A single-pixel edge imaging system and method based on quantum state encoding

Abstract

This invention discloses a single-pixel edge imaging system and method based on quantum state encoding. The system includes modules for illumination pattern generation, single-pixel detection and data acquisition, and edge image reconstruction. The core scheme is as follows: the illumination pattern generation module generates a spatiotemporally coupled light field, the spatial distribution of which is a convolution kernel of the edge detection operator, and the temporal dimension follows an orthogonal eigenmode sequence generated by the Sturm-Liouville operator. This light field interacts with the object to achieve parallel encoding of edge information. The detection module acquires temporal bucket signals, and the reconstruction module directly demodulates and images based on orthogonality. This invention combines quantum state orthogonality with temporal frequency division multiplexing, effectively suppressing noise under strong scattering, significantly reducing data redundancy, and achieving direct edge imaging with high signal-to-noise ratio and low latency, breaking through the dependence on spatial array sampling.

Description

Technical Field

[0001] This invention relates to the fields of computational optical imaging and machine vision technology, specifically to an imaging system and method that utilizes a quantum state light field to load a spatial differential operator, enabling direct physical extraction of target edge features without reconstructing the full image. Background Technology

[0002] Edge detection is a fundamental and crucial step in fields such as object recognition, autonomous driving, medical diagnostics, and industrial inspection. Traditional edge detection typically relies on an "image-first, process-later" approach, where a high-resolution array camera first acquires a complete intensity image of the scene, and then gradient operators are used in the digital domain for differentiation to extract edges. However, this traditional approach faces significant challenges in specific application scenarios. First, full-frame imaging acquires a large amount of redundant background data, which results in significant bandwidth waste and storage pressure for applications that only focus on contour features (such as foreign object detection on high-speed conveyor belts). Second, when imaging through strong scattering media such as smoke, biological tissue, or frosted glass, the ballistic light component of photons is drastically reduced, resulting in blurred images with extremely low signal-to-noise ratios from traditional area array cameras. In such cases, directly performing digital differentiation on the noisy image significantly amplifies the noise, causing edge extraction to completely fail. In addition, although traditional single-pixel imaging (SPI) has the advantage of high detection sensitivity by using spatial light modulators to encode the light field, existing SPI technologies based on Hadamard or Fourier basis are limited by the Heisenberg-Gabor inequality, which prevents them from achieving optimal localization in the time and frequency domains at the same time. This results in sidelobe crosstalk in the reconstructed image and insufficient edge sharpness at low sampling rates.

[0003] In recent years, a quantum state-encoded patterned illumination technique has emerged in the field of computational imaging. This technique utilizes quantum eigenstate wavefunctions as the encoding basis, achieving strict orthogonality in the time domain. On the other hand, methods that use the principle of optical convolution to directly load spatial differential operators into the illumination field to physically extract edges are also maturing. Current technologies have not yet effectively integrated the high fidelity and anti-scattering properties of quantum state encoding with the edge computing capabilities of spatial differential operators, and lack a real-time computing architecture based on dedicated hardware. Therefore, how to construct a system that combines anti-scattering capabilities, high signal-to-noise ratio, and real-time edge image output is a pressing technical problem to be solved in this field. Summary of the Invention

[0004] One objective of this invention is to propose a single-pixel edge imaging system and method based on quantum state encoding, which can directly realize the physical calculation and extraction of target edges under imaging-free conditions.

[0005] This invention is achieved through the following technical solution: a single-pixel edge imaging system based on quantum state modulation and edge detection operator convolution, comprising an illumination pattern generation module, an object to be measured, a single-pixel detection and data acquisition module, and an edge image reconstruction module;

[0006] The illumination pattern generation module is used to generate a spatiotemporally coupled light field. The spatiotemporally coupled light field follows an orthogonal sequence of quantum eigenstate wave functions in the time dimension and exhibits a convolution kernel distribution of edge detection operators in the spatial dimension. The illumination pattern generation module projects the spatiotemporally coupled light field onto the object under test or modulates the light field from the object under test.

