Dynamic scattering medium imaging method and system based on variable-length Rayleigh bridge

Abstract

The invention provides a dynamic scattering medium imaging method and system based on a variable-length Rayleigh bridge, and the method comprises the steps: carrying out the iterative sampling of a speckle pattern based on a denoising diffusion implicit model, and obtaining a final target image; and in the iterative sampling process, inputting the current image into the trained convolutional neural network to obtain the noise of the next iteration. Compared with the prior art, the imaging method based on Rayleigh distribution has the advantages that the target image is reconstructed through continuous iteration, noise in each iteration step is predicted through the neural network model, the characteristics of the neural network model and the Rayleigh bridge model are combined, and the accuracy of image reconstruction is improved; according to the method, the speckle degeneration degree is quantitatively evaluated by using a self-attention mechanism, so that the bridge length of the Rayleigh bridge model is adaptively adjusted, and the image reconstruction speed is greatly improved.

Description

Technical Field

[0001] The present application belongs to the field of optical imaging, and specifically relates to a method and system for imaging a dynamic scattering medium based on a variable-length Rayleigh bridge. Background Art

[0002] When light passes through dynamic media such as smoke, clouds, and fog, due to the presence of random scattering particles, its propagation path is no longer constrained by fixed rules. Especially when the thickness of the medium reaches a certain degree or the disturbance of the medium is relatively large, the propagation of light will show random characteristics, causing the spatial information of the light to become chaotic, which greatly limits the depth of optical imaging. Therefore, common dynamic scattering imaging technology often has limitations in practical applications in remote sensing, medical treatment, and military. There are many imaging technologies, including ballistic photon extraction, transmission matrix, deconvolution, memory effect, wavefront shaping, etc. These technologies use various physical properties of light in the scattering process to significantly improve the quality of scattering imaging. Among the above technologies, ballistic photons are suitable for imaging under dynamic scattering conditions. It extracts ballistic photons suitable for imaging from scattered light based on different physical properties such as polarization, flight time, and coherence. However, ballistic photons can only be extracted in large quantities at 10 optical thicknesses, which limits its imaging depth.

[0003] In recent years, with the growth of computing resources and data resources, deep learning-based methods have gradually been adopted by all walks of life. The deep learning-based approach does not require the setting of a complex imaging system to achieve high-quality imaging. Therefore, it has been widely used in various computational imaging studies. However, since the existing deep learning methods are purely data-driven, the performance of the model depends entirely on the training data, and the generalization is greatly limited, especially for dynamic scattering media, where the data distribution is constantly changing, it is often difficult to apply. At the same time, the existing deep learning-based methods often only focus on the input and output of the model, using a black box to fit the scattering process of light, which leads to the complete neglect of the physical process of imaging and poor interpretability.

[0004] In summary, for the problem of optical imaging of dynamic scattering media, the use of deep learning-based methods still has significant limitations. Summary of the invention

[0005] The purpose of this application is to overcome the defects of the existing technology that it is too dependent on training data, has weak generalization and weak interpretability.

[0006] In order to achieve the above objectives, the present application proposes a dynamic scattering medium imaging method based on a variable-length Rayleigh bridge, comprising:

[0007] The speckle pattern is iteratively sampled based on the denoising diffusion implicit model to obtain the final target image;

[0008] During the iterative sampling process, the current image is input into the trained convolutional neural network to obtain the noise for the next iteration.

[0009] As an improvement of the above method, the iterative sampling formula is:

[0010]

[0011] Among them, μ s-1 represents the mean value at the s-1th iteration; X(0) represents the target image. When it iterates to the last time, X(0) = μ 0 ; s represents the number of iterations, from T to 0; [1:S] is a subsequence of [1:T], s∈[1:S] uses skip sampling; X(s) represents the result of the sth iteration; X(T) represents the speckle pattern; δ s represents the variance at the sth iteration; σ s-1 Represents the noise added by the denoising diffusion implicit model at the s-1th iteration.

[0012] As an improvement of the above method, the variance δ s It is expressed as:

[0013]

[0014] As an improvement of the above method, in the iterative sampling formula, at each iteration, the target image X(0) is obtained by the following formula:

[0015]

[0016] Among them, ∈ θ (μ s ,s) represents a convolutional neural network.

