Near-infrared single-pixel front denoising imaging method based on source estimation and related device

Abstract

The embodiment of the invention relates to the technical field of data processing and imaging, and provides a near-infrared single-pixel front denoising imaging method based on source estimation and a related device, and the method comprises the steps: obtaining first light intensity information which is collected by a target detector and is reflected by projecting a speckle matrix to a target object, exporting the first light intensity information as a noise-containing sampling signal; based on a source number estimation module, carrying out analysis and decomposition processing on the noise-containing sampling signal to obtain an independent source number; determining a detector group corresponding to the number of the independent sources according to the number of the independent sources; acquiring second light intensity information which is acquired by the detector group and reflected by projecting the speckle matrix onto the target object, and exporting the second light intensity information as a multi-channel signal; based on a blind source separation module, carrying out separation and extraction processing on the multi-channel signal to obtain a target sampling signal; and on the basis of the reconstruction module, two-dimensional reconstruction processing is performed according to the target sampling signal to obtain the target object image corresponding to the target object, so that the imaging accuracy can be improved.

Description

Technical Field

[0001] This application relates to the fields of data processing and imaging technology, and particularly relates to a near-infrared single-pixel pre-denoising imaging method and related device based on source estimation. Background Art

[0002] With the rapid development of modern technology, imaging technology, as an important means of information acquisition, plays a crucial role in many fields. Traditional line and area array imaging technologies often face problems such as high cost, complex hardware, and large volume. Currently, a near-infrared single-pixel imaging technology with low cost, simple device, and strong penetration has been introduced. Single-pixel imaging technology is a technology that uses a detector without spatial resolution ability to collect the light intensity information reflected from a target object irradiated by a coded pattern for image reconstruction. However, in most existing studies, the near-infrared single-pixel imaging technology optimizes the modulation matrix and updates the reconstruction algorithm under noise-free and interference-free conditions. However, the actual imaging environment is not noise-free. In a near-infrared single-pixel imaging system, the main source of noise is the noise of the detector. Therefore, how to achieve more accurate imaging in the actual environment with noise has become an urgent problem to be solved. Summary of the Invention

[0003] The embodiments of this application provide a near-infrared single-pixel pre-denoising imaging method and related device based on source estimation, which can reduce the influence of noise in the imaging process, improve the accuracy of image reconstruction, obtain a more accurate image of the target object, and can improve the accuracy of imaging.

[0004] The first aspect of the embodiments of this application provides a near-infrared single-pixel pre-denoising imaging method based on source estimation, and the method includes:

[0005] Obtain the first light intensity information reflected from the speckle matrix projected onto the target object collected by the target detector, and export the first light intensity information as a noisy sampling signal;

[0006] Based on the source number estimation module, perform analytical decomposition processing on the noisy sampling signal to obtain the number of independent sources;

[0007] According to the number of independent sources, determine the detector group corresponding to the number of independent sources;

[0008] Obtain the second light intensity information reflected from the speckle matrix projected onto the target object collected by the detector group, and export the second light intensity information as a multi-channel signal;

[0009] Based on the blind source separation module, perform separation and extraction processing on the multi-channel signal to obtain the target sampling signal;

[0010] Based on a reconstruction module, perform two-dimensional reconstruction processing according to the target adopted signal to obtain the target object image corresponding to the target object.

[0011] In a second aspect of the embodiments of the present application, a near-infrared single-pixel pre-denoising imaging device based on source estimation is provided. The device includes:

[0012] A first acquisition unit, configured to acquire first light intensity information reflected by projecting a speckle matrix onto a target object collected by a target detector, and export the first light intensity information as a noisy sampling signal;

[0013] A first processing unit, configured to perform analytical decomposition processing on the noisy sampling signal based on a source number estimation module to obtain an independent source number;

[0014] A first determination unit, configured to determine a detector group corresponding to the independent source number according to the independent source number;

[0015] A second acquisition unit, configured to acquire second light intensity information reflected by projecting the speckle matrix onto the target object collected by the detector group, and export the second light intensity information as a multi-channel signal;

[0016] A second processing unit, configured to perform separation and extraction processing on the multi-channel signal based on a blind source separation module to obtain a target sampling signal;

[0017] A third processing unit, configured to perform two-dimensional reconstruction processing according to the target adopted signal based on a reconstruction module to obtain the target object image corresponding to the target object.

[0018] In a third aspect of the embodiments of the present application, a terminal is provided, including a processor, an input device, an output device, and a memory. The processor, input device, output device, and memory are interconnected. Among them, the memory is used to store a computer program, and the computer program includes program instructions. The processor is configured to call the program instructions to execute the step instructions in the first aspect of the embodiments of the present application.

[0019] In a fourth aspect of the embodiments of the present application, a computer-readable storage medium is provided. The computer-readable storage medium stores a computer program for electronic data exchange. The computer program enables a computer to execute some or all of the steps described in the first aspect of the embodiments of the present application.

[0020] A fifth aspect of the embodiments of the present application provides a computer program product. The computer program product includes a non-transitory computer-readable storage medium storing a computer program, and the computer program is operable to cause a computer to execute some or all of the steps described in the first aspect of the embodiments of the present application. The computer program product may be a software installation package.

[0021] Implementing the embodiments of the present application has the following beneficial effects:

[0022] By obtaining the first light intensity information reflected by projecting a speckle matrix onto a target object collected by a target detector and exporting the first light intensity information as a noisy sampling signal, it is possible to further perform analytical decomposition processing on the noisy sampling signal based on a source number estimation module to obtain the number of independent sources, and based on the number of independent sources, determine a detector group corresponding to the number of independent sources, thereby obtaining the second light intensity information reflected by projecting the speckle matrix onto the target object collected by the detector group and exporting the second light intensity information as a multi-channel signal. Further, based on a blind source separation module, perform separation and extraction processing on the multi-channel signal to obtain a target sampling signal, and then based on a reconstruction module, perform two-dimensional reconstruction processing according to the target sampling signal to obtain a target object image corresponding to the target object, and a more accurate target object image can be obtained, which can improve the accuracy of imaging. Description of the Drawings

[0023] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the following drawings are only some embodiments of the present application. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0024] Figure 1 It is a schematic flowchart of a near-infrared single-pixel imaging technology provided by the embodiments of the present application;

[0025] Figure 2 It is a schematic flowchart of a near-infrared single-pixel pre-denoising imaging method based on source estimation provided by the embodiments of the present application;

[0026] Figure 3 It is a schematic diagram of a speckle matrix provided by the embodiments of the present application;

[0027] Figure 4 It is a schematic flowchart of an adaptive empirical mode decomposition module provided by the embodiments of the present application;

[0028] Figure 5 It is a schematic flowchart of a singular value decomposition module provided by the embodiments of the present application;

[0029] Figure 6 It is a schematic flowchart of a blind source separation module provided by an embodiment of the present application;

[0030] Figure 7 It is a schematic flowchart of compressive sensing provided by an embodiment of the present application;

[0031] Figure 8 It is a schematic flowchart of a near-infrared single-pixel pre-denoising imaging device based on source estimation provided by an embodiment of the present application;

[0032] Figure 9 It is a schematic structural diagram of a terminal provided by an embodiment of the present application;

[0033] Figure 10 It is a schematic structural diagram of a near-infrared single-pixel pre-denoising imaging device based on source estimation provided by an embodiment of the present application. Specific embodiments

[0034] Next, the technical solutions in the embodiments of the present application will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present application.

[0035] The terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish different objects, rather than to describe a specific order. In addition, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device that includes a series of steps or units is not limited to the listed steps or units, but optionally further includes steps or units not listed, or optionally further includes other steps or units inherent to these processes, methods, products or devices.

