Plug-and-play single-pixel imaging method and system based on efficient mask sorting

Abstract

The invention provides a plug-and-play single-pixel imaging method and system based on efficient mask sorting, and the method comprises the steps: enabling a spatial light modulator to load the efficient mask sorting, carrying out the coding, and then carrying out the measurement, thereby obtaining a measurement value in an image reconstruction process, and carrying out the iteration solving through a PnP-ADMM frame. The system comprises a coding modulation module, an information acquisition module and an image reconstruction module. A measurement value of the image reconstruction module in an image reconstruction process is coded by loading a high-efficiency mask sequence by a spatial light modulator of the coding modulation module and is measured by the information acquisition module; and optimization iteration is iteratively solved by a PnP-ADMM framework. Redundant sampling is reduced through efficient mask sorting, and the information acquisition efficiency is improved to the maximum extent; and high-quality decoding of low-sampling data is realized by further combining a plug-and-play algorithm with an advanced deep denoising network, high-quality image reconstruction is realized at an extremely low sampling rate, and meanwhile, the method has robustness to noise.

Description

Technical Field

[0001] The present invention belongs to the technical field of optical imaging, and particularly relates to a plug-and-play single-pixel imaging method and system based on efficient mask sorting. Background Art

[0002] Single-pixel imaging technology uses coded wide-field illumination and a single-point detector to capture data, and has the advantages of low cost, high signal-to-noise ratio, and wide spectral range. By combining compressive sensing technology and utilizing the sparsity of natural images in the transform domain, high-quality images of the target scene can be reconstructed when the sampling rate is lower than the Nyquist sampling limit. It has been widely applied in fields such as scattering imaging, extremely weak light imaging, three-dimensional imaging, multispectral imaging, and near-field terahertz imaging.

[0003] Single-pixel imaging technology based on the compressive sensing theory can reconstruct high-quality images even at low sampling rates. The reconstruction quality at low sampling rates is affected by two main factors, namely the measurement matrix and the reconstruction algorithm. Currently, various methods have been proposed to reduce the sampling rate and improve the image reconstruction quality. On the one hand, in order to achieve reconstruction at low sampling rates, a measurement matrix most relevant to the basis mask must be selected, which has been widely studied and proven, such as Dyadic sorting, Russian Doll (RD) sorting, Origami Pattern (OP) sorting, Cake Cutting (CC) sorting, etc. in Hadamard matrix sorting. These sorting methods greatly facilitate matrix operations and basis mask extraction, avoid the storage of large-scale measurement matrices, and more importantly, these measurement matrices greatly improve the single-pixel imaging reconstruction quality at low sampling rates, which is very beneficial for the fast reconstruction of high-resolution compressive sensing.

[0004] On the other hand, in order to further improve the image reconstruction quality, a high-performance image reconstruction algorithm is also required. Existing reconstruction algorithms can be divided into model-based methods, deep learning-based methods, and physical prior plug-and-play methods. Model-based methods, such as conjugate gradient descent method, alternating projection method, and total variation augmented Lagrangian alternating direction algorithm based on compressive sensing (CS) (TVAL3), can show strong generalization ability, but there will be obvious performance degradation and long reconstruction time at low sampling rates. Deep learning-based methods can achieve fast operation and better performance at low sampling rates, but have limited generalization ability and need to be retrained when the sampling rate changes. In addition, the increase in the sampling rate does not necessarily lead to a significant improvement in the image reconstruction performance. The PnP method based on physical prior combines a regularization optimization iterative algorithm with an advanced denoising prior. By using a plug-in denoiser, it avoids network retraining when the sampling ratio changes, has higher generality and reconstruction efficiency, and improves the performance of model-based algorithms while maintaining noise robustness. Summary of the Invention

[0005] In order to obtain an imaging scheme for a single-pixel system that comprehensively considers the noise robustness and image reconstruction quality at a low sampling rate in optical modulation coding and computational reconstruction decoding, the present invention proposes a plug-and-play single-pixel imaging method and system based on efficient mask sorting. Aiming at the problem of poor image reconstruction quality under extremely low sampling rates and strong noise environments, the present invention comprehensively considers the influence of light field modulation and computational decoding on single-pixel imaging performance, reduces redundant sampling through efficient mask sorting, and maximizes the efficiency of information acquisition. Furthermore, by combining with a plug-and-play algorithm with an advanced deep denoising network, high-quality decoding of low-sampling data is achieved, high-quality image reconstruction is realized at an extremely low sampling rate, and at the same time, it is robust to noise.

