Fourier single-pixel imaging high-resolution reconstruction method based on multi-domain diffusion model

Abstract

The invention discloses a Fourier single-pixel imaging high-resolution reconstruction method based on a multi-domain diffusion model. The Fourier single-pixel imaging high-resolution reconstruction method comprises the following steps: 1) acquiring a target object light response signal through a single-pixel imaging system; 2) analyzing the photoelectric response value based on a phase modulation method, and generating a Fourier sparse sampling frequency spectrum of the target object and a corresponding low-resolution initial image; 3) performing data enhancement on the high-resolution image to obtain a data sample, and continuously adding Gaussian noise to image domain and frequency domain data in a prior learning process to disturb data distribution so as to obtain prior scores of the image domain and frequency domain data training samples; 4) realizing image high-resolution reconstruction through a serial or parallel cooperative strategy, and gradually iterating Gaussian noise from the learned priori score to obtain a target image; 5) introducing a PC sampler to correct errors in the reverse SDE process, and obtaining a corrected reconstructed image.The method can significantly improve the reconstruction precision and the high-frequency detail retention capability under ultra-sparse sampling.

Description

Technical Field

[0001] The invention relates to the technical field of optical imaging, and in particular to a Fourier single-pixel imaging high-resolution reconstruction method based on a multi-domain diffusion model. Background Art

[0002] Fourier single-pixel imaging technology can achieve high dynamic range imaging in low photon flux, wide spectrum and high noise scenes by combining sparse sampling theory with computational imaging. In recent years, it has shown important application value in biological microscopic observation, industrial non-destructive testing and remote sensing monitoring. The core challenge of this technology is how to reconstruct high-fidelity images from a very small amount of Fourier space (k-space) sampling data. Its performance is highly dependent on the efficiency of the reconstruction algorithm in utilizing prior information. Existing methods are mainly divided into two categories: one is based on compressed sensing theory and uses k-space sparsity constraints to build an optimization model; the other is based on deep learning and learns image domain feature mapping relationships through end-to-end networks. However, traditional compressed sensing methods suffer from a sharp drop in reconstruction quality due to serious underdetermination problems when ultra-sparse sampling (such as sampling rate less than 5%), while pure data-driven deep learning models lack physical imaging mechanism embedding, so their generalization ability is limited and artifacts are easily introduced. To overcome the above defects, the existing technology proposes a solution to integrate the generative model (such as the diffusion model) with the physical imaging model. By constructing a diffusion process in the image domain to iteratively optimize the reconstruction results, the imaging quality at low sampling rates is significantly improved. However, this type of method still has obvious limitations: its optimization process only relies on the image domain generation prior, and fails to simultaneously utilize the inherent sparsity characteristics of the k-space, resulting in insufficient ability to restore high-frequency details. Specifically, the single domain constraint forces the algorithm to make a trade-off between data consistency (k-space sampling matching) and image authenticity (generation prior), especially when faced with complex textures or non-stationary noise interference, it is easy to have problems such as blurred edges and residual artifacts in the reconstructed image. In addition, the existing system lacks a dynamic collaborative optimization mechanism between the hardware control module and the reconstruction algorithm, making it difficult to adapt to the rapidly changing imaging scene requirements. Summary of the invention

[0003] The purpose of the present invention is to propose a high-resolution reconstruction method for Fourier single-pixel imaging based on a multi-domain diffusion model in response to actual needs. By generating prior collaborative constraints through the multi-domain diffusion model, the reconstruction accuracy and high-frequency detail retention capability of Fourier single-pixel imaging under ultra-sparse sampling are significantly improved.

[0004] In order to achieve the above object, the present invention adopts the following technical solution.