[0036] Referring to "embodiment" in the present application means that a specific feature, structure or characteristic described in conjunction with the embodiment may be included in at least one embodiment of the present application. The phrase appears in various positions in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment mutually exclusive with other embodiments. Those skilled in the art explicitly and implicitly understand that the embodiments described in the present application may be combined with other embodiments.

[0037] To better understand a near-infrared single-pixel pre-denoising imaging method provided by an embodiment of the present application based on source estimation, the following first briefly introduces the near-infrared single-pixel pre-denoising imaging method in the existing solution. In the existing solution, traditional line and area array imaging technologies often face problems such as high cost, complex hardware, and large volume. Therefore, a near-infrared single-pixel imaging technology with low cost, simple device, and strong penetration is introduced. Single-pixel imaging technology is a technology that uses a detector without spatial resolution ability to collect the light intensity information reflected by a coded pattern irradiated on a target object to achieve image reconstruction. However, in most existing studies, the near-infrared single-pixel imaging technology optimizes the modulation matrix and updates the reconstruction algorithm under noise-free and interference-free conditions to improve the image reconstruction accuracy. However, the actual imaging environment is not noise-free. In a near-infrared single-pixel imaging system, the main source of noise is the noise of the detector.

[0038] Aiming at solving the above problems, an embodiment of the present application provides a near-infrared single-pixel pre-denoising imaging method based on source estimation, which can obtain a cleaner signal source through separating and analyzing the noisy sampling signal and blind denoising processing before imaging, providing sufficient preprocessing work for subsequent image reconstruction, so as to reconstruct a clearer and more accurate image.

[0039] Please refer to Figure 1 , Figure 1 which shows a schematic flow diagram of a near-infrared single-pixel imaging technology. As Figure 1 shown, first, a near-infrared light source is used to irradiate the target object to be imaged. The object scatters and absorbs the light emitted by the light source, forming a light field containing object information. Then, the light field is modulated by a spatial light modulator. The rows of the speckle matrix are loaded on the spatial light modulator, and only a part of the light field is allowed to pass through each time. After the modulated light field passes through the spatial light modulator, the near-infrared single-pixel detector collects its light intensity information, and the image of the object is reconstructed by combining the light intensity values corresponding to the spatial light modulator modes with the compressive sensing algorithm.

[0040] Please refer to Figure 2 , Figure 2 which is a schematic flow diagram of a near-infrared single-pixel pre-denoising imaging method based on source estimation provided by an embodiment of the present application. As Figure 2 shown, the near-infrared single-pixel pre-denoising imaging method based on source estimation includes:

[0041] 201. Obtain the first light intensity information reflected by projecting the speckle matrix onto the target object collected by the target detector, and export the first light intensity information as a noisy sampling signal.

[0042] Among them, the target detector can be a detection instrument for collecting light intensity information. Optionally, the target detector can be a near-infrared single-pixel detector, and the present application does not limit this. The speckle matrix can be a measurement matrix used to encode the spatial information of an object in single-pixel imaging technology. By optically mixing the target object with the speckle pattern, the overall light intensity information of the target object can be captured, and by combining the compressive sensing algorithm, the reconstruction of the target object image can be achieved.

[0043] Please refer to Figure 3 , Figure 3 which is a schematic diagram of a speckle matrix provided by an embodiment of the present application. As Figure 3 shown, the present application introduces a sliced-cake Hadamard matrix as the measurement matrix (i.e., the speckle matrix), which is an orthogonal matrix composed of +1 and -1, and each row is the Hadamard transform of other rows. Among them, the distribution characteristics of the sliced-cake Hadamard matrix are as follows:

[0044]

[0045] where CC(H n ) represents an n-order sliced-cake Hadamard matrix, n is the order of the matrix, CC(H n-1 ) represents an n - 1-order sliced-cake Hadamard matrix, H n represents an n×n Hadamard matrix, and H n-1 represents an (n - 1)×(n - 1) Hadamard matrix.

[0046] By using the sliced-cake Hadamard matrix as the measurement matrix, the measurement process can be simplified, and due to its orthogonality and uniformity, it helps to reconstruct a high-quality image with fewer measurement times. Optionally, the present application will take the speckle matrix as the sliced-cake Hadamard matrix as an example for illustration later, which does not limit the present application.

[0047] The target detector collects the light intensity information reflected by the speckle matrix projected onto the target object, that is, the first light intensity information. In this embodiment, in the process of collecting the light intensity information (i.e., the first light intensity information) returned by the speckle matrix irradiated on the target object through a near-infrared single-pixel detector, the sampling signal can include one or more groups, and one group of signals can correspond to one detector. After being converted by a data collector into an input data format applicable to the present application (i.e., a noisy sampling signal). It should be noted that at this time, the collected sampling signals are not all effective object-reflected light intensity information, and it can include noise in the detector, and the present application does not limit this.

[0048] 202. Based on the source number estimation module, perform an analysis and decomposition process on the noisy sampling signal to obtain the independent source number.

[0049] Among them, the source number estimation module, also known as the source estimation module, can be used to perform analytical decomposition processing on the noisy sampling signal. This source number estimation module can use methods such as adaptive empirical mode decomposition and singular value decomposition to decompose the noisy sampling signal into different components, analyze the signal characteristics, and determine the number of independent sources, that is, the independent source number (also known as the source number) in the mixed signal by finding the demarcation point between the signal and the noise. The independent source number can be used to reasonably set the number of detectors in the subsequent process, which is conducive to achieving a more effective signal-to-noise separation purpose.

[0050] In this embodiment, the source estimation module separates and analyzes the noisy sampling sequence collected by the near-infrared detector. This source estimation module can include an improved complete ensemble empirical mode decomposition with adaptive noise module and a singular value decomposition module. Among them, the improved complete ensemble empirical mode decomposition with adaptive noise module can decompose the sampling sequence into intrinsic modes, that is, intrinsic mode functions and residual signals, and combine them into a matrix, and then further calculate its covariance matrix to import it into the singular value decomposition module to decompose it into a diagonal matrix. Then, the number of independent sources is determined according to the adjacent singular value difference sequence.

[0051] 203. Determine a detector group corresponding to the number of independent sources according to the number of independent sources.

[0052] According to the number of independent sources obtained by the source number estimation module, the corresponding detector group can be determined. The number of detectors in this detector group can be equal to the number of independent sources, and each detector can be used to collect the light intensity information reflected by the speckle matrix projected onto the target object. By setting the detector group, more comprehensive light intensity information can be obtained, providing data support for more accurate signal and noise separation in the subsequent steps.

[0053] 204. Obtain the second light intensity information reflected by projecting the speckle matrix onto the target object collected by the detector group, and export the second light intensity information as a multi-channel signal.

[0054] The detector group collects the light intensity information reflected by the speckle matrix projected onto the target object, that is, the second light intensity information. By separately exporting and combining the light intensity information collected by multiple detectors in the detector group, a multi-channel signal can be formed. It should be noted that compared with the first light intensity information, the multi-channel signal can contain more-dimensional data, can more comprehensively reflect the information of the target object, and provides a richer data basis for the subsequent blind source separation step.

[0055] 205. Based on the blind source separation module, perform separation and extraction processing on the multi-channel signal to obtain the target sampling signal.

[0056] The blind source separation module can process multi-channel signals. That is to say, without knowing the original signals and the mixing method, the blind source separation module can separate the valid signals and noise in the mixed signals. Specifically, through techniques such as the independent component analysis optimized by particle swarm optimization, the valid sampling signals, that is, the target sampling signals, can be extracted from the multi-channel signals. The target sampling signal can be understood as a sampling signal that has removed most of the noise and retained the key information of the target object, which can provide guarantee for accurately reconstructing the image of the target object subsequently.