[0006] In terms of system design, the key to the present invention's solution lies in the cooperation between the encoding modulation module and the image reconstruction module. In the encoding modulation module, after a series of collimations, the light source passes through an optical mask plate carrying object information, projects the incident light with object information onto a spatial light modulator, and sequentially loads the generated efficient modulation mask patterns onto the spatial light modulator to perform optical modulation coding on the projected object information.

[0007] After that, an information acquisition module is adopted. A single-pixel detector detects the output light field modulated by the mask pattern, receives the output signal of the single-pixel imaging system, acquires optical intensity information less than the Nyquist sampling theorem, and digitizes the signal by a data acquisition card and transmits it to a computer for storage.

[0008] In the image reconstruction module, with the PnP-ADMM framework as the core, a data-model dual-driven plug-and-play algorithm is adopted. An imaging optimization model under undersampling conditions is established by the compressive sensing theory, and the data fidelity term and the prior regularization term are decomposed into two sub-problems using the alternating direction method of multipliers. By combining with the plug-and-play algorithm, the solution of the sub-problem of the prior regularization term is replaced by solving an image denoising problem, and finally high-quality image reconstruction at an extremely low sampling rate is achieved.

[0009] In summary, the present invention aims at the problem that the optical modulation coding and computational reconstruction decoding of the existing single-pixel imaging system are separately designed without global optimization, and the image reconstruction quality is poor under extremely low sampling rates and strong noise environments. A new single-pixel imaging scheme with joint optimization of modulation coding and computational decoding is designed and experimented. By combining the plug-and-play algorithm with efficient mask sorting, high-quality reconstruction at an extremely low sampling rate can be achieved. The designed method uses efficient mask sorting to sort and extract the importance of basis vectors from the highest to the lowest, and integrates an advanced deep learning denoiser to improve the image reconstruction quality while increasing the capture efficiency.

[0010] The technical solution specifically adopted by the present invention to solve its technical problems is as follows:

[0011] A plug-and-play single-pixel imaging method based on efficient mask sorting: The measurement values in the image reconstruction process are obtained by measuring after being encoded by loading an efficient mask sorting on a spatial light modulator, and the optimization iteration is solved iteratively by the PnP-ADMM framework.

[0012] Further, the specific encoding by loading an efficient mask sorting on the spatial light modulator is as follows: The light source passes through a series of collimations and then passes through an optical mask template carrying object information, projects the incident light with object information onto the spatial light modulator, and sequentially loads the generated efficient modulation mask patterns on the spatial light modulator to optically encode the projected object information.

[0013] Further, the generation method of the mask pattern is as follows:

[0014]

[0015] where is a set of one-dimensional basis vectors recursively expanded from the initial seed vector s, α k v is the basis vector v multiplied by the value of α k α = α1,..., α k ∈ {+1, -1} uniquely defines each basis vector in, and is the α-index of v. When the Hamming distance between the α-indices of two basis vectors is 1, the two vectors are α-related, and the set of continuously α-related is the Gray code sequence;

[0016] The two-dimensional form of is:

[0017]

[0018] where v1×v2 is the outer product of the one-dimensional basis vectors v1 and v2, which is applied to generate a two-dimensional Hadamard basis mask large matrix, and the Hadamard basis masks in the matrix are extracted by zigzag traversal to generate the required mask pattern.

[0019] Further, after generating the mask pattern, each basis mask is split into two complementary binary matrices by the following formula:

[0020]

[0021] where +1 in H + is mapped to 1, and -1 is mapped to 0; -1 in H - is mapped to 1, and +1 is mapped to 0. The measurement value is obtained by subtracting the measurement value encoded by H + from the measurement value encoded by H - .

[0022] Furthermore, the image reconstruction establishes an imaging optimization model under undersampling conditions based on the compressed sensing theory. The alternating direction method of multipliers (ADMM) is used to decompose the data fidelity term and the prior regularization term into two sub-problems. Combining with the plug-and-play (PnP) algorithm, the solution of the sub-problem of the prior regularization term is replaced by solving the image denoising problem. Finally, the sub-problems of the data fidelity term and the prior regularization term are alternately solved to achieve high-quality image reconstruction at a low sampling rate.