[0017] As an improvement of the above method, the convolutional neural network is a U-net neural network.

[0018] As an improvement of the above method, the speckle pattern degradation degree T is calculated based on the self-attention mechanism:

[0019] The speckle pattern pixels are spliced ​​into a long pixel vector by row, and then the cosine similarity between each pixel and all other pixels is calculated to obtain a similarity matrix; the distance between the corresponding two pixels in the original image is used as the weight to obtain a weight matrix;

[0020] The convolution result of the similarity matrix and the weight matrix is ​​taken as the degradation degree T.

[0021] The present application also provides a dynamic scattering medium imaging system based on a variable-length Rayleigh bridge, which is implemented based on the above method, and the system includes:

[0022] An iterative sampling module is used to iteratively sample the speckle pattern based on a denoising diffusion implicit model to obtain a final target image;

[0023] The noise prediction module is used to input the current image into the trained convolutional neural network during the iterative sampling process to obtain the noise for the next iteration.

[0024] As an improvement of the above system, the system further comprises:

[0025] The degradation degree calculation module is used to calculate the degradation degree T of the speckle pattern using the self-attention mechanism.

[0026] Compared with the prior art, the advantages of this application are:

[0027] 1. The imaging method based on Rayleigh distribution in the present application method reconstructs the target image through continuous iteration, wherein the noise in each iterative step is predicted by a neural network model, which combines the characteristics of the neural network model and the Rayleigh bridge model to improve the accuracy of the reconstructed image;

[0028] 2. The method of this application uses a self-attention mechanism to quantitatively evaluate the degree of speckle degradation, thereby adaptively adjusting the bridge length of the Rayleigh bridge model, greatly improving the speed of image reconstruction. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 The figure shows the network architecture of the neural network prediction model used in this application; wherein, Down SampleBlock: down sampling module; ResidualAttn Block: residual attention module; Up Sample Block: up sampling module; Up Sample Block (Residual+Attn): up sampling module combining residual structure and attention mechanism; Concat: skip connection structure;

[0030] Figure 2 The following is the architecture diagram of the Rayleigh bridge model; 0 to x T For the forward process, by adding noise and x that obeys the Rayleigh distribution T Component implementation, no need to use neural network; x T to x 0 The reverse process is the speckle reconstruction process, which predicts x through the neural network. t+1 The noise component added at time t further obtains x t ;

[0031] Figure 3The figure shows a schematic diagram of the speckle degradation degree calculation process based on self-attention; where Speckle Pattern: speckle pattern; Copy: copy; Similarity Matrix: similarity matrix; Weight Matrix: weight matrix; Degradation Level: degradation level;

[0032] Figure 4 The figure shows the VRDM model diagram combined with the degradation degree. The first column on the left is the input speckle image, and the first column on the right is the model reconstruction image. Among them, T is the speckle degradation degree. According to different scattering conditions, T will determine the number of steps of VRDM model iteration and make targeted reconstruction.

[0033] Figure 5 The figure shows the optical path diagram of the image acquisition of the time-varying scattering medium data set based on the fog chamber; where Seatteringspeckle: speckle pattern; Reconstructed image: reconstructed target image; Fog generator: fog generator; Fog chamber: fog chamber; Blower: blower; Original target image: original target image; CCD: charge coupled device; DMD: spatial light modulator; Laser: laser; L1, L2: lenses;

[0034] Figure 6(a) shows the speckle pattern of the partial reconstruction process of the random scattering model based on Rayleigh distribution;

[0035] Figure 6(b) shows the intermediate result of the partial reconstruction process of the random scattering model based on Rayleigh distribution;

[0036] Figure 6(c) shows the target image of the partial reconstruction process of the random scattering model based on Rayleigh distribution;

[0037] Figure 7 Shown is the frequency distribution histogram of the peak signal-to-noise ratio (PSNR) of the random scattering model based on Rayleigh distribution on the test set;

[0038] Figure 8 Shown is the frequency distribution histogram of the structural similarity index (SSIM) of the random scattering model based on Rayleigh distribution on the test set;

[0039] Fig. 9 The following are some examples of the degree of degradation of the speckle pattern. The numbers on the top of each image represent the degree of degradation of each image. The larger the number, the more serious the degradation.