[0005] A Fourier single pixel imaging high-resolution reconstruction method based on a multi-domain diffusion model comprises the following steps: Step S1, constructing a single pixel imaging system and collecting light response signals of target objects; The single-pixel imaging system includes a computer host, a helium-neon laser, a lens group, a digital micromirror device (DMD), and a photodetector; the helium-neon laser continuously emits red laser light, which is reflected by the lens group into the effective area of the digital micromirror device (DMD). The computer host transmits the Fourier basis pattern to the digital micromirror device (DMD) to modulate the light beam. The modulated light beam is focused by the lens onto the diaphragm and then irradiated onto the target object. The photodetector detects the light response signal and synchronously transmits it to the computer host; Step S2: Analyze the light response signal of the target object; Based on the collected light response signal of the target object, the computer host obtains the Fourier coefficients of the Fourier basis pattern through the four-step phase-shift method and recombines them to obtain the Fourier sparse sampling spectrum of the target object and the corresponding low-resolution initial image; Step S3: Use the multi-domain diffusion model to enhance the data of the high-resolution image; The multi-domain diffusion model includes an image-domain diffusion model and a frequency-domain diffusion model. The image-domain diffusion model and the frequency-domain diffusion model are trained simultaneously. During the forward diffusion process, data enhancement is performed on the high-resolution image-domain and frequency-domain training images. Gaussian noise is continuously added to the training set to perturb the data distribution and obtain the prior scores in the data sample image domain and frequency domain; Step S4: Multi-domain joint optimization based on a serial or parallel collaborative manner; Gradually iterate to obtain the target object image based on the obtained prior scores: In the serial collaborative manner, the initial reconstructed image is generated by the image-domain diffusion model, and then the spectral features are optimized by the frequency-domain diffusion model; in the parallel collaborative manner, the multi-domain diffusion model is run synchronously and the output results are fused; Step S5: Introduce a PC sampler during the process of reconstructing the image based on a serial or parallel collaborative manner to correct the error in the reverse SDE process and obtain the corrected reconstructed image.

[0006] Specifically, in step S1, the computer host transmits the binarized Fourier basis pattern to the digital micromirror device (DMD) to modulate the light beam; the photodetector is a single-pixel photodetector, and the single-pixel photodetector converts the detected light response signal into a voltage signal.

[0007] Specifically, the analysis of the light response signal of the target object in step S2 includes the following content: Based on the collected light response signal of the target object, four Fourier basis patterns with different phases are obtained, and the Fourier coefficients are calculated through the image intensity difference. The Fourier coefficients are calculated by the following formula: ; In the above formula, is the Cartesian coordinate of the Fourier basis pattern; is the imaginary unit; is the photoelectric response value of the Fourier basis pattern with frequency and initial phase ; or or or ; is the Fourier coefficient at the corresponding frequency ; is the system gain factor; By reorganizing the obtained Fourier coefficients, the Fourier spectrum of the target object image is obtained .

[0008] Specifically, in the forward diffusion process of step S3, Gaussian noise is gradually added to the high-resolution image domain and the frequency-domain training image to obtain a stochastic process evolving with time represented as: ; In the above formula, is the change amount of on , describing the change of during the noise addition process; is the drift term, is the diffusion coefficient, is the standard Brownian motion noise; The time schedule of the Gaussian noise variance is represented as: ; In the above formula, is the variance of the Gaussian noise at time ; is the minimum value of the Gaussian noise variance; is the minimum value of the Gaussian noise standard deviation; is the maximum value of the Gaussian noise standard deviation; Select and , then the forward diffusion process is represented by the stochastic differential equation SDE as: ; The prior scores in the data sample image domain and the frequency domain are estimated using the score model with parameters The score model is trained by optimizing the parameters through the following algorithm: ; In the above formula, is the optimal network parameter,​​ To find the parameters that minimize the expected loss, For the time variable expectation; is the time-dependent weighting function; is the expected value of the true data sample expectation; is given the expected value of the noisy data expectation; is the true score, i.e., given when the gradient of the log probability density, is the data after adding noise; is the transition probability distribution of the diffusion process; once the training is completed, , then use to represent the prior score; Apply a weighting strategy to the input data in the frequency domain space of the optimization process to narrow the dynamic range difference between high-frequency and low-frequency information, making the prior score more accessible. The frequency domain space weighting strategy is expressed as: ; In the above formula, is the input data in the weighted frequency domain space; is the input data in the frequency domain space of the training model; is the weight matrix: ; In the above formula, and are the number of frequency encoding lines and phase encoding lines respectively, and are used to adjust the weights, is the set cut-off value, is the smoothness of the weight boundary.