[0057] In this embodiment, the pre-denoising can be implemented by the blind source separation part, and the separation method based on independent component analysis has been optimized by particle swarm optimization. The whole optimization process scheme is clear, with high speed and good effect. The blind source separation module is mainly composed of the independent component analysis optimized by particle swarm optimization. This module inputs the multi-channel signals collected by multiple detectors and outputs an estimated source matrix. One of the sequences of the estimated source matrix can be the denoised sampling sequence, and the others can be the separated noise signals. Among them, the detectors collect the reflected light beams in the near-infrared band filtered and converged by the lens. The number of detectors and the number of channels can be equal to the number of estimated sources in the previous part, and this application does not limit this.

[0058] 206. Based on the reconstruction module, perform two-dimensional reconstruction processing according to the target sampling signal to obtain the target object image corresponding to the target object.

[0059] Based on the target sampling signal, the reconstruction module can perform two-dimensional reconstruction processing using algorithms such as compressive sensing. Specifically, through the analysis and operation of the target sampling signal, the image of the target object, that is, the target object image, can be reconstructed. During the reconstruction process, techniques such as the total variation regularization algorithm can be used to solve the target convex function under the constraint conditions, so as to effectively reconstruct the edge information of the target object and improve the clarity and accuracy of the image.

[0060] In this embodiment, the separated valid sampling sequence is input into the image reconstruction algorithm. The reconstruction algorithm is the most important step in the compressive sensing framework. This application uses the reconstruction algorithm to obtain the non-linear optimization solution of the target convex function under the constraint conditions, so as to achieve the purpose of reconstructing the edge information of the target object.

[0061] It can be seen that the present application supports image reconstruction of a target object in a low-light environment, and the overall volume and cost of the device are significantly lower than those of traditional line and area array sensor imaging technologies. The near-infrared single-pixel imaging device uses one pixel to sequentially record the light intensity information of the scene. This technology typically involves using a digital micromirror device to encode the light information of the scene, and then collecting it through a single-pixel detector to achieve the maximum performance and energy efficiency ratio. Secondly, the present application uses a source estimation model to accurately estimate the number of sources in the mixed signal. The mixed signal is usually composed of multiple signal sources through linear or nonlinear mixing. Importing the noisy sampling sequence into the source estimation model can obtain the adjacent singular value difference sequence. Then, the boundary space between the signal and the noise is found to infer the number of sources. Further, the present application proposes a denoising scheme based on single-pixel detection. This scheme does not traditionally perform image filtering and image enhancement after imaging, but processes from the source, purifies the signal and then reconstructs it, with extremely high performance.

[0062] In a possible implementation, after collecting the corresponding sampling sequence, in the improved adaptive noise complete ensemble empirical mode decomposition module, the separation of non-linear and non-stationary sequence data can be realized, and the original signal can be effectively separated into multiple intrinsic mode signals and a residual signal, and further singular value decomposition processing is performed through the singular value decomposition module to determine the number of independent sources. Specifically, a method for parsing and decomposing the noisy sampling signal based on the source number estimation module to obtain the number of independent sources includes:

[0063] A1. Based on the adaptive empirical mode decomposition module in the source number estimation module, decomposing the noisy sampling signal into intrinsic mode signals and a residual signal;

[0064] A2. Based on the source number estimation module, combining the noisy sampling signal, the intrinsic mode signal and the residual signal into a matrix to obtain virtual multi-channel data;

[0065] A3. Based on the singular value decomposition module in the source number estimation module, performing singular value decomposition on the covariance matrix of the virtual multi-channel data to obtain an adjacent singular value difference sequence;

[0066] A4. Determining the number of independent sources according to the adjacent singular value sequence.

[0067] Among them, the adaptive empirical mode decomposition module can be understood as a key component in the source number estimation module, and can be used to process nonlinear and non-stationary noisy sampling signals. The adaptive empirical mode decomposition module can decompose the complex noisy sampling signal into multiple intrinsic mode signals (also called intrinsic mode functions) (IMF) and residual signals by iteratively detecting the local maximum and minimum values of the signal in sequence and calculating the envelope of the signal. Among them, the intrinsic mode signal represents the fluctuation characteristics of the signal on different time scales, and the residual signal reflects the part of the signal that cannot be fully described by the intrinsic mode signal. This decomposition method can adaptively separate different components based on the characteristics of the signal itself, providing a basis for subsequent analysis.

[0068] It should be noted that the intrinsic mode signal may include one or more IMF components, each of which may have specific frequency characteristics and physical meanings. In the time series, the number of its zero crossing points is equal to the number of extreme points or differs by at most one, and the upper and lower envelopes are locally symmetric about the time axis. They can reflect the fluctuation characteristics of the original signal from different scales, which helps to analyze the signal more carefully. The residual signal may contain the residual part of the trend component or noise in the original signal that cannot be captured by the intrinsic mode signal, and may reflect the information in the signal that is relatively stable or difficult to accurately represent with the IMF component. This application does not limit this.

[0069] Specifically, in this embodiment, in the improved adaptive noise complete set empirical mode decomposition module (i.e., adaptive empirical mode decomposition module), the separation of nonlinear and non-stationary sequence data can be achieved, and the original signal can be effectively separated into multiple intrinsic mode signals and residual signals. The flow chart can be as follows: Figure 4 As shown. First, the adaptive empirical mode decomposition module can add adaptive noise, and the size of the adaptive noise is usually proportional to the amplitude of the input sampling signal to ensure that the addition of noise will not have too much impact on the characteristics of the signal; then, the signal with the added noise is subjected to multiple empirical mode decompositions, and each decomposition will generate a set of modal components and residual components. The residual signal continues to be subjected to empirical mode decomposition by adding adaptive noise until the stopping criterion (such as reaching the maximum number of iterations) is reached, and the decomposed modal components are summarized (such as averaging) to obtain the final result. The adaptive empirical mode decomposition module can effectively reduce the problem of mode mixing, and the adaptive empirical mode decomposition module improves the stability and repeatability of the decomposition and reduces the endpoint effect by adding different noises multiple times and taking the average.

[0070] Optionally, based on the adaptive empirical mode decomposition module in the source number estimation module, the process of decomposing the noisy sampling signal into intrinsic mode functions (IMFs) and a residual signal can be seen in the following formula:

[0071] I = IMF1 + IMF2 +... + IMF n + r

[0072] Where, I represents the noisy sampling signal; IMF1 represents the first decomposed intrinsic mode function; IMF2 represents the second decomposed intrinsic mode function; IMF n represents the nth decomposed intrinsic mode function; r is the residual signal.

[0073] The virtual multi-channel data can be understood as the matrix-form data formed by combining the noisy sampling signal, intrinsic mode functions, and residual signal. Through the above combination method, the original single noisy sampling signal can be expanded into virtual multi-channel data with multiple-dimensional information, which can increase the richness of the data and provide more comprehensive information for subsequent singular value decomposition, so as to better analyze the characteristics and correlations of the signal.

[0074] Optionally, based on the source number estimation module, the process of combining the intrinsic mode functions and the residual signal into a matrix to obtain virtual multi-channel data can be seen in the following formula:

[0075] X = [I; IMF1;...; IMF n ; r]

[0076] Where, X represents the virtual multi-channel data; I represents the original detected signal, that is, the noisy sampling signal; IMF1 represents the first decomposed intrinsic mode function; IMF n represents the nth decomposed intrinsic mode function; r is the residual signal.