[0007] The object under test is located between the lighting pattern generation module and the single pixel detection and data acquisition module. It physically interacts with the spatiotemporally coupled light field through reflection, transmission or self-radiation to achieve physical convolution between the edge operator and the spatial information of the object.

[0008] The single-pixel detection and data acquisition module is used to collect the total light intensity modulated by the object under test, convert the light signal into an analog electrical signal through single-point detection, perform analog-to-digital conversion acquisition, and output a time-domain bucket signal sequence containing edge feature information.

[0009] The edge image reconstruction module is used to receive a time-domain bucket signal sequence, demodulate the signal using the orthogonality of quantum eigenstate wave functions, separate the edge response values ​​of each spatial location from the bucket signal, and directly reconstruct the edge image of the object under test.

[0010] The lighting pattern generation module adopts an active projection structure or a passive imaging structure;

[0011] Active projection structure: includes a light source and a spatial light modulator, or a digital projector; used to actively project a spatiotemporally coupled light field loaded with an edge detection operator pattern onto the object under test;

[0012] Passive imaging structure: includes an imaging lens and a spatial light modulator; light emitted by the object under test or reflected light from the environment is imaged onto the plane of the spatial light modulator by the imaging lens, and the edge detection operator pattern loaded by the spatial light modulator is transmitted or reflected for modulation.

[0013] The lighting pattern generation module includes a spatial light modulator; the lighting pattern generation module includes a light source; the spatial light modulator is selected from any one of digital micromirror array (DMD), liquid crystal spatial light modulator (SLM), or rotating encoder disk; the light source is selected from any one of broadband light source, laser light source, or LED.

[0014] The spatiotemporal coupled optical field is constructed by a selected spatial edge operator matrix and orthogonal quantum state eigenfunctions; spatial frequency domain modulation is converted into time frequency domain modulation, and parallel physical convolution and extraction of edge information of the entire field of view are achieved by distinguishing the time-domain sinusoidal frequencies corresponding to different spatial locations.

[0015] The edge detection operator is a spatial filtering matrix with positive and negative weights, selected from any one or a combination of first-order gradient operators, second-order differential operators, directional detection operators, and custom operators;

[0016] The first-order gradient operator includes any one or a combination of the Sobel operator, Prewitt operator, Roberts crossover operator, isotropic Sobel operator, Scharr operator, Krisch operator, and Robinson operator;

[0017] The second-order differential operators include any one or a combination of the Laplacian operator, the Gaussian-Laplace LoG operator, the difference Gaussian DoG operator, and the Marr-Hildreth operator;

[0018] The directional detection operator includes any one or a combination of the Nevatia-Babu operator, the Compass operator, and the gradient magnitude and direction calculation kernel in the Canny edge detection algorithm.

[0019] The custom operator is any spatial convolution kernel with high-pass filtering characteristics, customized according to the target features.

[0020] The spatiotemporal coupled light field generated by the lighting pattern generation module follows orthogonal eigenmodes in the time dimension, constituting a temporal orthogonal eigenmode sequence; the temporal orthogonal eigenmode sequence It is generated by a Sturm-Liouville type differential operator and satisfies the eigenvalue equation. and weighted inner product orthogonality The sequence is selected from one or a combination of the following sets of eigenstates of physical systems: a) Translation-generated type: Eigenstates corresponding to momentum or angular momentum operators, manifested as plane waves or spiral phase wavefronts; b) Potential well-constrained type: Boundary-constrained eigenstates corresponding to one-dimensional infinite potential wells or finite-width waveguides, manifested as sine or cosine sequences; c) Resonant-bound type: Eigenstates corresponding to quantum harmonic oscillators or parabolic potential wells, manifested as Hermitian-Gaussian modes or Laguerre-Gaussian modes; d) Discrete-degree-of-freedom type: Spin or polarization eigenstates corresponding to finite-dimensional Hilbert spaces; and the actual projected light intensity distribution. By the intrinsic modes By applying a DC bias or using differential measurement, the data is mapped to physical quantities that satisfy nonnegativity constraints.

[0021] The hardware of the edge image reconstruction module includes an embedded real-time processing architecture and an offline processing architecture.