[0009] Specifically, in step S4, the target object image is gradually iteratively obtained from the obtained prior score. The target object image reconstruction process is expressed as: ; In the above formula, is the change amount on , describing the change during the denoising process; represents the variance of the noise at time ; represents the prior score, represents the standard Brownian process noise of time reversal; In each reconstruction iteration, to ensure that the generated image is consistent with the actually acquired Fourier spectrum data, a data consistency update process is introduced: Let the current reconstructed image be , and its Fourier transform be . The actually acquired low-frequency Fourier data is . Define the masking function as . The masking function takes the value of 1 in the low-frequency region and 0 in the remaining regions. Then the reconstructed image after data consistency update is expressed as: .

[0010] Furthermore, in the serial cooperation method described in step S4, the iterative processes of the image-domain and frequency-domain diffusion models are alternately executed. After each iteration, through data consistency update, the low-frequency spectrum components are constrained and corrected: First, use the image-domain diffusion model for preliminary reconstruction to obtain an intermediate result . Subsequently, use the frequency-domain diffusion model to further process to obtain the final image .

[0011] Furthermore, in the parallel cooperation method, the intermediate results and of preliminary reconstruction are independently obtained by using the image-domain diffusion model and the frequency-domain diffusion model respectively. Then, they are weighted and synthesized to obtain the final reconstructed image . The synthesis formula is as follows: ; In the above formula, is the weight coefficient; is the inverse Fourier transform.

[0012] Specifically, in step S5, during the process of reconstructing the image based on the serial or parallel cooperation method, a PC sampler is introduced to correct the error in the reverse SDE process to obtain the corrected reconstructed image. The PC sampler includes a predictor and a corrector, and is iteratively corrected by the Markov chain Monte Carlo (MCMC) method. The prediction algorithm provides directional updates through the prior score to generate a preliminary sample update for iterative correction. The prediction algorithm is expressed as: ; In the above formula, is the sample state of predicting the previous time step ; is the sample state of the current time step ; is the discretized time step; is with respect to The prior score; is Gaussian noise with a mean of 0 and a covariance matrix of ; I is the identity matrix; The correction algorithm is based on the same prior score , and corrects the prediction result along the gradient direction to reduce the deviation caused by random noise in the prediction step; The correction algorithm is expressed as: ; In the above formula, is the corrected sample; is the gradient ascent step size.

[0013] Compared with the prior art, the present invention has the following beneficial effects: 1. The method of the present invention breaks through the limitations of traditional diffusion models in a single domain by combining double diffusion models in the image domain and the frequency domain, and adopts a serial and parallel collaborative training method to effectively improve the accuracy of image reconstruction under the condition of an extremely low sampling rate.

[0014] 2. The method of the present invention greatly improves the imaging quality of the system through multi-domain collaborative training, making it have a wider application prospect, especially in the fields of medical imaging, physical detection, and computer vision. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 is a flowchart of the Fourier single-pixel imaging high-resolution reconstruction method based on the multi-domain diffusion model of the present invention; Figure 2 is a block diagram of the architecture of the single-pixel imaging system and the multi-domain diffusion model of the present invention; Figure 3 is a flowchart of the high-resolution iterative reconstruction parallel collaboration based on the multi-domain diffusion model of the present invention; Figure 4 is a flowchart of the reconstruction part of the serial collaboration mode and the parallel collaboration mode of the present invention; Figure 5 is a reconstruction result diagram of the target object being a cat in the embodiment of the present invention; Figure 6 is a comparison diagram of the SSIM index of the reconstruction results of the method of the present invention and the traditional method. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0016] To facilitate the understanding and implementation of the present invention by those of ordinary skill in the art, the following provides a detailed description of each step of the method proposed by the present invention. It should be understood that these embodiments are only used to illustrate the present invention and not to limit the scope of the present invention. In addition, it should be understood that after reading the content taught by the present invention, those skilled in the art can make various changes or modifications to the present invention, and these equivalent forms also fall within the scope defined by the appended claims of this application.