[0077] In a possible implementation, the singular value decomposition module can be understood as a functional module that performs singular value decomposition on the covariance matrix of the virtual multi-channel data. Singular value decomposition is a matrix decomposition technique that can decompose a matrix into the product of three matrices, namely the left singular matrix, the singular value matrix (diagonal matrix), and the right singular matrix. In this process, the singular values represent the important features of the matrix, with larger singular values corresponding to the main components of the signal and smaller singular values usually related to noise. Through singular value decomposition, the main information and noise in the data can be effectively distinguished. Specifically, a method for performing singular value decomposition on the covariance matrix of the virtual multi-channel data based on the singular value decomposition module in the source number estimation module to obtain a sequence of differences between adjacent singular values includes:

[0078] B1. Determine the right matrix according to the covariance matrix of the virtual multi-channel data;

[0079] B2. Perform eigen-decomposition on the right matrix to obtain a right singular matrix;

[0080] B3. Determine a left matrix according to the covariance matrix of the virtual multi-channel data;

[0081] B4. Perform eigen-decomposition on the left matrix to obtain a left singular matrix;

[0082] B5. Perform singular value decomposition on the covariance matrix of the virtual multi-channel data according to the right singular matrix and the left singular matrix to obtain a singular value decomposition diagonal matrix;

[0083] B6. Perform singular value decomposition operation on the covariance matrix of the virtual multi-channel data according to the singular value decomposition diagonal matrix to obtain a sequence of adjacent singular value differences.

[0084] The virtual multi-channel data is a matrix composed of a noisy sampling signal, an intrinsic mode signal, and a residual signal. The covariance matrix of the virtual multi-channel data can be used to measure the correlation between each dimension of these data (i.e., the noisy sampling signal, the intrinsic mode signal, and the residual signal), and can reflect the mutual relationship between different signal components. By calculating the covariance matrix, the correlation information of the data can be quantified, providing a basis for analyzing the signal structure.

[0085] The right matrix can be understood as an intermediate matrix determined according to the covariance matrix of the virtual multi-channel data. The right singular matrix can be understood as the matrix obtained by performing eigen-decomposition on the right matrix. Eigen-decomposition can be to decompose a matrix into the product form of eigenvalues and eigenvectors, and the right singular matrix is composed of the eigenvectors of the right matrix.

[0086] The left matrix can be an intermediate matrix determined based on the covariance matrix of the virtual multi-channel data. Corresponding to the right matrix, it is an important link in the singular value decomposition process. The left singular matrix can be understood as obtained by performing eigen-decomposition on the left matrix and is composed of the eigenvectors of the left matrix. The left singular matrix cooperates with the right singular matrix in the singular value decomposition and can, together with the singular value matrix (diagonal matrix) in the subsequent singular decomposition process, decompose the original covariance matrix to reveal the internal characteristics of the data.

[0087] The singular value decomposition diagonal matrix (which can be simply referred to as the diagonal matrix, i.e., the aforementioned singular value matrix) can be understood as one of the results obtained by performing singular value decomposition on the covariance matrix of virtual multi-channel data based on the right singular matrix and the left singular matrix. It should be noted that the main diagonal elements of the singular value decomposition diagonal matrix are singular values, and these singular values are arranged in descending order, which can reflect the importance of the data in different characteristic directions. That is to say, larger singular values can correspond to the main signal components in the data, and smaller singular values can be related to noise or secondary information.

[0088] In this embodiment, the flowchart of the singular value decomposition module is as Figure 5 shown, where the signals of the decomposed intrinsic modes are recombined with the original sampled signals to form multi-channel X, and the covariance matrix A of X is obtained as the input. According to the covariance matrix A of X and the transpose of the covariance matrix A, the left matrix AA T is calculated, and the eigenvalues and eigenvectors of the left matrix AA T are calculated through eigenvalue decomposition, and the left singular matrix U is obtained after normalization. Also, according to the transpose of the covariance matrix A of X and the covariance matrix A, the right matrix A T A is calculated, and the eigenvalue decomposition is performed on the right matrix A T A to obtain eigenvalues and eigenvectors, which constitute the right singular matrix.

[0089] Specifically, eigenvalue decomposition is performed on the left matrix AA T and the right matrix A T A to calculate the eigenvalues of AA T and A T A and the corresponding eigenvectors. Then, the left singular matrix U and the right singular matrix V are constructed for singular value decomposition to obtain the diagonal matrix Λ. The process can be seen in the following formula:

[0090]

[0091] where A is the matrix to be decomposed, i.e., the covariance matrix of virtual multi-channel data; U is the left singular matrix, Λ is the singular value matrix, V is the right singular matrix; V T is the transpose of the right singular matrix.

[0092] That is to say, the diagonal matrix Λ can be seen in the following formula:

[0093]

[0094] where the elements other than the main diagonal are all zero, and the elements on the main diagonal are called singular values, λ1,Lλ n represents the 1st to the nth singular values.

[0095] Further, the diagonal elements in the diagonal matrix after singular value decomposition are arranged in descending order, and the differences between adjacent elements are further calculated. According to the information theory criterion, larger singular values usually correspond to signals, while smaller singular values correspond to noise. The points close to zero in the differences are the spatial boundaries of the source signal and the adaptive noise in the sampling sequence. We can determine how many significant singular values correspond to independent signal sources, and thus determine the number of sources.

[0096] The sequence of differences between adjacent singular values can be understood as the singular values obtained after singular value decomposition, arranged in descending order, and then the differences between adjacent singular values are calculated in turn. The sequence composed of these differences is the sequence of differences between adjacent singular values. Since there are differences in the magnitudes of the singular values corresponding to signals and noise, this sequence can reflect this change and provide a basis for determining the boundary point between signals and noise.

[0097] In the sequence of differences between adjacent singular values, when the difference is close to zero, it means a transition from the signal-dominated region to the noise-dominated region. By finding these points where the difference is close to zero, the number of singular values corresponding to the signal part can be determined, and this number is the number of independent sources. The number of independent sources reflects the number of independent signal components contained in the mixed signal, which is crucial for subsequent blind source separation and accurate image reconstruction.

[0098] In a possible implementation, pre-denoising is achieved by the blind source separation part, and the separation method based on independent component analysis is optimized by particle swarm optimization. The blind source separation module is mainly composed of independent component analysis optimized by particle swarm optimization. This module inputs the multi-channel signals collected by multiple detectors and outputs an estimated source matrix, where one sequence is the denoised sampling sequence and the others are the separated noise signals. Specifically, a method for separating and extracting the multi-channel signals based on the blind source separation module to obtain the target sampling signal includes:

[0099] C1. Based on the blind source separation module, perform initialization particle swarm processing on the multi-channel signals to obtain an initial particle set;

[0100] C2. Based on the blind source separation module, randomly generate the positions and velocities of each initial particle in the initial particle set to obtain an initial position set and an initial velocity set;

[0101] C3. Based on the blind source separation module, calculate the fitness of each initial particle in the initial particle set according to the initial position set and the initial velocity set to obtain a fitness set;

[0102] C4. Based on the blind source separation module, update the positions and velocities of each initial particle in the initial particle set according to each fitness in the fitness set, to obtain an updated position set and an updated velocity set;

[0103] C5. Based on the blind source separation module, perform an update process on the multi-channel signal according to the updated position set and the updated velocity set, to obtain an updated multi-channel signal;

[0104] C6. Based on the blind source separation module, perform an effective sampling signal screening process on the updated multi-channel signal, to obtain a target sampling signal.

[0105] Among them, the blind source separation module can be understood as a module in the entire imaging system that is responsible for separating effective sampling signals and noise from the mixed multi-channel signals. Its principle is to use an algorithm (such as the particle swarm optimization algorithm) to process the multi-channel signal without knowing the original signal and the mixing method, and through iterative optimization and other means, to achieve the separation of signals and noise, and provide high-quality sampling signals for subsequent image reconstruction.