[0023] In addition, a plug-and-play single-pixel imaging system based on efficient mask sorting includes: an encoding and modulation module, an information acquisition module, and an image reconstruction module; the measurement value in the image reconstruction process of the image reconstruction module is encoded by loading an efficient mask sorting on the spatial light modulator of the encoding and modulation module and measured by the information acquisition module, and the optimization iteration is solved iteratively by the PnP-ADMM framework.

[0024] Furthermore, the generation of the mask and the image reconstruction process are implemented by a computer.

[0025] Furthermore, the measurement value in the image reconstruction process of the image reconstruction module is encoded by loading an efficient mask sorting on the spatial light modulator of the encoding and modulation module: the light source passes through a series of collimations and then passes through an optical mask template carrying object information, projects the incident light with object information onto the spatial light modulator, and sequentially loads the generated efficient modulation mask pattern on the spatial light modulator to optically encode the projected object information.

[0026] Furthermore, the generation method of the mask pattern is:

[0027]

[0028] where is a set of one-dimensional basis vectors recursively expanded from the initial seed vector s, α k v is the basis vector v multiplied by α k value, α = α1,…,α k ∈{+1, -1} uniquely defines each basis vector in, the α-index of v, when the Hamming distance of the α-indices of two basis vectors is 1, the two vectors are α-related, and the set of continuously α-related is the Gray code sequence;

[0029] The two-dimensional form of is:

[0030]

[0031] Among them, v1×v2 is the outer product of one-dimensional basis vectors v1 and v2, which is applied to generate a large two-dimensional Hadamard basis mask matrix. The Hadamard basis masks in the matrix are extracted by zigzag traversal to generate the required mask pattern.

[0032] Further, after generating the mask pattern, each basis mask is split into two complementary binary matrices by the following formula:

[0033]

[0034] where +1 in H + is mapped to 1, and -1 is mapped to 0; -1 in H - is mapped to 1, and +1 is mapped to 0. The measured value is obtained by subtracting the measured value encoded by H + from the measured value encoded by H - .

[0035] Compared with the prior art, the present invention and its preferred solutions reduce redundant sampling through efficient mask sorting, maximize the efficiency of information acquisition, optimize the decoding process of low-sampled data through the PnP algorithm with an advanced deep denoising network to improve the reconstruction quality, do not require retraining the network when the sampling rate changes, have good generalization, and achieve excellent image imaging quality at extremely low sampling rates, thus being able to better complete more complex optical information processing tasks. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] The present invention will be further described in detail below with reference to the drawings and specific embodiments:

[0037] Figure 1 It is a schematic diagram of the Plug-and-Play single-pixel imaging method and system based on efficient mask sorting according to an embodiment of the present invention.

[0038] Figure 2 It is a schematic diagram of the generation and extraction of efficient mask sorting (EMS) according to an embodiment of the present invention, where Figure 2 (a) is for GCS to create one-dimensional basis vectors, Figure 2 (b) is for GCS to create a two-dimensional basis mask set, Figure 2 (c) is a schematic diagram of zigzag traversal.

[0039] Figure 3 It is a schematic diagram of the correlation between encoding modulation and image reconstruction according to an embodiment of the present invention.

[0040] Figure 4 It is a numerical simulation comparison diagram of the "Cameraman" image under different methods and different typical sampling rates according to an embodiment of the present invention.

[0041] Figure 5Numerical simulation comparison diagram of the "House" image of the embodiment of the present invention under different methods and extremely low sampling rates.

[0042] Figure 6 Numerical simulation comparison diagram of the "Cameraman" image of the embodiment of the present invention under different methods and different noise levels.

[0043] Figure 7 Optical experimental light path diagram of single-pixel imaging of the embodiment of the present invention.

[0044] Figure 8 Optical experimental comparison diagram of the "Fu" image of the embodiment of the present invention under different methods and different sampling rates, where Figure 8 (a) is the single-pixel imaging results of the "Fu" target image (64×64) under different typical sampling rates and different methods, Figure 8 (b) is the single-pixel imaging results of the "Fu" target image (64×64) under different extremely low sampling rates and different methods. Detailed implementation manners

[0045] To make the features and advantages of this patent more obvious and understandable, specific embodiments are given below for detailed description as follows:

[0046] It should be noted that the following detailed descriptions are all illustrative and are intended to provide further explanations for this application. Unless otherwise specified, all technical and scientific terms used in this specification have the same meaning as commonly understood by those of ordinary skill in the technical field to which this application belongs.