[0022] The embedded real-time processing architecture adopts a field-programmable gate array (FPGA) or application-specific integrated circuit (ASIC), and is internally configured with a pipelined multiply-accumulate unit (MAC) or a fast Fourier transform (FFT) core, which is used to simultaneously perform time-domain and frequency-domain demodulation while acquiring bucket signals, so as to achieve "imaging while acquiring".

[0023] The offline processing architecture described above uses a data acquisition card in conjunction with an external computer to store the acquired bucket signals and perform subsequent high-precision reconstruction algorithm processing.

[0024] The imaging method of the single-pixel edge imaging system utilizing the quantum state modulation and edge detection operator convolution comprises the following steps:

[0025] Step a) Encoding construction steps: Selecting target edge detection operators And generate a set of quantum state time basis functions that satisfy the orthogonality condition. Constructing a spatiotemporal coupling modulation function ;

[0026] Step b) Optical field modulation and interaction step: The spatiotemporal coupling modulation function is modulated using a spatial light modulation device. Physical loading is applied to the light field, making the light field and the imaging object function. A physical multiplication effect occurs;

[0027] Step c) Detection step: Spatial integration of the applied light field is performed using a single-pixel detector to obtain a time-varying voltage signal. ;

[0028] Step d) Edge image reconstruction steps: Based on and The orthogonality relationship is used to calculate the convolution value between the object and the operator. Generate an edge image.

[0029] In step d), the edge image reconstruction module runs a computational reconstruction algorithm, which includes any one or a combination of the following:

[0030] Direct reconstruction algorithm based on orthogonal projection: Utilizes the orthogonality of quantum state wave functions in the time domain, and extracts edge response values ​​by calculating the inner product;

[0031] Sparse reconstruction algorithm based on compressed sensing: It utilizes the spatial sparsity of edge images to solve for edges from undersampled data by minimizing an optimization objective function that incorporates a total variation regularization term;

[0032] Algorithms based on deep learning networks: using pre-trained networks or end-to-end mapping networks to recover high signal-to-noise ratio edge images from noisy bucket signals.

[0033] The specific execution method of step d) is as follows:

[0034] d1) Offline reconstruction: The electrical signal of the single-pixel detector is amplified and converted from analog to digital and then transmitted to an external computer, where the computer runs the computational reconstruction algorithm.

[0035] d2) Real-time reconstruction: The electrical signal from the single-pixel detector is input to an application-specific integrated circuit (ASIC), which performs analog-to-digital conversion on-chip and utilizes hardware pipeline technology to run the computational reconstruction algorithm on-chip in real time.

[0036] The beneficial effects of this invention are as follows:

[0037] 1. Direct edge imaging: Using optical analog convolution instead of digital differentiation, edge information is directly output without reconstructing the full image, which greatly reduces the amount of data processing and transmission bandwidth.

[0038] 2. Strong anti-scattering capability: The time-domain orthogonality of quantum state encoding based on the Sturm-Liouville operator can effectively suppress pixel crosstalk and DC background noise caused by scattering medium; combined with the high-pass filtering characteristics of second-order operators such as LoG, it can still extract clear edges in low signal-to-noise ratio environments.

[0039] 3. Flexible reconstruction architecture: The system supports both PC-based offline high-precision reconstruction, suitable for scientific research and medical analysis, and FPGA / ASIC-based on-chip real-time reconstruction, suitable for industrial online inspection, meeting the different speed and accuracy requirements of different scenarios. Attached Figure Description

[0040] Figure 1 This is a framework diagram of the single-pixel edge imaging system based on quantum state encoding of the present invention;

[0041] Figure 2 This is a schematic diagram of a single-pixel edge imaging system with an active quantum state-encoded illumination mode;

[0042] in:

[0043] a is a schematic diagram of a single-pixel edge imaging system based on quantum state illumination modulated by a spatial light modulator; b is a schematic diagram of a single-pixel edge imaging system based on quantum state illumination encoded by a commercial projector.