[0040] Fig.10It is a schematic diagram showing the effect of degradation on time; VRDM is the model proposed by the present invention, the left side is the original model, the right side is the effect of adding degradation on the model training and testing time overhead, the vertical axis is time, the unit is minutes;

[0041] Fig.11 The figure shows the effect of degradation degree on peak signal-to-noise ratio and structural similarity index. The horizontal axis represents the number of speckle patterns in the test set, and the vertical axes represent peak signal-to-noise ratio and structural similarity index, respectively. DETAILED DESCRIPTION

[0042] The technical solution of the present application is described in detail below with reference to the accompanying drawings.

[0043] In view of the difficulty of imaging dynamic scattering media, the present application proposes a method and system for imaging dynamic scattering media based on a variable-length Rayleigh bridge. When a signal is composed of a large number of complex components with independent phases (components with both amplitude and phase), speckle will appear in the signal, and the probability density of the amplitude obeys the Rayleigh distribution. The imaging method based on the Rayleigh bridge distribution is based on the Brownian bridge model in statistics and the currently popular diffusion model, and a random process is established through the Rayleigh distribution to simulate the process of light scattering. When the number of optical speckle scattering times is too large, it is quite difficult to reconstruct the target image directly from the speckle. However, knowing the speckle at a certain time t, it is relatively easy to restore the time immediately adjacent to t-1. The imaging method based on the Rayleigh distribution proposed in the present invention reconstructs the target image by continuously iterating this process. In addition, according to the different degrees of scattering, the present invention also proposes a dynamic scattering medium imaging method based on a variable-length Rayleigh bridge (Variable-Length Rayleigh Bridge Denoising Model, VRDM). Based on the Rayleigh bridge model, this method proposes to use a self-attention mechanism to quantitatively evaluate the degree of speckle degradation, thereby adaptively adjusting the bridge length of the Rayleigh bridge model. The variable-length Rayleigh bridge-based dynamic scattering medium imaging method (VRDM) provides a new solution for dynamic scattering imaging.

[0044] 1. Diffusion model;

[0045] The idea of ​​iteratively reconstructing the target image proposed in the present invention is based on the Denoising Diffusion Probabilistic Models (DDPM). The basic idea of ​​the diffusion model is that in order to know how to decode noise, we should first know how the image is destroyed into noise. Therefore, it consists of a forward process (noise addition) and a reverse process (noise removal). Similar to the diffusion model, the study of the speckle formation process is very important for speckle reconstruction.

[0046] The forward process (noise addition) is shown in Formula 1. Randomly extract a sample x from the data0 , and then add Gaussian noise to it to get x 1 , repeating this process, the image is eventually destroyed into noise x that obeys the standard Gaussian distribution T .x 0 to x T Construct a Markov chain.

[0047]

[0048] In the above formula, Follows normal distribution, β t is a custom hyperparameter, which is a constant. 0 Gradually add noise that follows a standard Gaussian distribution to obtain noise x T , x T Obeying the standard Gaussian distribution, p latent (x T )=N(0,I). The reverse process (denoising) uses the knowledge learned in the forward process to extract the Gaussian noise x from the Gaussian noise x through another Markov chain. T Rebuild x 0 .

[0049]

[0050] In the forward process, Give x t Add Gaussian noise to get x t+1 . Among them α t is also a custom hyperparameter, so the forward process is a deterministic process without the involvement of neural networks. In the reverse process, when x is known t+1 In the case of t For example, if a speckle pattern is taken at a certain moment, it is difficult to obtain the speckle pattern at the previous moment adjacent to this moment. Therefore, a neural network is used to predict x t , the neural network is based on x t+1 and time t to predict x t Of course, the real DDPM does not directly predict x t , but by predicting the noise ∈ added at time t in the forward process t , through x t =x t+1 -∈ t To find x t .

[0051] 2. Random scattering model based on Rayleigh distribution

[0052] In order to model the mapping between paired speckle images and target images, the present invention proposes a Rayleigh bridge model based on the diffusion model DDPM and Rayleigh distribution, and its mathematical expression is as follows.