[0106] The particle swarm optimization algorithm is the method adopted by the blind source separation module. When initializing the particle swarm processing, the number of the particle swarm will be set, and each particle represents a possible demixing matrix. These particles together form the initial particle set, which is the starting state of subsequent iterative optimization. The state (position and velocity) of each particle will be continuously adjusted in the subsequent process to find the optimal solution.

[0107] Randomly generate positions and velocities for each particle in the initial particle set. The positions of all particles form the initial position set, and the velocities of all particles form the initial velocity set. The initial positions and velocities determine the starting search directions of the particles in the solution space, and can provide a basis for subsequent calculation of particle fitness and update of particle states.

[0108] According to the initial positions and velocities of each particle, its fitness can be calculated. Fitness can be used to measure the quality of the solution (demixing matrix) represented by each particle for the separation effect of the multi-channel signal. Usually, the fitness will be calculated according to a specific objective function. For example, it is hoped that the separated signal can restore the real signal to the greatest extent and minimize the influence of noise. The fitness values of each particle form the fitness set, and by comparing the values in the fitness set, it is judged which particles are closer to the optimal solution.

[0109] According to each fitness value in the fitness set, update the positions and velocities of each particle in the initial particle set according to the rules of the particle swarm optimization algorithm. Generally, factors such as the particle's own historical optimal position and the global optimal position are referred to for adjustment. The updated particle positions form the updated position set, and the updated particle velocities form the updated velocity set. By continuously updating the particle positions and velocities, the particles gradually approach the optimal solution in the solution space, improving the signal separation effect.

[0110] Process the multi-channel signal using the particle states (demixing matrices) corresponding to the updated position set and the updated velocity set. These demixing matrices can act on the multi-channel signal to attempt to separate the signal and noise more accurately, thereby obtaining an updated multi-channel signal. The updated multi-channel signal has improved separation between the signal and noise compared to the original multi-channel signal.

[0111] Screen the updated multi-channel signal to find the valid sampling signal from the multiple separated signals. Methods such as sample entropy discrimination can be used. The valid signal has a certain regularity and structure, with a smaller sample entropy value, while the noise sequence is the opposite. The screened valid sampling signal is the target sampling signal, which removes most of the noise and contains the main information of the target object for subsequent two-dimensional reconstruction of the image.

[0112] In this embodiment, the flowchart of blind source separation is as Figure 6 shown. First, input the multi-channel signal into the blind source separation module, initialize the particle swarm, where each particle represents a possible demixing matrix. Then, set parameters (such as setting the fitness convergence value for stopping iteration). Next, randomly generate the positions and velocities of the particles, calculate the fitness of the particles, update the optimal positions and optimal velocities of each particle, and update the separated source matrix. Finally, determine whether the termination condition is reached (such as whether the fitness convergence value for stopping iteration is reached). If not, calculate and update the iteration. If reached (such as below the convergence value), stop the iteration and output the estimated source matrix.

[0113] The blind source separation module sets the number of detectors equal to the number of sources to collect the corresponding multi-channel sampling sequences for import to smoothly carry out the task of separating signals and noise. There is randomness in the signal separation process. Even if the input matrix is the same, different source matrices may be output, and moreover, the results of each separation may not exactly match. Therefore, the separation results show uncertainty and instability, affecting the quality of the two-dimensional reconstructed image of the target image. To address this problem, the independent component analysis module optimized by particle swarm is adopted in the present invention, aiming to achieve nearly complete separation of the signal and noise, thereby improving the stability and efficiency of blind source separation.

[0114] Particle swarm is an optimization algorithm based on swarm intelligence, which simulates the foraging behavior of bird flocks. It uses the cooperation and information sharing capabilities among individuals to find the optimal solution. After initializing the number of particle swarms, random positions, and random velocities, the positions and velocities of the particles are updated by calculating the fitness of each particle, and the iterative update is repeated until the optimal solution confusion matrix is output.

[0115] Optionally, the algorithm flow formula for particle swarm update iteration can be as follows:

[0116] P g (t) ∈ {P1(t), P2(t),..., P N (t)|f(P g (t)) = min{f(P1(t)), f(P2(t)),..., f(P N (t))}}

[0117] Among them, i = 1,..., N represents the number of current iterations, P g (t) is the corresponding optimal particle with the minimum fitness selected from P1(t), P2(t),..., P N (t), P1(t), P2(t),..., P N (t) represents the first, second to the Nth particles at the t-th moment, f(x) is the fitness function, and f(P1(t)), f(P2(t)),..., f(P N (t)) represents the fitness functions of the first, second to the Nth particles.

[0118] Optionally, the update steps are as follows:

[0119] a i (t + 1) = a i (t) + c1d1(P i - W i ) + c2d2(P g - W i )

[0120] Among them, a i (t + 1), a i (t) represent the velocities of the i-th particle at the (t + 1)-th and t-th iterations, c1 and c2 represent acceleration constants, c1 and c2 are random numbers uniformly distributed within 0 - 1, P i represents the optimal position of the i-th particle, W i is the position of the i-th particle, and P g is the global optimal position.

[0121] First, initialize the number of particle swarms k, including the random position W i and the random velocity a i, and satisfying 1 ≤ i ≤ k, calculate the fitness of each particle, update the position and velocity of the particle according to the expression until the judgment condition is reached, that is, the fitness converges to 10 -5 , if the judgment condition has been reached, the current optimal solution can be output. If the judgment condition cannot be reached, re-iterate and update until the judgment condition is satisfied.

[0122] Furthermore, the multi-channel sampled signals are separated and output as one-dimensional signals with the same number of channels. To perform the subsequent reconstruction work, a denoised sampling sequence needs to be found. Therefore, this application introduces a sample entropy discrimination method to screen the effective sampling signals. The smaller the value of the sample entropy, the more ordered the system is. After calculating the sample entropy of each separated one-dimensional signal, the sequence corresponding to the minimum value is the effective sampling signal, which is used for the two-dimensional reconstruction of the subsequent image.

[0123] The process of the sample entropy screening effective sequence module is as follows. The optimal solution unmixing matrix is the separated signal and noise sequences, both of which are one-dimensional sequences. To ensure the smooth progress of the subsequent reconstruction work, this application uses the sample entropy discrimination method for the unmixed signal and noise sequences. The sample entropy is based on the concept of approximate entropy and is a method for measuring the complexity of time series, with better consistency. Effective signals usually have a certain degree of regularity and structure, showing a certain complexity in the time series. The noise sequence is random and irregular, with low complexity and no useful information. Therefore, the sample entropy can measure the complexity of the time series and is calculated by comparing the similarity degree between subsequences of length m in the sequence and their subsequent subsequences. The specific calculation formula is as follows:

[0124]

[0125] where SE is the sample entropy value, C(m,r) is the proportion of similar vector pairs with a distance less than or equal to r between vectors of length m in the time series, m is the embedding dimension, generally taken as 2, and r is the similarity tolerance, usually taken as 0.2 times the standard deviation.

[0126] In a possible implementation, the reconstruction module performs an inverse operation on the one-dimensional sampled signal according to the speckle matrix information using the compressive sensing algorithm to obtain the estimated target image. Specifically, a method for performing two-dimensional reconstruction processing on the target sampled signal based on the reconstruction module to obtain the target object image corresponding to the target object includes:

[0127] Based on the reconstruction module, use the reconstruction algorithm to perform non-linear optimization and solution on the target convex function under the constraint conditions to obtain the target object image corresponding to the target object;

[0128] The steps of realizing the non - linear optimization solution of the target convex function under constraints by using a reconstruction algorithm based on a reconstruction module to obtain the target object image corresponding to the target object are as follows:

[0129]

[0130] where, R n represents the light intensity information of the n - th pixel of the estimated target object image; n represents the current sampling times during the acquisition process, n = 1, 2, 3,..., M; M represents the total sampling times, φ n represents the light intensity information of the n - th speckle matrix, and I n represents the light intensity information collected by the detector for the n - th time.