[0044] 1. Light source; 2. Spatial light modulation device; 3. Imaging lens; 4. Object under test; 5. Converging lens; 6. Single pixel detector; 7. Data processing module; 8. Projector;

[0045] Figure 3 This is a schematic diagram of a single-pixel edge imaging system with a passive quantum state encoded illumination mode. Detailed Implementation

[0046] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be described in detail below with reference to the accompanying drawings and embodiments.

[0047] Example 1: Active Anti-scattering Edge Imaging System

[0048] This embodiment demonstrates a system for directly extracting object edges through a structured light field using active illumination. Depending on the application scenario and cost requirements, this embodiment provides two specific hardware implementations: a discrete device-based configuration and an integrated configuration based on a commercial projector.

[0049] In discrete device-based configurations (such as...) Figure 2 As shown in (a), the system is provided with a light source (1), a spatial light modulator (2), and an imaging lens (3) in sequence along the optical axis. The light beam emitted by the light source (1) is collimated and uniformly illuminates the photosensitive surface of the spatial light modulator (2). The spatial light modulator (2) is the core encoding device and is loaded with a preset spatiotemporal coupling matrix. The physical meaning of this matrix is ​​that, in terms of spatial distribution, it presents the shape of the convolution kernel of a selected edge operator (such as the Gaussian-Laplace LoG operator); in terms of time dimension, its light intensity change follows the quantum eigenstate wave function of a one-dimensional infinite potential well particle (i.e., a set of orthogonal sine wave sequences). The modulated light beam is projected through the imaging lens (3) and illuminates the object under test (4) through the possible scattering medium. The light reflected from the surface of the object carries edge convolution information, which is collected by the converging lens (5) and converged to the single-pixel detector (6). The analog barrel signal output by the detector is transmitted to the edge image reconstruction module, and the edge image is directly output through temporal orthogonal demodulation.

[0050] In configurations based on integrated devices (such as...) Figure 2 As shown in (b), to simplify the optical path and improve the portability of the system, the aforementioned light source (1), spatial light modulation device (2), and imaging lens (3) are replaced with an integrated projector (8). The projector (8) projects a preset spatiotemporal coupling matrix directly onto the object under test (4). The detector transmits the detection barrel signal to the edge image reconstruction module to obtain an anti-scattering edge image.

[0051] Example 2: FPGA-based Real-time Edge Computing System

[0052] This embodiment demonstrates a passive imaging scheme (such as...). Figure 3 As shown in the image, this is particularly suitable for scenarios where the target object is self-illuminating or illuminated by ambient light and active lighting is not advisable. Figure 2 As shown, the system is provided with a target object (4), an imaging lens (3), a spatial light modulation device (2), a converging lens (5), and a single-pixel detector (6) in sequence along the optical path.

[0053] In the specific process, the radiated or reflected light (scene light) from the object under test (4) first passes through the imaging lens (3) and forms a clear real image on the modulation plane of the spatial light modulator (2). At this time, the spatial light modulator (2) (such as DMD) acts as a light field mask to perform transmission or reflection modulation on the real image. Similar to Example 1, the modulation matrix loaded here still adopts a combination of spatial edge operators and temporal quantum sine wave matrix. The beam modulated by DMD is focused by the converging lens (5) and enters the single-pixel detector (6). The edge image reconstruction module collects the light intensity changes of the detector and transmits them to the edge image reconstruction module to obtain the edge image. This embodiment proves that the method is not only applicable to active projection, but also to passive detection, and has wide applicability.

[0054] Example 3: FPGA-based single-pixel edge imaging system

[0055] This embodiment focuses on improving the real-time performance of imaging and computing through hardware acceleration, and constructing an edge device that is a "sensor as a computer". An FPGA embedded development board can be used, and the core of the system lies in the pipeline computing architecture built inside the FPGA. The analog signal output by the single-pixel detector (6) is digitized by a high-speed analog-to-digital converter (ADC) and then directly enters the programmable logic terminal of the FPGA in the form of streaming data.