[0017] Embodiment As Figure 1 shown, the present invention discloses a high-resolution reconstruction method for Fourier single-pixel imaging based on a multi-domain diffusion model, including the following steps: Step S1, constructing a single-pixel imaging system and collecting the optical response signal of the target object; As Figure 2 shown, Figure 2 the single-pixel imaging system in the left block diagram in includes a computer host, a helium-neon laser, a lens group, a digital micromirror device DMD, and a photodetector; the helium-neon laser continuously emits red laser light, and the laser light is reflected by the lens group into the effective area of the digital micromirror device DMD. The computer host transmits the Fourier basis pattern to the digital micromirror device DMD to modulate the light beam. The modulated light beam is focused by the lens to the aperture and then irradiated onto the target object. The photodetector detects the optical response signal and synchronously transmits it to the computer host; Step S2, analyzing the optical response signal of the target object; The computer host obtains the Fourier coefficients of the Fourier basis pattern through the four-step phase-shift method according to the collected optical response signal of the target object, and reorganizes them to obtain the Fourier sparse sampling spectrum of the target object and the corresponding low-resolution initial image; Step S3, using a multi-domain diffusion model to perform data enhancement on the high-resolution image; As Figure 2 shown, Figure 2 the multi-domain diffusion model in the right block diagram in includes an image-domain diffusion model and a frequency-domain diffusion model. The image-domain diffusion model and the frequency-domain diffusion model are trained simultaneously. During the forward diffusion process, data enhancement is performed on the high-resolution image-domain and frequency-domain training images, and Gaussian noise is continuously added to the training set to perturb the data distribution, obtaining the prior scores on the data sample image domain and frequency domain; Step S4, multi-domain joint optimization based on a serial or parallel collaborative manner; The target object image is gradually obtained by iterative calculation based on the obtained prior scores: in the serial collaborative manner, the initial reconstructed image is generated by the image-domain diffusion model, and then the spectral features are optimized by the frequency-domain diffusion model; in the parallel collaborative manner, the multi-domain diffusion model is run synchronously and the output results are fused; Step S5: During the process of reconstructing the image based on the serial or parallel collaboration method, introduce a PC sampler to correct the error in the reverse SDE process, and obtain the corrected reconstructed image.

[0018] Specifically, in step S1, the computer host transmits the binarized Fourier basis pattern to the digital micromirror device (DMD) to modulate the light beam; the photodetector is a single-pixel photodetector, and the single-pixel photodetector converts the detected light response signal into a voltage signal.

[0019] Specifically, the analysis of the target object light response signal in step S2 includes the following content: According to the collected target object light response signal, obtain four Fourier basis patterns with different phases, and calculate the Fourier coefficients through the image intensity difference. The Fourier coefficients are calculated by the following formula: ; In the above formula, is the Cartesian coordinate of the Fourier basis pattern; is the imaginary unit; is the light response value of the Fourier basis pattern with frequency and initial phase ; or or or ; is the Fourier coefficient corresponding to the frequency ; is the system gain factor; By reorganizing the obtained Fourier coefficients, the Fourier spectrum of the target object image is obtained .