[0047] It should be noted that the terms used here are only for describing specific implementation manners and are not intended to limit the exemplary implementation manners according to this application. As used here, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0048] Aiming at the technical problems existing in the background art, the embodiment of the present invention reduces redundant sampling through efficient mask sorting, maximizes the efficiency of information acquisition, optimizes the decoding process of low-sampling data through the PnP algorithm with an advanced depth denoising network to improve the reconstruction quality, does not require retraining the network when the sampling rate changes, has good generalization, and achieves excellent image imaging quality at extremely low sampling rates.

[0049] Next, a plug-and-play single-pixel imaging method and system based on efficient mask sorting proposed according to the embodiments of the present invention will be described with reference to the accompanying drawings.

[0050] As Figure 1As shown in the figure, the single-pixel imaging system provided in this embodiment includes:

[0051] An encoding and modulation module. After a series of collimations, the light source passes through an optical mask template carrying object information, projects the incident light with object information onto a spatial light modulator, and sequentially loads the generated high-efficiency modulation mask patterns onto the spatial light modulator to optically encode the projected object information.

[0052] An information acquisition module. A single-pixel detector detects the output light field modulated by the mask pattern, receives the output signal of the single-pixel imaging system, acquires optical intensity information less than the Nyquist sampling theorem, and digitizes the signal by a data acquisition card and transmits it to a computer for storage.

[0053] An image reconstruction module. An imaging optimization model under undersampling conditions is established by the compressive sensing theory. The data fidelity term and the prior regularization term are decomposed into two sub-problems using the Alternating Direction Method of Multiplier (ADMM). Combining the Plug-and-Play (PnP) algorithm, the solution of the sub-problem of the prior regularization term is replaced by solving an image denoising problem. Finally, the sub-problems of the data fidelity term and the prior regularization term are alternately solved to achieve high-quality image reconstruction at a low sampling rate.

[0054] The relationship among the mask modulation pattern, the target object, and the measurement values of the single-pixel detector in the above single-pixel imaging system is:

[0055] Ax = y (1)

[0056] Where represents the n-pixel target image to be reconstructed, represents m mask modulation patterns, each pattern consisting of n pixels, represents the m measurement values detected by the single-pixel detector.

[0057] Such as Figure 2 As shown in the figure, in a preferred embodiment of the present invention, the generation method of the mask pattern selected and loaded by the encoding and modulation module is as follows:

[0058]

[0059] Where is a set of one-dimensional basis vectors recursively expanded from the initial seed vector s, α k v is the basis vector v multiplied by the value of α k The value of α = α1,..., α k ∈ {+1, -1} uniquely defines Each basis vector in it is the α-index of v. When the Hamming distance between the α-indices of two basis vectors is 1, the two vectors are α-related. The set of continuously α-related vectors is the Gray code sequence (GCS). Its two-dimensional form is as follows:

[0060]

[0061] Where v1×v2 is the outer product of the one-dimensional basis vectors v1 and v2. Applying the concept of GCS to the two-dimensional version generates a two-dimensional Hadamard basis mask large square matrix. Using a zigzag traversal to extract the Hadamard basis masks in this matrix generates the required mask pattern. This is the Efficient Mask Sorting (EMS) for improving the system information capture efficiency preferably provided by the embodiments of the present invention. It should be noted that this implementation method of this scheme currently achieves the optimal effect of EMS+PnP, but the proposed EMS scheme is not limited to this specific implementation method. Various equivalent replacements of the above steps (such as other efficient measurement matrix sorting schemes with approximate effects) also fall within the protection scope of the present invention.

[0062] Finally, each basis mask is split into two complementary binary matrices by the following formula to implement the hardware adaptation strategy and noise suppression:

[0063]

[0064] Where H + The +1 in it is mapped to 1, and -1 is mapped to 0; for H - The -1 in it is mapped to 1, and +1 is mapped to 0. The measured value is obtained by subtracting the measured value encoded by H + from the measured value encoded by H -

[0065] The above EMS method is generally implemented by a computer.