[0056] The FPGA internally incorporates a massively parallel multiply-accumulate array. When the light intensity sampling data arrives at time t, the logic unit immediately compares it with the current time quantum basis function value pre-stored in the on-chip RAM. The multiplication and accumulation operations are performed. This process is completely synchronized with signal acquisition and does not require waiting for the entire sampling period to end. Once the time-coded sequence of the nth pixel has been transmitted, its corresponding edge response integral value is immediately calculated. The FPGA then calls on-chip DSP resources to perform thresholding or gradient synthesis on this value and writes the final edge pixel into the frame buffer.

[0057] Example 4: Multi-operator switching single-pixel edge imaging system

[0058] This embodiment demonstrates the passive scheme and operator switching capability of the present invention.

[0059] The light beam from the object under test (4) is imaged onto the spatial light modulator (2) by the imaging lens (3).

[0060] This system supports software-defined operator switching: users can achieve different detection functions simply by changing the matrix loaded by the programmable structured light emission module through software instructions, without modifying the hardware. For example, when detecting longitudinal cracks, the system generates an illumination pattern combining the Sobel-Y operator and quantum state encoding, outputting a vertical edge gradient map; when detecting isotropic micro-defects, the system switches to the Laplacian or LoG operator. This flexibility allows the same hardware to adapt to a variety of complex machine vision tasks.

[0061] Image edge reconstruction algorithm

[0062] The edge image reconstruction steps described in this invention are detailed below. The execution unit of this reconstruction step depends on different embodiments of this invention: in an offline scheme, this step is executed by an external computer in the edge image reconstruction module; while in a real-time scheme, this step is executed on-chip by an application-specific integrated circuit (such as an FPGA or ASIC) in the edge image reconstruction module.

[0063] The core of the computational reconstruction step is to run a computational reconstruction algorithm to solve the mathematical inverse problem defined by the imaging model. In this invention, the imaging model is defined by a spatiotemporal coupling matrix composed of temporal quantum eigenstate functions and spatial edge detection operators. The computational reconstruction algorithms that can be used in this invention include, but are not limited to, combinations of one or more of the following categories:

[0064] Direct reconstruction algorithm based on orthogonal projection:

[0065] Such algorithms utilize the orthogonality of quantum state wavefunctions in the time domain. The module calculates the bucket signal. With each basis function The inner product of directly physically extracts the response value of the body to the edge operator at each spatial location.

[0066] Sparse reconstruction algorithm based on compressed sensing:

[0067] These algorithms leverage the high spatial sparsity of edge images. Edges are solved from undersampled data by minimizing an optimization objective that combines data fidelity and regularization terms. For example:

[0068] in, For bucket signal, For the spatiotemporal coding matrix, The edge graph to be determined can be represented by total variational regularization (TVM). Algorithms such as ADMM can be used for solving the problem.

[0069] Algorithms based on deep learning networks:

[0070] These algorithms utilize neural networks to process data with extremely low signal-to-noise ratios.

[0071] a) Data-driven pre-trained networks: These networks require end-to-end pre-training on large-scale "bucket signal-edge image" datasets to learn a nonlinear mapping from one-dimensional temporal measurements (or their preliminary backprojections) to two-dimensional true edge images. The network architecture can be (but is not limited to) convolutional neural networks (CNNs), U-Nets, generative adversarial networks (GANs), Vision Transformers (ViTs), or denoising autoencoders. This method implicitly learns the point spread function (PSF) of the scattering medium, thereby effectively removing speckle noise while reconstructing edges.

[0072] b) Non-pretrained networks based on physical priors: These networks utilize randomly initialized neural network structures as priors for image generation, eliminating the need for pre-collecting large-scale training datasets. The process includes: using the output of a randomly initialized network (e.g., the U-Net architecture) as an estimate of the edge image to be reconstructed; substituting this estimate into the spatiotemporal physical transport model of this invention to derive the predicted bucket signal; iteratively optimizing the network's weight parameters by minimizing the loss function between the predicted bucket signal and the actual measured bucket signal until the network output converges to the final sharp edge image.

[0073] c) Hybrid Algorithms: These algorithms deeply integrate the physical model encoded by quantum states with deep learning networks, combining physical interpretability with data-driven denoising capabilities.