[0020] As Figure 3 shown, Figure 3 in which FFT represents the fast Fourier transform, and IFFT represents the inverse fast Fourier transform; in the forward diffusion process of step S3, Gaussian noise is gradually added to the high-resolution image domain and the frequency domain training image respectively, to obtain a random process evolving with time which is expressed as: ; In the above formula, is the change amount of on , describing the change of during the noise addition process; is the drift term, is the diffusion coefficient, is the standard Brownian motion noise; The time scheduling of the Gaussian noise variance is expressed as: ; In the above formula, is the variance of the Gaussian noise at time ; is the minimum value of the Gaussian noise variance; is the minimum value of the Gaussian noise standard deviation; is the maximum value of the Gaussian noise standard deviation; Select and , then the forward diffusion process is represented by the stochastic differential equation SDE as: ; The prior scores in the data sample image domain and frequency domain are estimated using the score model with parameters . The score model is trained by optimizing the parameters through the following algorithm: ; In the above formula, is the optimal network parameter, is to find the parameter that minimizes the expected loss, is the expectation with respect to the time variable ; is the time-dependent weighting function; is the expectation of the true data sample ; is the expectation of the noise-added data given ; is the true score, that is, the gradient of the log probability density when is given, is the data after adding noise; is the transition probability distribution of the diffusion process; once the training is completed, , then is used to represent the prior score; Since in the frequency domain space, the low-frequency and high-frequency components are located in the central region and the peripheral region respectively, resulting in a large dynamic range change between the low spatial frequency and high spatial frequency of the frequency domain space data, a weighting strategy is imposed on the input data in the frequency domain space of the optimization process to narrow the dynamic range difference between the high-frequency and low-frequency information, making the prior score easier to obtain. The frequency domain space weighting strategy is expressed as: ; In the above formula, is the input data in the weighted frequency domain space; is the input data of the frequency domain space for training the model; is the weight matrix: ; In the above formula, and are the numbers of the frequency encoding line and the phase encoding line respectively, and are used to adjust the weights, is the set cut-off value, is the smoothness of the weight boundary.

[0021] Specifically, in step S4, the target object image is gradually iteratively obtained based on the obtained prior score, and the target object image reconstruction process is expressed as: ; In the above formula, is the variation of describing the variation in the denoising process; is the variance of the Gaussian noise at time ; represents the prior score, represents the standard Brownian process noise of time reversal; In each reconstruction iteration, in order to ensure that the generated image is consistent with the actually acquired Fourier spectrum data, a data consistency update process is introduced: Let the current reconstructed image be , and its Fourier transform is , the actually acquired low-frequency Fourier data is , the masking function is defined as , and the masking function takes the value of 1 in the low-frequency region and 0 in the remaining regions. After the data consistency update, the reconstructed image is expressed as: .

[0022] As Figure 4 shown, Figure 4 in which FFT represents the fast Fourier transform, IFFT represents the inverse fast Fourier transform, and Mean represents the average value; in the serial cooperation method, the iterative processes of the image domain and the frequency domain diffusion models are alternately executed. After each iteration, through the data consistency update, the low-frequency spectrum components are constrained and corrected: first, the image domain diffusion model is used for preliminary reconstruction to obtain the intermediate result , and then the frequency domain diffusion model is used to further process to obtain the final image ; In the parallel collaborative method, the intermediate results of preliminary reconstruction are separately and independently obtained by using the image domain diffusion model and the frequency domain diffusion model and , and the two are weighted and synthesized to obtain the final reconstructed image , and the synthesis formula is as follows: ; In the above formula, is the weight coefficient; is the inverse Fourier transform.

[0023] Specifically, in step S5, a PC sampler is introduced during the process of reconstructing the image based on the serial or parallel collaborative method to correct the error in the reverse SDE process, and the corrected reconstructed image is obtained; the PC sampler includes a predictor and a corrector, and is iteratively corrected by the Markov chain Monte Carlo MCMC method. The prediction algorithm provides directional updates through the prior score to generate a preliminary sample update for iterative correction. The prediction algorithm is expressed as: ; In the above formula, is the sample state at the previous time step ; is the sample state at the current time step ; is the discretized time step; is the prior score with respect to ; is Gaussian noise with a mean of 0 and a covariance matrix of , I is the identity matrix; The correction algorithm is based on the same prior score , and corrects the prediction result along the gradient direction to reduce the deviation caused by random noise in the prediction step; The correction algorithm is expressed as: ; In the above formula, is the corrected sample; is the gradient ascent step size.

[0024] Next, through an example, the technical effects of the method of the present invention are further described.