[0066] As Figure 3 shown, in a preferred embodiment of the present invention, the imaging optimization model under the undersampling condition of the image reconstruction module is as follows:

[0067]

[0068] Where f(x) is the data fidelity term determined by the single-pixel imaging measurement model in formula (1), g(x) is the regularization term, which contains some prior information of the unknown image, and λ is the regularization parameter, used to constrain the solution in the ideal signal space. Introducing the auxiliary parameter z transforms the unconstrained optimization problem into a constrained optimization problem:

[0069] ​

[0070] wherein represents an n-pixel target image to be reconstructed, represents m mask modulation patterns, each pattern consisting of n pixels, represents m measurement values detected by single-pixel detection. By introducing auxiliary variables z and w, the augmented Lagrangian function of formula (4) can be expressed as:

[0071]

[0072] Let u = w / ρ, then the iterative steps of ADMM can be divided into the update steps of the following three sub-problems:

[0073]

[0074]

[0075] u (k+1) = u (k) - (x (k+1) - z (k+1) ) (11)

[0076] In a preferred embodiment of the present invention, the solution processes of the data fidelity term sub-problem and the prior regularization term sub-problem of the image reconstruction module are as follows:

[0077] The data fidelity term sub-problem is in a quadratic form and has a closed-loop solution. Specifically, given z (k) , the Euclidean projection of z (k) onto the linear manifold:

[0078] x (k+1) = (A T A + ρI) -1 (A T y + ρz (k) ) (12)

[0079] The prior regularization term sub-problem is a denoising problem for x (k+1) - u (k) . By combining the PnP algorithm and the ADMM algorithm, the update can be replaced by an image denoiser:

[0080]

[0081] wherein is an image denoiser, σ is a noise estimation parameter dependent on λ / ρ, ρ is a noise penalty coefficient, and the value of ρ is adjusted to match the noise level introduced in the compressive measurements. The deep denoising network adopts the SCU-Net network. Specifically, the SCU-Net network is trained on a training dataset composed of the Waterloo Exploration Database, DIV2K, and Flick2K to endow the SCU-Net network with denoising ability.

[0082] In a preferred embodiment of the present invention, the reconstruction process of the image reconstruction module is as follows: The data fidelity term sub-problem of formula (9), the image denoising sub-problem of the prior regularization term of formula (10), and the dual variable of formula (11) are alternately iteratively updated until the error is within a set threshold to obtain a reconstructed high-quality image. The measurement values in the image reconstruction process are obtained by encoding and measuring the high-efficiency mask sorting loaded by the spatial light modulator, and the optimization iteration is solved iteratively by the PnP-ADMM method.

[0083] The above image reconstruction process is generally implemented by a computer.

[0084] The effects of the embodiments are verified through numerical simulation and optical experiments.

[0085] 1. Numerical simulation results

[0086] To verify the image reconstruction performance of the proposed plug-and-play single-pixel imaging method based on high-efficiency mask sorting in a noise-free environment with different sampling rates and in a noisy environment, the Set12 dataset is used as the target image, and horizontal and vertical comparisons are made with the results of reconstruction using different coding methods and traditional decoding algorithms (using binary sorting coding and decoding with the TVAL3 algorithm, using cake-cutting sorting coding and decoding with the TVAL3 algorithm, using GCSS sorting coding and decoding with the TVAL3 algorithm, using random pattern coding and decoding with the PnP algorithm). The peak signal-to-noise ratio and structural similarity index are used to quantitatively measure the reconstruction performance, and the peak signal-to-noise ratio and structural similarity index of the reconstructed images of each method are averaged as the performance of this method at this sampling rate. The image size is 64×64, the typical sampling rate is set from 0.2 to 0.9, the very low sampling rate is set from 0.02 to 0.09, and the sampling rate in the noisy environment is set to 0.5.

[0087] Table 1 Comparison of average PSNR of all images in single-pixel reconstruction on the Set12 dataset (sampling rate from 0.2 to 0.9)

[0088]

[0089] Table 2 Comparison of average SSIM of all images for single-pixel reconstruction on the Set12 dataset (sampling rate from 0.2 to 0.9)

[0090]

[0091]

[0092] As can be seen from Table 1 and Table 2, the reconstruction performance of the proposed plug-and-play single-pixel imaging method based on efficient mask sorting in the present invention is significantly better than the other four comparison methods. The reconstructed images of the "Cameraman" image in the Set12 dataset under different typical sampling rates of different methods are as Figure 4 shown. It can be seen that the image reconstruction performance of the present invention is better than the other four comparison methods at all sampling rates, especially at low sampling rates.