[0131] In this embodiment, the principle of compressive sensing is as Figure 7 shown. The reconstruction algorithm is the most important step in the compressive sensing framework. The present invention uses the total variation regularization algorithm to obtain a non - linear optimization solution for the target convex function under constraints. The total variation regularization method is a regularization method based on the L1 norm. Compared with the Tikhonov regularization method based on the L2 norm, this method can solve discontinuous solutions and can effectively reconstruct the edge information of the target object.

[0132] As Figure 8 shown, the working flow chart of a near - infrared single - pixel pre - denoising imaging device based on source estimation provided by the present application. It includes a near - infrared light source, an imaging target, an imaging lens, a spatial light modulator, a focusing lens, a single - point detector, and a data collector. First, the light emitted by the near - infrared light source irradiates the imaging object, and then the projected light beam passes through the imaging lens. The filtered light beam irradiates the spatial light modulator for spatial modulation. A modulation matrix generated by a computer is preset on the spatial light modulator, as Figure 2 shown. One of the reflected light beams modulated by the spatial light modulator is focused by the focusing lens and collected by the single - point detector. The data acquisition card completes analog - to - digital conversion and digital signal acquisition, which is called a sampling sequence. To overcome the influence of various noises in the acquisition environment, the collected sampling sequence is imported into the source estimation module to estimate the number of sources. After measuring the number of sources, for further efficient blind source separation, the corresponding number of detectors is configured to collect the corresponding sampling sequences and import them into the blind source separation model for dissociation of source signals and noises. After blind source separation, the mixed signal is decomposed into multiple one - dimensional signals. Among them, one is a valid sampling sequence, and the rest are noises. Therefore, a sample entropy discrimination module is used to screen out the valid sampling sequence for reconstruction. Finally, the valid sequence is imported into the reconstruction module for denoised image reconstruction.

[0133] In this embodiment, a novel single-pixel imaging technology is employed. It does not require the use of a complex sensor array and only needs one detector, which simplifies the structure of the imaging system, reduces costs and complexity. Since there is no need to arrange a dense sensor array, the single-pixel imaging technology can achieve a larger field of view angle, which is particularly important for application fields such as monitoring and remote sensing. The single-pixel imaging technology can adapt to different imaging requirements by changing the position of the detector and the parameters of the imaging system, with relatively high flexibility. Subsequently, the introduction of the near-infrared band compared with traditional visible-light single-pixel imaging technology has stronger penetration ability, can perform effective imaging under low-light conditions, and can penetrate deeper into biological tissues, which is of great significance for fields such as night monitoring and biomedical imaging. Secondly, to overcome the influence of noise in the acquisition environment on the imaging quality under low-light conditions, the one-dimensional sampling sequence collected is subjected to blind source separation based on source estimation, and the denoised one-dimensional signal is reconstructed to obtain a two-dimensional target image with higher quality, bringing great advantages to subsequent image processing and improving the imaging efficiency at the same time. In summary, the novel imaging technology in this embodiment not only simplifies the hardware design, reduces costs, but also significantly improves the imaging performance. By introducing the near-infrared band and advanced signal processing technology, the present invention brings new possibilities to the imaging field, especially showing great potential in low-light and biomedical imaging.

[0134] Optionally, the present application uses a single-pixel platform based on the near-infrared band to implement the pre-denoising imaging method of the present invention, and the process includes:

[0135] 1. Build a single-pixel imaging system, and the specific devices mainly include a near-infrared light source, an imaging target, an imaging lens, a spatial light modulator, a collecting lens, a single-point detector, and a data collector. The physical components of the hardware include software pointed to by a central processing unit, a digital signal processor, or a microprocessor. Computer storage media include, but are not limited to, random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory, or other memory technologies, compact disc read-only memory (CD-ROM), digital versatile disc or other optical disc storage, magnetic cassette, or any other medium for storing desired information and accessible by a computer.

[0136] 2. Edit the code modules, perform formula derivation for the theory of each module, and conduct mathematical modeling for each module in the Matrix Laboratory (MATLAB) simulation software. After completing each code block, perform preliminary debugging to check for syntax errors or logical errors. After the code block is debugged without errors, conduct simulation tests to simulate the actual operating environment, verify the functions and performance of the code block, and then complete the entire system and conduct simulation tests.

[0137] 3. Joint debugging: Place the target object at the target location, turn on the system platform, and the near-infrared single-pixel detector collects the light intensity information reflected from it. The analog-to-digital conversion is performed by the data collector to convert it into a processable digital signal and import it into the computer.

[0138] 4. Process the signal: Import the sampling sequence into the code area for processing, and finally display the reconstructed target image on the computer platform.

[0139] In this embodiment, a near-infrared single-pixel pre-denoising imaging device based on source estimation mainly includes two parts: a projection device and a collection device, which jointly achieve efficient imaging of the target object. The main function of the projection device is to project a speckle matrix onto the target object to facilitate subsequent light intensity information acquisition and processing. The projection device includes a spatial light modulator and a digital micromirror device. Among them, the spatial light modulator is the key part of the projection device. It dynamically modulates the phase, amplitude, or polarization state of the incident light according to the input digital signal to form a speckle pattern on the target object. The generation of the speckle pattern replaces the reference light field in traditional ghost imaging, making the imaging process no longer rely on complex interference or correlation measurements, and simplifying the structure of the imaging system. The digital micromirror device works in cooperation with the spatial light modulator to precisely control the reflection direction of the light beam. The main task of the collection device is to collect the light intensity information reflected or transmitted by the target object and convert it into a processable digital signal. The collection device mainly includes an imaging lens, a near-infrared single-pixel detector, a data collector, and a computer.

[0140] Consistent with the above embodiment, please refer to Figure 9 , Figure 9 which is a schematic structural diagram of a terminal provided by an embodiment of the present application. As Figure 9 shown, it includes a processor, an input device, an output device, and a memory. The processor, input device, output device, and memory are interconnected. Among them, the memory is used to store a computer program, and the computer program includes program instructions. The processor is configured to call the program instructions, and the above program includes instructions for performing the following steps;

[0141] Obtain the first light intensity information reflected by projecting the speckle matrix onto the target object collected by the target detector, and export the first light intensity information as a noisy sampling signal;

[0142] Based on the source number estimation module, the noisy sampling signal is analyzed and decomposed to obtain the number of independent sources;

[0143] According to the number of independent sources, a detector group corresponding to the number of independent sources is determined;

[0144] Obtain the second light intensity information reflected by projecting the speckle matrix onto the target object collected by the detector group, and export the second light intensity information as a multi-channel signal;

[0145] Based on the blind source separation module, the multi-channel signal is separated and extracted to obtain the target sampling signal;

[0146] Based on the reconstruction module, two-dimensional reconstruction processing is performed according to the target sampling signal to obtain the target object image corresponding to the target object.

[0147] The above mainly introduces the solution of the embodiment of the present application from the perspective of the execution process on the method side. It can be understood that in order for the terminal to implement the above functions, it includes the corresponding hardware structure and / or software module for executing each function. Those skilled in the art should easily realize that, combining the units and algorithm steps of each example described in the embodiments provided in this article, the present application can be implemented in the form of hardware or a combination of hardware and computer software. Whether a certain function is executed in the way of hardware or computer software driving hardware depends on the specific application and design constraints of the technical solution. Professional technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of the present application.