[0074] Using the algorithm described above, the system ultimately outputs a high signal-to-noise ratio edge image of the target object. Those skilled in the art will further recognize that, by combining the algorithm steps described in the embodiments disclosed in this invention, different network structures can be selected to solve the inverse problem of the physical model. The specific network configuration used to perform these functions depends on the specific application and design constraints of the technical solution.

Claims

1. A single-pixel edge imaging system based on quantum state modulation and edge detection operator convolution, characterized in that: It includes a lighting pattern generation module, a test object module, a single pixel detection and data acquisition module, and an edge image reconstruction module; The lighting pattern generation module is used to generate a spatiotemporally coupled light field. The spatiotemporally coupled light field follows an orthogonal sequence of quantum eigenstate wave functions in the time dimension and presents a convolution kernel distribution of edge detection operators in the spatial dimension. The lighting pattern generation module projects a spatiotemporally coupled light field onto the object under test, or modulates the light field from the object under test; The object under test is located between the lighting pattern generation module and the single pixel detection and data acquisition module. It physically interacts with the spatiotemporally coupled light field through reflection, transmission or self-radiation to achieve physical convolution between the edge operator and the spatial information of the object. The single-pixel detection and data acquisition module is used to collect the total light intensity modulated by the object under test, convert the light signal into an analog electrical signal through single-point detection, perform analog-to-digital conversion acquisition, and output a time-domain bucket signal sequence containing edge feature information. The edge image reconstruction module is used to receive a time-domain bucket signal sequence, demodulate the signal using the orthogonality of quantum eigenstate wave functions, separate the edge response values ​​of each spatial location from the bucket signal, and directly reconstruct the edge image of the object under test.

2. The single-pixel edge imaging system based on quantum state modulation and edge detection operator convolution according to claim 1, characterized in that: The lighting pattern generation module adopts an active projection structure or a passive imaging structure; Active projection structure: includes a light source and a spatial light modulator, or a digital projector; used to actively project a spatiotemporally coupled light field loaded with an edge detection operator pattern onto the object under test; Passive imaging structure: includes an imaging lens and a spatial light modulator; light emitted by the object under test or reflected light from the environment is imaged onto the plane of the spatial light modulator by the imaging lens, and the edge detection operator pattern loaded by the spatial light modulator is transmitted or reflected for modulation.

3. The single-pixel edge imaging system based on quantum state modulation and edge detection operator convolution according to claim 1, characterized in that: The lighting pattern generation module includes a spatial light modulator; the lighting pattern generation module includes a light source; the spatial light modulator is selected from any one of digital micromirror array (DMD), liquid crystal spatial light modulator (SLM), or rotating encoder disk; the light source is selected from any one of broadband light source, laser light source, or LED.

4. The single-pixel edge imaging system based on quantum state modulation and edge detection operator convolution according to claim 1, characterized in that: The spatiotemporal coupled optical field is constructed by a selected spatial edge operator matrix and orthogonal quantum state eigenfunctions; spatial frequency domain modulation is converted into time frequency domain modulation, and parallel physical convolution and extraction of edge information of the entire field of view are achieved by distinguishing the time-domain sinusoidal frequencies corresponding to different spatial locations.

5. The single-pixel edge imaging system based on quantum state modulation and edge detection operator convolution according to claim 1, characterized in that: The edge detection operator is a spatial filtering matrix with positive and negative weights, selected from any one or a combination of first-order gradient operators, second-order differential operators, directional detection operators, and custom operators; The first-order gradient operator includes any one or a combination of the Sobel operator, Prewitt operator, Roberts crossover operator, isotropic Sobel operator, Scharr operator, Krisch operator, and Robinson operator; The second-order differential operators include any one or a combination of the Laplacian operator, the Gaussian-Laplace LoG operator, the difference Gaussian DoG operator, and the Marr-Hildreth operator; The directional detection operator includes any one or a combination of the Nevatia-Babu operator, the Compass operator, and the gradient magnitude and direction calculation kernel in the Canny edge detection algorithm. The custom operator is any spatial convolution kernel with high-pass filtering characteristics, customized according to the target features.