[0025] Experimental equipment configuration and preparation: A helium-neon laser (model: JDSU-1137, manufacturer: Power Technology, wavelength: 632.8 nm, beam diameter: 0.84 mm) emits a beam; this beam passes through a beam expander system composed of plano-convex lenses L1 (focal length: 18 mm) and L2 (focal length: 150 mm), with a magnification of 8.3 times; then, the beam is reflected by mirror M1 to a digital micromirror device DMD (model: V-7001VIS, manufacturer: Aunion Tech, resolution: 1024×768); the computer transmits the binarized Fourier basis pattern to the digital micromirror device DMD to modulate the beam; the modulated Fourier basis pattern is magnified by plano-convex lens L3 (focal length: 150 mm); its first-order diffracted light passes through the aperture and irradiates the target, and plano-convex lens L4 is used to collect the diffuse reflection light signal of the reconstructed target and uniformly focus it on an A4 paper; a single-pixel photodetector (model: DET025A / M, manufacturer: Thorlabs, wavelength range: 400 - 1700 nm) converts the detected light response signal into a voltage signal, and a data acquisition card DAQ (model: PicoScope 3405D, manufacturer: Pico Technology, bandwidth: 100 MHz, sampling rate: 1 GS / s) synchronously transmits it to the computer host.

[0026] As Figure 5 shown is the reconstruction result diagram of the target object being a cat in this embodiment; Figure 5 the first column in it represents the sparse sampling rate (5%, 3% and 1%); Figure 5 the second column in it is the reconstruction result diagram using the traditional FSPI method, and the subsequent two columns are a group. Above each group, the model used for reconstruction is marked. In the left column of each group is the reconstructed picture, and the number above the reconstructed picture represents the iteration times for obtaining the best result. The right column is the spectrogram, and the last group is the ground truth GT of the picture; the SSIM value and PSNR value are marked in the upper left corner of the reconstructed picture.

[0027] From the reconstructed pictures and spectrograms, it can be clearly seen that whether it is the SSIM evaluation index or the PSNR evaluation index, the reconstruction result of the multi-domain diffusion model proposed by the method of the present invention is better than that of the single-domain diffusion model, and due to the loss of high-frequency information during the reconstruction process of the traditional FSPI method, the obtained reconstruction result is relatively poor, while the pictures reconstructed by the method of the present invention have richer high-frequency information.

[0028] Furthermore, to verify the effectiveness of the multi-domain diffusion model proposed in the present invention, a series of ablation experiments were conducted in this embodiment. Based on the same data, the serial collaboration and parallel collaboration methods of the single-frequency domain diffusion model, the single-image domain diffusion model, and the multi-domain diffusion model were compared, and the structural similarity index SSIM between the reconstructed image and the real image was statistically analyzed. At the same time, to evaluate the robustness of the model, eight different animal (cat) images were selected for experimental analysis.

[0029] As Figure 6 shown, the experimental results are presented in the form of a bar chart, where the abscissa represents the sparse sampling rate (5%, 3%, 1%, and the average value), and the ordinate is the SSIM value. The reconstruction effects of different methods such as untreated (DC), single-frequency domain diffusion model (single-frequency domain), single-image domain diffusion model (single-image domain), parallel collaboration (parallel connection), and serial collaboration (series connection) of the multi-domain diffusion model are shown respectively at each sampling rate. From Figure 6 it can be clearly seen that compared with the traditional FSPI method, the image reconstruction results using the multi-domain diffusion model proposed in the present invention are significantly better than the traditional method, and the reconstruction quality of the multi-domain diffusion model is better than that of the single-domain diffusion model, further verifying the effectiveness and robustness of the multi-domain diffusion model in the image reconstruction task.

[0030] The above are only the preferred embodiments of the present invention, and are not intended to limit the present invention in other forms. Any person skilled in the art may use the disclosed technical content to make changes or modifications into equivalent embodiments with equivalent changes. However, any simple modification, equivalent change, and modification made to the above embodiments based on the technical essence of the present invention without departing from the technical solution content of the present invention still fall within the protection scope of the technical solution of the present invention.