[0053]

[0054] Among them, X(0) and X(T) are paired target images and speckle images, R(t) and R(T) are Rayleigh distributions at a certain moment, the image X(0) at time 0 is the target image, the image X(T) at time T is the speckle pattern, and X(t) is the intermediate image at a certain moment in the scattering process, t∈[0,T]. The expectation and variance of X(t) in Formula 3 are obtained as Formula 4 and Formula 5 respectively.

[0055]

[0056]

[0057] Where σ is the standard deviation of the Rayleigh distribution. At t = 0, E(X(t)) = X(0), where X(0) is the target of reconstruction. When t = T, E(X(t)) = X(T), where X(T) is the speckle pattern. The present invention uses formula 6 for sampling.

[0058]

[0059] where μ t and σ t They represent the mean E(X(t)) and the variance D(X(t)) respectively. It should be noted that in the speckle reconstruction task, the sampling formula should satisfy X(0) = μ when t = 0. 0 When t = T, X(T) = μ T This is because the target image at time 0 and the speckle pattern at time T are both determined in the model of the present invention, so the variance σ is required t = 0. From formula 5, we can see that the variance of the established model meets the requirement at time 0, but not at time T. Therefore, it is necessary to transform the variance formula to obtain the final variance formula 7.

[0060]

[0061] Here σ = 1. At this time, Formula 6 can be used to perform sampling training for the entire process, where σ is the standard deviation.

[0062] In the actual training process of the model of the present invention, Formula 6 is not used directly, but a neural network is used to learn μ t , which is the loss function in Algorithm 1, is used for sampling only when the neural network is trained and tested. This is because ∈ tis a noise that obeys the Rayleigh distribution with σ=1. If a neural network is directly used to fit Formula 6, the training time will be too long.

[0063]

[0064] X(0) in Formula 8 is the reconstruction target, which is unknown, so a neural network is needed to predict it. The loss function of the model is shown in Formula 9.

[0065]

[0066] where ∈ θ Represents a neural network, with input at time t and μ t , the output is μ t-1 .

[0067] During the test, since t needs to iterate from T to 0, and T is generally very large, such as T = 1000 steps, the model is very slow. The improved version of DDPM, Denoising Diffusion Implicit Models (DDIM), can solve this problem well. According to the sampling method proposed by DDIM, the sampling of the present invention can be obtained as shown in Formula 10.

[0068]

[0069] Among them, the subsequence [1:S] is extracted from [1:T], s∈[1:S] realizes skip sampling, δ s represents the variance in Formula 7, σ s-1 Here, the random noise is inherited from the DDIM algorithm. When sampling, the target image is obtained by equation 9, where The entire training and testing process is summarized in Algorithm 1 and Algorithm 2.

[0070]

[0071]

[0072] In the above algorithm, ∈ θ For neural networks, residual networks and self-attention mechanisms are added to the basic U-net architecture. The complete network architecture is as follows Figure 1 As shown in Figure 1, it includes four downsampling layers and four upsampling layers. Residual and self-attention modules are used in each downsampling layer and upsampling layer. The activation function between adjacent modules is ReLU. Jump connections are established between corresponding downsampling layers and upsampling layers. The activation function of the last layer of the network is Tanh.

[0073] use Figure 1The network predicts the noise during the inverse reconstruction process. The overall architecture of the Rayleigh bridge model is as follows: Figure 2 As shown. In the forward process, according to the Rayleigh bridge formula, x 0 Add noise and x T Components to simulate the scattering process of light. Since the inverse solution process is ill-conditioned, a neural network is used according to x t+1 To predict the previous moment x t , iterate this process until the reconstructed image x is obtained 0 .