[0148] The embodiment of the present application can divide the functions of the terminal according to the above method examples. For example, each function unit can be divided corresponding to each function, or two or more functions can be integrated into one processing unit. The above integrated unit can be implemented in the form of hardware or in the form of a software function unit. It should be noted that the division of units in the embodiment of the present application is illustrative, only a logical function division, and there can be other division methods in actual implementation.

[0149] Consistent with the above, please refer to Figure 10 , Figure 10 FIG. is a schematic structural diagram of a near-infrared single-pixel pre-denoising imaging device based on source estimation provided by an embodiment of the present application. As Figure 10 shown, the device includes:

[0150] The first acquisition unit 101 is configured to acquire the first light intensity information reflected by projecting the speckle matrix onto the target object collected by the target detector, and export the first light intensity information as a noisy sampling signal;

[0151] The first processing unit 102 is configured to perform an analytical decomposition process on the noisy sampling signal based on a source number estimation module to obtain an independent source number;

[0152] The first determination unit 103 is configured to determine a detector group corresponding to the independent source number according to the independent source number;

[0153] The second acquisition unit 104 is configured to acquire second light intensity information reflected by projecting the speckle matrix onto a target object collected by the detector group, and export the second light intensity information as a multi-channel signal;

[0154] The second processing unit 105 is configured to perform a separation and extraction process on the multi-channel signal based on a blind source separation module to obtain a target sampling signal;

[0155] The third processing unit 106 is configured to perform a two-dimensional reconstruction process on the target object based on a reconstruction module according to the target sampling signal to obtain a target object image corresponding to the target object.

[0156] In a possible implementation manner, the first processing unit 102 is configured to perform an analytical decomposition process on the noisy sampling signal based on a source number estimation module to obtain an independent source number, and specifically configured to:

[0157] Decompose the noisy sampling signal into an intrinsic mode function signal and a residual signal based on an adaptive empirical mode decomposition module in the source number estimation module;

[0158] Based on the source number estimation module, combine the noisy sampling signal, the intrinsic mode function signal, and the residual signal into a matrix to obtain virtual multi-channel data;

[0159] Perform a singular value decomposition on a covariance matrix of the virtual multi-channel data based on a singular value decomposition module in the source number estimation module to obtain a sequence of differences between adjacent singular values;

[0160] Determine the independent source number according to the sequence of differences between adjacent singular values.

[0161] In a possible implementation manner, the first processing unit 102 is configured to perform a singular value decomposition on a covariance matrix of the virtual multi-channel data based on a singular value decomposition module in the source number estimation module to obtain a sequence of differences between adjacent singular values, and specifically configured to:

[0162] Determine a right matrix according to the covariance matrix of the virtual multi-channel data;

[0163] Perform an eigen decomposition process on the right matrix to obtain a right singular matrix;

[0164] Determine a left matrix according to the covariance matrix of the virtual multi-channel data;

[0165] Perform eigen-decomposition processing on the left matrix to obtain a left singular matrix;

[0166] Perform singular value decomposition on the covariance matrix of the virtual multi-channel data according to the right singular matrix and the left singular matrix to obtain a singular value decomposition diagonal matrix;

[0167] Perform singular value decomposition operation processing on the covariance matrix of the virtual multi-channel data according to the singular value decomposition diagonal matrix to obtain a neighboring singular value difference sequence.

[0168] In a possible implementation manner, the second processing unit 105 is configured to perform separation and extraction processing on the multi-channel signal based on a blind source separation module to obtain a target sampling signal, including:

[0169] Perform initial particle swarm processing on the multi-channel signal based on the blind source separation module to obtain an initial particle set;

[0170] Randomly generate the positions and velocities of each initial particle in the initial particle set based on the blind source separation module to obtain an initial position set and an initial velocity set;

[0171] Calculate the fitness of each initial particle in the initial particle set based on the blind source separation module according to the initial position set and the initial velocity set to obtain a fitness set;

[0172] Update the positions and velocities of each initial particle in the initial particle set based on the blind source separation module according to each fitness in the fitness set to obtain an updated position set and an updated velocity set;

[0173] Perform update processing on the multi-channel signal based on the blind source separation module according to the updated position set and the updated velocity set to obtain an updated multi-channel signal;

[0174] Perform effective sampling signal screening processing on the updated multi-channel signal based on the blind source separation module to obtain a target sampling signal.

[0175] In a possible implementation manner, the third processing unit 106 is configured to perform two-dimensional reconstruction processing on the target object according to the target sampling signal based on a reconstruction module to obtain a target object image corresponding to the target object, including:

[0176] Perform non-linear optimization and solution on the target convex function under constraints based on the reconstruction module using a reconstruction algorithm to obtain a target object image corresponding to the target object;

[0177] The steps of realizing non - linear optimization and solution of the target convex function under constraints by using a reconstruction algorithm based on a reconstruction module to obtain the target object image corresponding to the target object are as follows:

[0178]

[0179] Among them, R n represents the light intensity information of the nth pixel of the estimated target object image; n represents the current sampling times during the acquisition process, n = 1, 2, 3,..., M; M represents the total sampling times, and φ n represents the light intensity information of the nth speckle matrix, and I n represents the light intensity information collected by the detector for the nth time.

[0180] An embodiment of the present application also provides a computer storage medium. The computer storage medium stores a computer program for electronic data exchange, and the computer program enables a computer to execute some or all of the steps of any one of the near - infrared single - pixel pre - denoising imaging methods based on source estimation described in the above - mentioned method embodiments.

[0181] An embodiment of the present application also provides a computer program product. The computer program product includes a non - transitory computer - readable storage medium storing a computer program, and the computer program enables a computer to execute some or all of the steps of any one of the near - infrared single - pixel pre - denoising imaging methods based on source estimation described in the above - mentioned method embodiments.

[0182] It should be noted that for the foregoing method embodiments, for the sake of simple description, they are all expressed as a series of action combinations. However, those skilled in the art should know that the present application is not limited by the described action sequence, because according to the present application, certain steps can be performed in other sequences or simultaneously. Secondly, those skilled in the art should also know that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily essential to the present application.

[0183] In the above embodiments, the descriptions of each embodiment have their own emphases. For the parts not detailed in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0184] In several embodiments provided in the present application, it should be understood that the disclosed device can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of the units is only a logical function division. In actual implementation, there may be other division methods. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed couplings or direct couplings or communication connections between each other can be through some interfaces. The indirect couplings or communication connections of the devices or units can be in electrical or other forms.

[0185] The units described as separate components may or may not be physically separated. The components displayed as units may or may not be physical units, that is, they can be located in one place or distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0186] In addition, each functional unit in the various embodiments of the application can be integrated in a processing unit, or each unit can exist physically alone, or two or more units can be integrated in one unit. The above-mentioned integrated units can be implemented in the form of hardware or in the form of software program modules.

[0187] If the above-mentioned integrated unit is implemented in the form of a software program module and sold or used as an independent product, it can be stored in a computer-readable memory. Based on such an understanding, the technical solution of the present application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a memory and includes several instructions for causing a computer device (which can be a personal computer, a server or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present application. And the aforementioned memory includes: USB flash drives, read-only memories (ROM), random access memories (RAM), mobile hard disks, magnetic disks or optical disks and other media that can store program codes.

[0188] Those of ordinary skill in the art can understand that all or part of the steps in the various methods of the above embodiments can be completed by instructing relevant hardware through a program. This program can be stored in a computer-readable memory. The memory can include: flash drives, read-only memories, random access memories, magnetic disks or optical disks, etc.