Claims

1. A high-resolution reconstruction method for Fourier single-pixel imaging based on a multi-domain diffusion model, characterized in that It includes the following steps: Step S1, constructing a single-pixel imaging system and collecting the optical response signal of the target object; The single-pixel imaging system includes a computer host, a helium-neon laser, a lens group, a digital micromirror device (DMD), and a photodetector; the helium-neon laser continuously emits red laser light, and the laser light is reflected by the lens group into the effective area of the digital micromirror device (DMD). The computer host transmits the Fourier basis pattern to the digital micromirror device (DMD) to modulate the light beam. The modulated light beam is focused by the lens onto the diaphragm and then irradiated onto the target object. The photodetector detects the optical response signal and synchronously transmits it to the computer host; Step S2, analyzing the optical response signal of the target object; Based on the collected optical response signal of the target object, the computer host obtains the Fourier coefficients of the Fourier basis pattern through the four-step phase-shift method and reorganizes them to obtain the Fourier sparse sampling spectrum of the target object and the corresponding low-resolution initial image; Step S3, using a multi-domain diffusion model to enhance the data of the high-resolution image; The multi-domain diffusion model includes an image-domain diffusion model and a frequency-domain diffusion model. The image-domain diffusion model and the frequency-domain diffusion model are trained simultaneously. In the forward diffusion process, the data of the high-resolution image domain and the frequency-domain training images are enhanced, and Gaussian noise is continuously added to the training set to perturb the data distribution and obtain the prior scores in the data sample image domain and frequency domain; Step S4, multi-domain joint optimization based on a serial or parallel collaborative manner; The target object image is gradually obtained through iterative calculations based on the obtained prior scores: in the serial collaborative manner, the initial reconstructed image is generated by the image-domain diffusion model, and then the spectral features are optimized by the frequency-domain diffusion model; in the parallel collaborative manner, the multi-domain diffusion model is run synchronously and the output results are fused; Step S5, introducing a PC sampler during the process of reconstructing the image based on a serial or parallel collaborative manner to correct the error in the reverse SDE process and obtain the corrected reconstructed image.

2. A high-resolution reconstruction method for Fourier single-pixel imaging based on a multi-domain diffusion model according to claim 1, characterized in that, In step S1, the computer host transmits the binarized Fourier basis pattern to the digital micromirror device (DMD) to modulate the light beam; the photodetector is a single-pixel photodetector, and the single-pixel photodetector converts the detected optical response signal into a voltage signal.

3. A high-resolution reconstruction method for Fourier single-pixel imaging based on a multi-domain diffusion model according to claim 1, characterized in that In step S2, the analysis of the optical response signal of the target object includes the following content: Based on the collected optical response signal of the target object, four Fourier basis patterns with different phases are obtained, and the Fourier coefficients are calculated through the image intensity difference. The Fourier coefficients are calculated by the following formula: ； In the above formula, is the Cartesian coordinate of the Fourier basis pattern; is the imaginary unit; is the optoelectronic response value of the Fourier basis pattern with a frequency of and an initial phase of , or or or ; is the Fourier coefficient at the corresponding frequency ; is the system gain factor; By reorganizing the obtained Fourier coefficients, the Fourier spectrum of the target object image is obtained .