[0093] Table 3 Comparison of average PSNR of all images for single-pixel reconstruction on the Set12 dataset (sampling rate from 0.02 to 0.09)

[0094]

[0095] Table 4 Comparison of average SSIM of all images for single-pixel reconstruction on the Set12 dataset (sampling rate from 0.02 to 0.09)

[0096]

[0097]

[0098] As can be seen from Table 3 and Table 4, the reconstruction performance of the proposed plug-and-play single-pixel imaging method based on efficient mask sorting in the present invention is significantly better than the other four comparison methods. The reconstructed images of the "House" image in the Set12 dataset under different extremely low sampling rates of different methods are as Figure 5 shown. It can be seen that the image reconstruction performance of the present invention is better than the other four comparison methods at all sampling rates, especially at extremely low sampling rates.

[0099] Table 5 Comparison of average PSNR of all images for single-pixel reconstruction on the Set12 dataset under noisy environment (sampling rate 0.5)

[0100]

[0101] Table 6 Comparison of average SSIM of all images for single-pixel reconstruction on the Set12 dataset under noisy environment (sampling rate 0.5)

[0102]

[0103]

[0104] As can be seen from Table 5 and Table 6, the reconstruction performance of the plug-and-play single-pixel imaging method based on efficient mask sorting proposed by the present invention is significantly better than that of the other four comparison methods. The reconstructed images of the "Cameraman" image in the Set12 dataset under different methods and different noise levels are as Figure 6 shown. It can be seen that the image reconstruction performance of the present invention is better than that of the other four comparison methods at all noise levels, especially in a high-noise environment.

[0105] 2. Optical experimental results

[0106] To verify the image reconstruction performance of the plug-and-play single-pixel imaging method based on efficient mask sorting proposed by the present invention, a single-pixel imaging experimental optical path was built for experimental verification. The system optical path diagram is as Figure 7 shown. Among them, the number of pixels of the digital micromirror device is 1920×1080 pixels, the model of the single-pixel detector is Thorlabs (PDA100A2), and the model of the data acquisition card is NI (USB-6211). The plug-and-play single-pixel imaging method based on efficient mask sorting proposed by the present invention was compared with the reconstruction results of typical binary sorting coding and decoding using the TVAL3 algorithm, cake-cutting sorting coding and decoding using the TVAL3 algorithm, GCSS sorting coding and decoding using the TVAL3 algorithm, and random pattern coding and PnP algorithm decoding in the actual single-pixel imaging experiment, as Figure 8 shown.

[0107] Figure 8 (a) shows the single-pixel imaging results of the "Fu" target image (64×64) under different typical sampling rates and different methods, Figure 8 (b) shows the single-pixel imaging results of the "Fu" target image (64×64) under different extremely low sampling rates and different methods. It can be seen that the plug-and-play single-pixel imaging method based on efficient mask sorting proposed by the present invention is superior to other algorithms in terms of reconstruction quality, especially at extremely low sampling rates.

[0108] It should be noted that, unless otherwise defined, the technical terms or scientific terms used in the present invention shall have the ordinary meanings understood by those of ordinary skill in the field to which the present invention pertains. The "first", "second" and similar terms used in the present invention do not denote any order, quantity or importance, but are only used to distinguish different components. Words such as "comprising" or "including" mean that the elements or objects appearing before this word cover the elements or objects listed after this word and their equivalents, without excluding other elements or objects. Words such as "connected" or "coupled" are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. "Upper", "lower", "left", "right", etc. are only used to indicate relative position relationships, and when the absolute position of the object being described changes, the relative position relationship may also change accordingly.

[0109] As described above, it is only the preferred embodiment of the present invention, and it is not a limitation to the present invention in other forms. Any person skilled in the art may use the technical content disclosed above 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.

[0110] This patent is not limited to the above best implementation mode. Anyone inspired by this patent can obtain various other forms of a plug-and-play single-pixel imaging method and system based on efficient mask sorting. All equal changes and modifications made according to the scope of the patent application of the present invention shall fall within the scope covered by this patent.

Claims

1. A plug-and-play single-pixel imaging method based on efficient mask sorting, characterized in that: The measured values ​​in the image reconstruction process are obtained by encoding the spatial light modulator with efficient mask sorting, and the optimization iteration is iteratively solved by the PnP-ADMM framework.