[0189] The above has introduced the embodiments of the present application in detail. Specific examples are used herein to elaborate on the principle and implementation manner of the present application. The description of the above embodiments is only used to help understand the method and its core idea of the present application; at the same time, for those of ordinary skill in the art, according to the idea of the present application, there will be changes in the specific implementation manner and application scope. In summary, the content of this specification should not be construed as a limitation to the present application.

Claims

1. A near-infrared single-pixel pre-denoising imaging method based on source estimation, characterized in that The method includes: Obtaining first light intensity information reflected by projecting a speckle matrix onto a target object and collected by a target detector, and exporting the first light intensity information as a noisy sampling signal; Based on a source number estimation module, performing an analytical decomposition process on the noisy sampling signal to obtain an independent source number; According to the independent source number, determining a detector group corresponding to the independent source number; Obtaining second light intensity information reflected by projecting the speckle matrix onto the target object and collected by the detector group, and exporting the second light intensity information as a multi-channel signal; Based on a blind source separation module, performing a separation and extraction process on the multi-channel signal to obtain a target sampling signal; Based on a reconstruction module, performing a two-dimensional reconstruction process according to the target sampling signal to obtain a target object image corresponding to the target object.

2. The near-infrared single-pixel pre-denoising imaging method based on source estimation according to claim 1, wherein The step that the source number estimation module performs an analytical decomposition process on the noisy sampling signal to obtain an independent source number includes: Based on an adaptive empirical mode decomposition module in the source number estimation module, decomposing the noisy sampling signal into intrinsic mode signals and a residual signal; Based on the source number estimation module, combining the noisy sampling signal, the intrinsic mode signals, and the residual signal into a matrix to obtain virtual multi-channel data; Based on a singular value decomposition module in the source number estimation module, performing a singular value decomposition on the covariance matrix of the virtual multi-channel data to obtain a sequence of adjacent singular value differences; Determining the independent source number according to the sequence of adjacent singular values.

3. The near-infrared single-pixel pre-denoising imaging method based on source estimation according to claim 2, wherein The step that the singular value decomposition module in the source number estimation module performs a singular value decomposition on the covariance matrix of the virtual multi-channel data to obtain a sequence of adjacent singular value differences includes: Determining a right matrix according to the covariance matrix of the virtual multi-channel data; Performing an eigen decomposition process on the right matrix to obtain a right singular matrix; Determining a left matrix according to the covariance matrix of the virtual multi-channel data; Performing an eigen decomposition process on the left matrix to obtain a left singular matrix; According to the right singular matrix and the left singular matrix, performing a singular value decomposition on the covariance matrix of the virtual multi-channel data to obtain a singular decomposition diagonal matrix; According to the singular decomposition diagonal matrix, performing a singular value decomposition operation process on the covariance matrix of the virtual multi-channel data to obtain a sequence of adjacent singular value differences.

4. The near-infrared single-pixel pre-denoising imaging method based on source estimation according to claim 3, characterized in that The step that the blind source separation module performs a separation and extraction process on the multi-channel signal to obtain a target sampling signal includes: Based on the blind source separation module, performing an initial particle swarm process on the multi-channel signal to obtain an initial particle set; Based on the blind source separation module, randomly generating the positions and velocities of each initial particle in the initial particle set to obtain an initial position set and an initial velocity set; Based on the blind source separation module, calculating the fitness of each initial particle in the initial particle set according to the initial position set and the initial velocity set to obtain a fitness set; Based on the blind source separation module, updating the positions and velocities of each initial particle in the initial particle set according to each fitness in the fitness set to obtain an updated position set and an updated velocity set; Based on the blind source separation module, update and process the multi-channel signals according to the updated position set and the updated speed set to obtain updated multi-channel signals; Based on the blind source separation module, perform effective sampling signal screening on the updated multi-channel signals to obtain target sampling signals.

5. The near-infrared single-pixel pre-denoising imaging method based on source estimation according to any one of claims 1-4, characterized in that The two-dimensional reconstruction process based on the reconstruction module to obtain the target object image corresponding to the target object according to the target sampling signal includes: Based on the reconstruction module, use the reconstruction algorithm to perform non-linear optimization on the target convex function under constraints to obtain the target object image corresponding to the target object; The steps of using the reconstruction algorithm to perform non-linear optimization on the target convex function under constraints based on the reconstruction module to obtain the target object image corresponding to the target object are realized through the following formula: Among them, R n represents the light intensity information of the n-th pixel of the estimated target object image; n represents the current sampling number during the acquisition process, n = 1, 2, 3,..., M; M represents the total sampling number, φ n represents the light intensity information of the n-th speckle matrix, I n represents the light intensity information collected by the detector for the n-th time.

6. A near-infrared single-pixel pre-denoising imaging device based on source estimation, characterized in that, The device includes: The first acquisition unit is used to acquire the first light intensity information reflected by projecting the speckle matrix onto the target object collected by the target detector and export the first light intensity information as a noisy sampling signal; The first processing unit is used to perform analytical decomposition on the noisy sampling signal based on the source number estimation module to obtain the independent source number; The first determination unit is used to determine the detector group corresponding to the independent source number according to the independent source number; The second acquisition unit is used to acquire the second light intensity information reflected by projecting the speckle matrix onto the target object collected by the detector group and export the second light intensity information as multi-channel signals; The second processing unit is used to perform separation and extraction on the multi-channel signals based on the blind source separation module to obtain target sampling signals; The third processing unit is used to perform two-dimensional reconstruction on the target object according to the target sampling signal based on the reconstruction module to obtain the target object image corresponding to the target object.

7. The near-infrared single-pixel pre-denoising imaging device based on source estimation according to claim 6, wherein The first processing unit is used to perform analytical decomposition on the noisy sampling signal based on the source number estimation module to obtain the independent source number, specifically: Based on the adaptive empirical mode decomposition module in the source number estimation module, decompose the noisy sampling signal into intrinsic mode functions and residual signals; Based on the source number estimation module, combine the noisy sampling signal, the intrinsic mode functions and the residual signals into a matrix to obtain virtual multi-channel data; Based on the singular value decomposition module in the source number estimation module, perform singular value decomposition on the covariance matrix of the virtual multi-channel data to obtain a sequence of adjacent singular value differences; Determine the independent source number according to the sequence of adjacent singular values.

8. The near-infrared single-pixel pre-denoising imaging device based on source estimation according to claim 7, wherein The first processing unit is used to perform singular value decomposition on the covariance matrix of the virtual multi-channel data based on the singular value decomposition module in the source number estimation module to obtain a sequence of adjacent singular value differences, specifically: Determine the right matrix according to the covariance matrix of the virtual multi-channel data; Perform eigenvalue decomposition on the right matrix to obtain the right singular matrix; Determine the left matrix according to the covariance matrix of the virtual multi-channel data; Perform eigenvalue decomposition on the left matrix to obtain the left singular matrix; Perform singular value decomposition on the covariance matrix of the virtual multi-channel data according to the right singular matrix and the left singular matrix to obtain a singular value decomposition diagonal matrix; Perform singular value decomposition operation processing on the covariance matrix of the virtual multi-channel data according to the singular value decomposition diagonal matrix to obtain a sequence of adjacent singular value differences.

9. A terminal, characterized in that, It includes a processor, an input device, an output device, and a memory. The processor, the input device, the output device, and the memory are interconnected. Among them, the memory is used to store a computer program, the computer program includes program instructions, and the processor is configured to call the program instructions to execute the near-infrared single-pixel pre-denoising imaging method based on source estimation according to any one of claims 1-5.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, the computer program includes program instructions, and the program instructions, when executed by the processor, cause the processor to execute the near-infrared single-pixel pre-denoising imaging method based on source estimation according to any one of claims 1-5.

Images