[0074] 3. VRDM model of variable-length Rayleigh bridge combined with self-attention mechanism

[0075] The Rayleigh bridge model uses the same processing steps for all speckle images, but for actual situations, especially dynamic scattering media, we need to dynamically adjust the processing steps according to the different degrees of scattering. If you want to make targeted reconstructions for images with different scattering degrees, you need appropriate evaluation indicators. Although there are many indicators currently used to quantitatively analyze imaging quality, such as peak signal-to-noise ratio (PSNR), structural similarity index (SSIM), etc., these indicators often require speckle patterns and target images. As supervised indicators, they are only suitable for evaluating the quality of the scheme and are difficult to apply to the reconstruction of real reconstruction tasks. The present invention proposes to use a self-attention mechanism to quantitatively calculate the degree of degradation of the speckle image. When humans observe things, they often only focus on the key features of the object. For a speckle image with a low degree of degradation, human attention will be attracted by the edge contour, and the attention will be focused on certain specific parts. However, as the degree of degradation of the speckle gradually increases, the edge contour of the image will gradually blur, and human visual attention will be dispersed on the entire speckle image. Therefore, the degree of degradation of the speckle image can be measured according to the degree of distraction of attention.

[0076] The process of quantitatively calculating the degree of speckle image degradation based on the self-attention mechanism is as follows: Figure 3 As shown. For a speckle image, first straighten its pixels by row, then calculate the cosine similarity between each pixel and all other pixels, and the result is the similarity matrix (Similarity Matrix). At the same time, the distance between the corresponding two pixels in the original image is used as the weight to obtain the weight matrix (Weight Matrix). Finally, the convolution result of the similarity matrix and the weight matrix is ​​used as the degradation degree. This calculation process fully simulates the changes in human visual attention when observing things. At the same time, this indicator only requires speckle images. As an unsupervised calculation scheme, it is particularly suitable for the process of speckle reconstruction. VRDM combined with speckle degradation degree is shown as Figure 4 As shown in the figure, according to the degradation degree of the speckle, the reconstruction step number T is dynamically adjusted to perform targeted reconstruction. Tis the speckle pattern, x 0 To reconstruct the image. During the entire reconstruction process, the model continuously t+1 To predict x t , until the reconstructed image x is obtained 0 At the same time, some images during the VRDM iteration process are randomly selected for display, making the model reconstruction process more transparent.

[0077] The technical effect of this application is demonstrated through experiments.

[0078] 1. Experimental setup

[0079] Data acquisition optical path Figure 5 As shown. The laser with a wavelength of 532nm is expanded by an optical system composed of lenses L1 and L2 and then irradiated onto a digital micromirror device (DMD). The DMD is used to display the original target image. The target image is selected from the MNIST (Modified National Institute of Standards and Technology) data set. The target image passes through a fog chamber filled with water mist to form speckles. In order to make the scattering medium (fog) change more drastically, a fan that can rotate 360° is started when the fog generator enters the fog chamber. The speckles after passing through the fog chamber are collected by a charge coupled device (CCD). The present invention collects a total of 7033 pairs of images, including a test set (702 pairs), a validation set (705 pairs), and a training set (5626 pairs). In order to train the model, the network weights are updated with the L1 loss between the model output and the target. In order to facilitate the calculation of the degradation degree, only one pair of images is processed at a time. Adam is used as the default optimizer and trained for 10 cycles (epochs). The model was implemented using Pytorch (2.1.1+cu121) in Ubuntu 20.04 and accelerated by GeForce RTX 4090.

[0080] 2. Reconstruction results of random scattering model based on Rayleigh distribution

[0081] The partial reconstruction results of the random scattering model based on Rayleigh distribution are shown in Figure 6. According to the reconstruction results, the model can reconstruct images with a high degree of scattering. At the same time, the frequency distribution histograms of the peak signal-to-noise ratio (PSNR) and structural similarity index (SSIM) of the reconstruction results are shown in Figure 7 and 8 As shown in the figure, the peak signal-to-noise ratio of the reconstructed image is concentrated between 23 and 33, with an average of 28.18. The structural similarity index is concentrated between 0.85 and 0.96, with an average of 0.91, and a good reconstruction result is achieved overall.

[0082] 3. Reconstruction results of the variable-length Rayleigh bridge dynamic scattering medium imaging method (VRDM)

[0083] Since the spatial structure of the speckle image is randomly disrupted during the scattering process, the self-attention of the image is dispersed. Therefore, the self-attention of the speckle pattern can be used to calculate the degradation degree. However, directly calculating the self-attention of the speckle pattern is very time-consuming, and the time complexity is proportional to (h×w) 2 , h and w are the resolutions of the image. Therefore, the hierarchical self-attention calculation based on the sliding window is actually used. The degradation degree of some speckle images is shown in the following example: Fig. 9 As shown in the figure, it can be seen that for speckles with more severe scattering, the degradation value is larger.

