then stripping the films outside the electrode area to obtain a first electrode (730) and a second electrode (740); And carrying out annealing treatment on the samples obtained by the first electrode (730) and the second electrode (740) to obtain the graphene detector (700).
10. The imaging system of claim 5 or 9, wherein the graphene detector (700) comprises a substrate layer (710), and a patterned graphene layer (720) arranged on the substrate layer (710), wherein both ends of the patterned graphene layer (720) are plated with a first electrode (730) and a second electrode (740), and the first electrode (730) and the second electrode (740) are Ti/Au electrodes.

Description

Multi-wavelength single-pixel imaging method and system based on graphene detector Technical Field The invention relates to the technical field of optical imaging, in particular to a multi-wavelength single-pixel imaging method and system based on a graphene detector. Background In the region of the electromagnetic spectrum from the visible to the terahertz band, various types of detectors are designed and prepared based on the physical properties of materials in different wavelength ranges. For the visible light band, the mainstream detection technology is mainly based on silicon-based semiconductor photodetectors, such as Complementary Metal Oxide Semiconductor (CMOS) and Charge Coupled Devices (CCD), which exhibit excellent photoelectric conversion efficiency and quantum efficiency in a specific wavelength range, but the spectral response range is limited by the intrinsic band gap characteristics of the material, and it is difficult to realize wide-band detection. The terahertz wave band (0.1-10 THz) has obvious application prospect in the fields of biomedical imaging, safety detection, nondestructive detection and the like due to the non-ionizing radiation characteristic and the good penetrating capability of the terahertz wave band to most of nonpolar media. Currently, terahertz detection mainly depends on a heat detector, a photon detector and an electronic device detector, but the schemes of the detectors have technical bottlenecks that the heat detector generally shows the characteristics of long response time and low sensitivity under the room temperature condition, the real-time imaging capability is restricted, the working frequencies of the photon detector and the electronic detector are generally limited to below 1 THz, and high-spatial resolution imaging is difficult to realize due to diffraction limit effect in a high-frequency band. Graphene is used as a zero-band-gap two-dimensional carbon material, has ultrahigh carrier mobility and unique electrical and optical characteristics, and provides a novel material platform for developing broadband response and high-speed photoelectric detectors. However, the prior art still has the problems that the prior single-pixel imaging system is difficult to realize single-pixel imaging of visible light and terahertz waves at the same time, has obvious limitations in terms of imaging quality and sampling efficiency, and cannot meet the application requirements of high-spatial-resolution imaging. Disclosure of Invention The invention aims to overcome the defects of limitation of imaging quality and imaging speed of single-pixel imaging of visible light and terahertz waves in the prior art, and provides a multi-wavelength single-pixel imaging method and system based on a graphene detector, which can be applied to single-pixel imaging of visible light and terahertz wave bands, and can improve the imaging quality and imaging speed of single-pixel imaging. In order to solve the technical problems, the invention adopts the following technical scheme: The multi-wavelength single-pixel imaging method based on the graphene detector comprises the following steps: S1, constructing M binary mask patterns, wherein P= { P m |m=1, 2, 3,. M } represents M binary mask patterns formed by elements 0 and 1; S2, coding an image O with pixel resolution of N multiplied by N by using an mth binary mask pattern P m to obtain a single-pixel measured value for coding the image O; s3, selecting an image data set as a training set for training a neural network, wherein the training set with the number of images being K is defined as { I k=1, 2, 3,..k }, wherein,Representing an image with the resolution of NxN of the kth pixel in the training set; S4, obtaining K pairs of mark data pairs omega T= {(Ik of the training set according to the coding process of the step S2 by the images of the training set in the step S3, ) |k = 1, 2, 3, ...,K}; S5, constructing a loss function through root mean square error of an output image and an input image of a neural network, wherein the neural network can learn from statistical information of K pairs of mark data pairs omega T to map single-pixel measured values lower than a set sampling rate toAnd obtaining an optimized neural network mapping function model and an optimized binary mask pattern by minimizing the loss function. The multi-wavelength single-pixel imaging method based on the graphene detector can obtain high-efficiency binary mask patterns from the neural network through optimization learning, can obtain more image information in a small amount of mask patterns, further meets the high-quality imaging requirement under the low sampling rate, has excellent generalization performance, and can be applied to single-pixel imaging of visible light and terahertz wave bands. Further, in step S1: in the formula, An ith row and jth column element representing an mth binary mask pattern P m; Trainable weights representing the ith row and jth column