6. The single-pixel edge imaging system based on quantum state modulation and edge detection operator convolution according to claim 1, characterized in that: The spatiotemporal coupled light field generated by the lighting pattern generation module follows orthogonal eigenmodes in the time dimension, constituting a temporal orthogonal eigenmode sequence; the temporal orthogonal eigenmode sequence It is generated by a Sturm-Liouville type differential operator and satisfies the eigenvalue equation. and weighted inner product orthogonality The sequence is selected from one or a combination of the following sets of eigenstates of physical systems: a) Translation-generated type: Eigenstates corresponding to momentum or angular momentum operators, manifested as plane waves or spiral phase wavefronts; b) Potential well-constrained type: Boundary-constrained eigenstates corresponding to one-dimensional infinite potential wells or finite-width waveguides, manifested as sine or cosine sequences; c) Resonant-bound type: Eigenstates corresponding to quantum harmonic oscillators or parabolic potential wells, manifested as Hermitian-Gaussian modes or Laguerre-Gaussian modes; d) Discrete-degree-of-freedom type: Spin or polarization eigenstates corresponding to finite-dimensional Hilbert spaces; and the actual projected light intensity distribution. By the intrinsic modes By applying a DC bias or using differential measurement, the data is mapped to physical quantities that satisfy nonnegativity constraints.

7. The single-pixel edge imaging system based on quantum state modulation and edge detection operator convolution according to claim 1, characterized in that: The hardware of the edge image reconstruction module includes an embedded real-time processing architecture and an offline processing architecture. The embedded real-time processing architecture adopts a field-programmable gate array (FPGA) or application-specific integrated circuit (ASIC), and is internally configured with a pipelined multiply-accumulate unit (MAC) or a fast Fourier transform (FFT) core, which is used to simultaneously perform time-domain and frequency-domain demodulation while acquiring the bucket signal, so as to achieve "imaging while acquiring". The offline processing architecture described above uses a data acquisition card in conjunction with an external computer to store the acquired bucket signals and perform subsequent high-precision reconstruction algorithm processing.

8. An imaging method using a single-pixel edge imaging system based on quantum state modulation and edge detection operator convolution as described in any one of claims 1-7, characterized in that, The steps are as follows: Step a) Encoding construction steps: Selecting target edge detection operators And generate a set of quantum state time basis functions that satisfy the orthogonality condition. ; Constructing a spatiotemporal coupling modulation function ; Step b) Optical field modulation and interaction step: The spatiotemporal coupling modulation function is modulated using a spatial light modulation device. Physical loading is applied to the light field, making the light field and the imaging object function. A physical multiplication effect occurs; Step c) Detection step: Spatial integration of the applied light field is performed using a single-pixel detector to obtain a time-varying voltage signal. ; Step d) Edge image reconstruction steps: Based on and The orthogonality relationship is used to calculate the convolution value between the object and the operator. Generate an edge image.

9. The imaging method according to claim 8, characterized in that: In step d), the edge image reconstruction module runs a computational reconstruction algorithm, which includes any one or a combination of the following: Direct reconstruction algorithm based on orthogonal projection: Utilizes the orthogonality of quantum state wave functions in the time domain, and extracts edge response values ​​by calculating the inner product; Sparse reconstruction algorithm based on compressed sensing: It utilizes the spatial sparsity of edge images to solve for edges from undersampled data by minimizing an optimization objective function that incorporates a total variation regularization term; Algorithms based on deep learning networks: using pre-trained networks or end-to-end mapping networks to recover high signal-to-noise ratio edge images from noisy bucket signals.

10. The imaging method according to claim 8, characterized in that: The specific execution method of step d) is as follows: d1) Offline reconstruction: The electrical signal of the single-pixel detector is amplified and converted from analog to digital and then transmitted to an external computer, where the computer runs the computational reconstruction algorithm. d2) Real-time reconstruction: The electrical signal of the single-pixel detector is input to the application-specific integrated circuit (ASIC), which performs analog-to-digital conversion on the chip and uses hardware pipeline technology to run the computational reconstruction algorithm on the chip in real time.

Images