[0084] In addition to the self-attention mechanism proposed in the present invention to measure the degree of speckle degradation, there are many other schemes that can be used to evaluate the degree of speckle degradation. The present invention uses the structural similarity index (SSIM) as a benchmark to compare many unsupervised evaluation schemes, and the results are shown in Table 1. The table uses 702 speckle images as test data for degradation. In the table, the average error refers to the average value of the difference between various unsupervised algorithms and the degradation degree calculated based on the structural similarity index. From the experimental results, it can be seen that the test results of the self-attention mechanism are better, so the present invention uses this index to control the size of parameters T and S in formula (10). For speckle images with different degrees of degradation, the length of the model path will be dynamically adjusted according to its specific degree of degradation. Specifically, the model established by formula (10) is equivalent to building a bridge between the speckle image and the target image. The original bridge length is fixed, but when the degradation index proposed in the present invention is added, the bridge length will change dynamically and adaptively with the intensity of the scattering degree. The effects of degradation on the time and imaging quality of the model are as follows: Fig.10 , Fig.11 As shown in the results, we can see that while ensuring the reconstruction quality is equivalent or better, the time overhead of the training process and the testing process is shortened by 19.25% and 4.56% respectively.

[0085] Table 1. Average error of unsupervised evaluation scheme

[0086]

[0087]

[0088] The present application also provides a dynamic scattering medium imaging system based on a variable-length Rayleigh bridge, which is implemented based on the above method, and the system includes:

[0089] An iterative sampling module is used to iteratively sample the speckle pattern based on a denoising diffusion implicit model to obtain a final target image;

[0090] The noise prediction module is used to input the current image into the trained convolutional neural network during the iterative sampling process to obtain the noise for the next iteration.

[0091] The degradation degree calculation module is used to calculate the degradation degree of the speckle pattern using the self-attention mechanism.

[0092] The present application may also provide a computer device, comprising: at least one processor, a memory, at least one network interface and a user interface. The various components in the device are coupled together through a bus system. It is understood that the bus system is used to achieve connection and communication between these components. In addition to the data bus, the bus system also includes a power bus, a control bus and a status signal bus.

[0093] The user interface may include a display, a keyboard or a pointing device, such as a mouse, a trackball, a touch pad or a touch screen.

[0094] It is understood that the memory in the embodiments disclosed in the present application may be a volatile memory or a non-volatile memory, or may include both volatile and non-volatile memories. Among them, the non-volatile memory may be a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), or a flash memory. The volatile memory may be a random access memory (RAM), which is used as an external cache. By way of example and not limitation, many forms of RAM are available, such as static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDRSDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous link dynamic random access memory (SLDRAM), and direct memory bus random access memory (DRRAM). The memories described herein are intended to include, but are not limited to, these and any other suitable types of memories.

[0095] In some embodiments, the memory stores the following elements, executable modules or data structures, or a subset thereof, or an extended set thereof: an operating system and applications.

[0096] The operating system includes various system programs, such as a framework layer, a core library layer, a driver layer, etc., which are used to implement various basic services and process hardware-based tasks. The application includes various application programs, such as a media player (Media Player), a browser (Browser), etc., which are used to implement various application services. The program for implementing the method of the embodiment of the present disclosure can be included in the application.

[0097] In the above embodiment, the processor may also call a program or instruction stored in the memory, specifically, a program or instruction stored in an application program, and is used to:

[0098] Execute the steps of the above method.

[0099] The above method can be applied to a processor or implemented by a processor. The processor may be an integrated circuit chip with signal processing capabilities. In the implementation process, each step of the above method can be completed by an integrated logic circuit of hardware in the processor or an instruction in the form of software. The above processor may be a general processor, a digital signal processor (Digital Signal Processor, DSP), an application specific integrated circuit (Application Specific Integrated Circuit, ASIC), a field programmable gate array (Field Programmable Gate Array, FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components. The above-disclosed methods, steps and logic block diagrams can be implemented or executed. The general processor may be a microprocessor or the processor may also be any conventional processor, etc. The steps of the above-disclosed method can be directly embodied as a hardware decoding processor to execute, or the hardware and software modules in the decoding processor are combined to execute. The software module can be located in a mature storage medium in the field such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory or an electrically erasable programmable memory, a register, etc. The storage medium is located in the memory, and the processor reads the information in the memory and completes the steps of the above method in combination with its hardware.