2. The plug-and-play single-pixel imaging method based on efficient mask sorting according to claim 1, characterized in that: The encoding by loading an efficient mask sequence on a spatial light modulator is specifically as follows: the light source undergoes a series of collimations and then passes through an optical mask plate carrying object information, the incident light carrying the object information is projected onto the spatial light modulator, and the generated efficient modulation mask patterns are sequentially loaded onto the spatial light modulator to optically encode the projected object information.

3. The plug-and-play single-pixel imaging method based on efficient mask sorting according to claim 2, characterized in that: The mask pattern is generated as follows: in is the set of one-dimensional basis vectors recursively expanded from the initial seed vector s, α k v is the basis vector v multiplied by α k The value of α=α1,…,α k ∈{+1,-1} uniquely defines Each basis vector in is the α-index of v. When the Hamming distance of the α-index of two basis vectors is 1, the two vectors are α-correlated. The set of continuous α-correlated is the Gray code sequence. The two-dimensional form of is: Among them, v1×v2 is the outer product of the one-dimensional basis vectors v1 and v2, which is used to generate a two-dimensional Hadamard basis mask large square matrix. The Hadamard basis mask in the square matrix is ​​extracted using a zigzag traversal to generate the required mask pattern.

4. The plug-and-play single-pixel imaging method based on efficient mask sorting according to claim 3, characterized in that: After the mask pattern is generated, each base mask is split into two complementary binary matrices by the following formula: Among them, H + +1 is mapped to 1, -1 is mapped to 0; H - The -1 in the H is mapped to 1 and the +1 is mapped to 0. The measured value is + The encoded measurement value minus H - The encoded measurements are obtained.

5. The plug-and-play single-pixel imaging method based on efficient mask sorting according to claim 1, characterized in that: The image reconstruction is based on compressed sensing theory to establish an imaging optimization model under undersampling conditions, and uses the alternating direction multiplier method ADMM to decompose the data fidelity term and the prior regularization term into two sub-problems. The plug-and-play PnP algorithm is combined to replace the solution of the sub-problem of the prior regularization term with the solution of the image denoising problem. Finally, the data fidelity term sub-problem and the prior regularization term sub-problem are alternately solved to achieve high-quality image reconstruction under low sampling rate.

6. A plug-and-play single-pixel imaging system based on efficient mask sorting, characterized in that: include: Coding and modulation module, information acquisition module and image reconstruction module; the measurement value of the image reconstruction module during the image reconstruction process is encoded by the spatial light modulator of the coding and modulation module loading the efficient mask sorting and is measured by the information acquisition module, and the optimization iteration is iteratively solved by the PnP-ADMM framework.

7. The plug-and-play single-pixel imaging system based on efficient mask sorting according to claim 6, characterized in that: The mask generation and image reconstruction process are implemented by computer.

8. The plug-and-play single-pixel imaging system based on efficient mask sorting according to claim 6, characterized in that: The measurement values ​​of the image reconstruction module during the image reconstruction process are encoded by the spatial light modulator of the coding and modulation module by loading the efficient mask sequence: the light source passes through an optical mask plate carrying object information after a series of collimations, and the incident light with the object information is projected onto the spatial light modulator, and the generated efficient modulation mask pattern is sequentially loaded onto the spatial light modulator to optically encode the projected object information.

9. The plug-and-play single-pixel imaging system based on efficient mask sorting according to claim 8, characterized in that: The mask pattern is generated as follows: in is the set of one-dimensional basis vectors recursively expanded from the initial seed vector s, α k v is the basis vector v multiplied by α k The value of α=α1,…,α k ∈{+1,-1} uniquely defines Each basis vector in is the α-index of v. When the Hamming distance of the α-index of two basis vectors is 1, the two vectors are α-correlated. The set of continuous α-correlated is the Gray code sequence. The two-dimensional form of is: Among them, v1×v2 is the outer product of the one-dimensional basis vectors v1 and v2, which is used to generate a two-dimensional Hadamard basis mask large square matrix. The Hadamard basis mask in the square matrix is ​​extracted using a zigzag traversal to generate the required mask pattern.

10. The plug-and-play single-pixel imaging system based on efficient mask sorting according to claim 9, characterized in that: After the mask pattern is generated, each base mask is split into two complementary binary matrices by the following formula: Among them, H + +1 is mapped to 1, -1 is mapped to 0; H - The -1 in the H is mapped to 1 and the +1 is mapped to 0. The measured value is + The encoded measurement value minus H - The encoded measurements are obtained.

Images