4. A high-resolution reconstruction method for Fourier single-pixel imaging based on a multi-domain diffusion model according to claim 1, characterized in that In the forward diffusion process in step S3, for the high-resolution image domain and the frequency-domain training image add Gaussian noise step by step to obtain a stochastic process evolving over time represented as: ； In the above formula, is the change amount of describing the change during the noise addition process; is the drift term, is the diffusion coefficient, is the standard Brownian motion noise;​ The time scheduling of the Gaussian noise variance is expressed as: ； In the above formula, is the variance of Gaussian noise at time ; is the minimum value of the variance of Gaussian noise; is the minimum value of the standard deviation of Gaussian noise; is the maximum value of the standard deviation of Gaussian noise; Select and , the forward diffusion process is represented by the stochastic differential equation SDE as follows: ； The prior scores in the data sample image domain and frequency domain are estimated using a score model with parameters of the score model The parameters are optimized through the following algorithm for training: ； In the above formula, is the optimal network parameter, is to find the parameter that minimizes the expected loss, is the expectation with respect to the time variable ; is the time-dependent weighting function; is the expectation of the true data sample ; is the expectation of the noisy data given ; is the true score, that is, the gradient of the log probability density when is given , is the data after adding noise; is the transition probability distribution of the diffusion process; once the training is completed, , then is used to represent the prior score; Apply a weighting strategy to the input data in the frequency domain space of the optimization process to narrow the dynamic range difference between high-frequency and low-frequency information, making the prior score more easily obtainable. The weighting strategy in the frequency domain space is expressed as: ； In the above formula, is the input data in the weighted frequency domain space; is the input data in the frequency domain space of the training model; is the weight matrix: ； In the above formula, and are the numbers of the frequency encoding line and the phase encoding line respectively, and are used to adjust the weights, is the set cut-off value, is the smoothness of the weight boundary.

5. A high-resolution reconstruction method for Fourier single-pixel imaging based on a multi-domain diffusion model according to claim 1, characterized in that, In step S4, the process of gradually obtaining the target object image through iterative calculations based on the obtained prior scores is expressed as: ； In the above formula, is the change in , describing the change during the denoising process; is the variance of Gaussian noise at time ; represents the prior score, represents the standard Brownian motion noise in reverse time; In each reconstruction iteration, in order to ensure that the generated image is consistent with the actually collected Fourier spectrum data, a data consistency update process is introduced: Let the current reconstructed image be , and its Fourier transform be . The actually acquired low-frequency Fourier data is . Define the mask function as . The mask function takes the value of 1 in the low-frequency region and 0 in the remaining regions. After data consistency update, the reconstructed image is expressed as: 。 6. A high-resolution reconstruction method for Fourier single-pixel imaging based on a multi-domain diffusion model according to claim 5, characterized in that, In the serial cooperation method described in step S4, the iterative processes of the image-domain and frequency-domain diffusion models are alternately executed. After each iteration, data consistency is used for updating to constrain and correct the low-frequency spectral components: First, the image-domain diffusion model is used for preliminary reconstruction to obtain an intermediate result , and then the frequency-domain diffusion model is used to perform further processing to obtain the final image .

7. A high-resolution reconstruction method for Fourier single-pixel imaging based on a multi-domain diffusion model according to claim 5, characterized in that, In the parallel cooperation method described in step S4, the intermediate results of preliminary reconstruction are obtained separately and independently by using the image domain diffusion model and the frequency domain diffusion model and , and the two are weighted and synthesized to obtain the final reconstructed image , and the synthesis formula is as follows: ； In the above formula, is the weight coefficient; is the inverse Fourier transform.

8. A high-resolution reconstruction method for Fourier single-pixel imaging based on a multi-domain diffusion model according to claim 1, characterized in that In step S5, during the process of reconstructing the image based on the serial or parallel cooperation mode, a PC sampler is introduced to correct the error in the reverse SDE process, and a corrected reconstructed image is obtained; the PC sampler includes a predictor and a corrector, and is iteratively corrected by the Markov chain Monte Carlo (MCMC) method. The prediction algorithm is based on the prior score provides a directional update to generate a preliminary sample update for iterative correction. The prediction algorithm is expressed as: ； In the above formula, is the sample state at the previous time step ; is the sample state at the current time step ; is the discretized time step; is the prior score with respect to ; is Gaussian noise with a mean of 0 and a covariance matrix of ; I is the identity matrix; The calibration algorithm is based on the same prior score , and corrects the prediction result along the gradient direction to reduce the deviation caused by random noise in the prediction step; The correction algorithm is expressed as: ； In the above formula, is the corrected sample; is the gradient ascent step size.

Images