[0100] It is understood that the embodiments described in the present application can be implemented by hardware, software, firmware, middleware, microcode or a combination thereof. For hardware implementation, the processing unit can be implemented in one or more application specific integrated circuits (ASIC), digital signal processors (DSP), digital signal processing devices (DSPD), programmable logic devices (PLD), field programmable gate arrays (FPGA), general-purpose processors, controllers, microcontrollers, microprocessors, other electronic units for performing the functions described in the present application or a combination thereof.

[0101] For software implementation, the technology of the present application can be implemented by executing the functional modules (such as procedures, functions, etc.) of the present application. The software code can be stored in a memory and executed by a processor. The memory can be implemented in the processor or outside the processor.

[0102] The present application may also provide a non-volatile storage medium for storing a computer program. When the computer program is executed by a processor, each step in the above method embodiment can be implemented.

[0103] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present application and are not intended to limit it. Although the present application is described in detail with reference to the embodiments, a person skilled in the art should understand that any modification or equivalent replacement of the technical solution of the present application does not depart from the spirit and scope of the technical solution of the present application and should be included in the scope of the claims of the present application.

Claims

1. A dynamic scattering medium imaging method based on a variable-length Rayleigh bridge, comprising: The speckle pattern is iteratively sampled based on the denoising diffusion implicit model to obtain the final target image; During the iterative sampling process, the current image is input into the trained convolutional neural network to obtain the noise for the next iteration.

2. The dynamic scattering medium imaging method based on variable length Rayleigh bridge according to claim 1, characterized in that: The iterative sampling formula is: Among them, μ s-1 represents the mean value at the s-1th iteration; X(0) represents the target image. When it is the last iteration, X(0) = μ0; s represents the number of iterations, from the speckle pattern degradation degree T to 0; [1:S] is a subsequence of [1:T], s∈[1:S] uses skip sampling; X(s) represents the result of the sth iteration; X(T) represents the speckle pattern; δ s represents the variance at the sth iteration; σ s-1 Represents the noise added by the denoising diffusion implicit model at the s-1th iteration.

3. The dynamic scattering medium imaging method based on variable length Rayleigh bridge according to claim 2, characterized in that: The variance δ s It is expressed as:

4. The method for dynamic scattering medium imaging based on variable length Rayleigh bridge according to claim 2, characterized in that: In the iterative sampling formula, at each iteration, the target image X(0) is obtained by the following formula: Among them, ∈ θ (μ s ,s) represents a convolutional neural network.

5. The method for dynamic scattering medium imaging based on variable length Rayleigh bridge according to claim 1, characterized in that: The convolutional neural network is a U-net neural network.

6. The method for dynamic scattering medium imaging based on variable length Rayleigh bridge according to claim 1, characterized in that: Also includes: The speckle pattern degradation degree T is calculated based on the self-attention mechanism: The speckle pattern pixels are spliced ​​into a long pixel vector by row, and then the cosine similarity between each pixel and all other pixels is calculated to obtain a similarity matrix; the distance between the corresponding two pixels in the original image is used as the weight to obtain a weight matrix; The convolution result of the similarity matrix and the weight matrix is ​​taken as the degradation degree T.

7. A dynamic scattering medium imaging system based on a variable length Rayleigh bridge, implemented based on the method according to any one of claims 1 to 6, characterized in that: The system comprises: an iterative sampling module, used for iteratively sampling the speckle pattern based on a denoising diffusion implicit model to obtain a final target image; and The noise prediction module is used to input the current image into the trained convolutional neural network during the iterative sampling process to obtain the noise for the next iteration.

8. The dynamic scattering medium imaging system based on variable length Rayleigh bridge according to claim 7, characterized in that: The system further comprises: The degradation degree calculation module is used to calculate the degradation degree T of the speckle pattern using the self-attention mechanism.

Images