Unsupervised domain adaptive image non-blind deconvolution method, device and medium

Abstract

The present invention discloses an unsupervised domain adaptive image non-blind deconvolution method, device and medium, wherein the method comprises: determining source domain data and target domain data; constructing a non-blind deconvolution model; using the source domain data to perform supervised training on the non-blind deconvolution model to obtain a pre-trained model; applying the obtained pre-trained model to the target domain data, and using unsupervised adaptive loss to train the pre-trained model to obtain a migration model; judging whether the deconvolution accuracy of the migration model meets the preset conditions, if so, stopping the training, otherwise continuing the training; applying the migration model after the migration training to the target domain data, and outputting the deconvolution result of the test image. The present invention migrates the pre-trained model without accessing the clear images in the source domain data and the target domain data, thereby improving the non-blind deconvolution performance of the model in the target domain. The present invention can be widely used in the field of image non-blind deconvolution.

Description

Technical Field

[0001] The present invention relates to the field of non-blind image deconvolution, and in particular to an unsupervised domain adaptive non-blind image deconvolution method, device and medium. Background Art

[0002] Image non-blind deconvolution refers to a method of restoring a clear image from a given blurred image and its corresponding blur kernel. As an important issue in scientific imaging and image restoration, non-blind image deconvolution has important research significance and application value. The challenge of non-blind image deblurring mainly comes from how to suppress noise amplification while restoring image details. Traditional non-blind deblurring methods, such as regularization methods based on image priors, solve the problem by manually defining image priors. Its iterative optimization process takes a long time, and the regularization parameters need to be optimized for each image, so satisfactory results cannot be achieved in terms of running time and restoration effect.

[0003] In recent years, with the rapid development of deep learning in the field of image processing, methods based on deep learning have gradually become the mainstream of image non-blind deconvolution algorithms. At present, most of the commonly used image non-blind deconvolution methods based on deep learning are supervised methods, which are trained based on a large amount of blurred and clear image data to obtain deblurring models. However, in many application fields of scientific imaging, such as microscopy imaging and remote sensing imaging, it is extremely difficult or even impractical to obtain clear images, which limits the widespread application of supervised deep learning non-blind image deblurring. Directly calling the pre-trained model on the source domain on the target domain may lead to a sharp drop in performance, especially when the images and blur kernels between the two domains are very different. Summary of the invention

[0004] In order to solve at least one of the technical problems existing in the prior art to a certain extent, the present invention aims to provide an unsupervised domain adaptive image non-blind deconvolution method, device and medium.

[0005] The technical solution adopted by the present invention is:

[0006] An unsupervised domain adaptive image non-blind deconvolution method comprises the following steps:

[0007] Determining source domain data and target domain data, wherein the source domain data includes a blurred image, a blur kernel, and a clear image, and the target domain data includes a blurred image and a blur kernel;

[0008] Build a non-blind deconvolution model;

[0009] The source domain data is used to perform supervised training on the non-blind deconvolution model to obtain a pre-trained model;

[0010] Apply the obtained pre-trained model to the target domain data and use unsupervised adaptive loss to train the pre-trained model to obtain a migration model.

[0011] Determine whether the deconvolution accuracy of the migration model meets the preset conditions. If so, stop training; otherwise, continue training.

[0012] The transfer model after transfer training is applied to the target domain data and the deconvolution result of the test image is output.

[0013] Furthermore, the non-blind deconvolution model is constructed by alternately iterating an inversion module and a denoising module;

[0014] Among them, the inversion module solves the quadratic problem by using the fast Fourier transform method to obtain the inversion result, and the denoising module is composed of a UNet network.

[0015] Furthermore, the expression of the unsupervised adaptive loss is as follows:

[0016]

[0017] Among them, M represents the non-blind deconvolution model, and the input of the non-blind deconvolution model includes the blurred image and blur kernel Γ1 and Γ2 represent data enhancement methods, including image rotation and folding. represents the image noise, represents the convolution operation, Represents the square of the Frobenius norm of the matrix.

[0018] Furthermore, the non-blind deconvolution model is based on a semi-quadratic splitting scheme to expand the regularization model, and the expression of the non-blind deconvolution model is as follows:

[0019]

[0020] in, represents some image prior, W p It is a set of wavelet high-pass filters, X represents a clear image, Y represents a blurred image, and K represents a blur kernel. represents the square of the Frobenius norm of the matrix, and p represents the number of the wavelet high-pass filter;

[0021] The iterative scheme based on semi-quadratic splitting is as follows: for t=1,…T,

[0022]

[0023]

[0024] where λt is a hyperparameter, T is the total number of iterations, A p Represents a set of auxiliary variables.

[0025] Furthermore, the inversion module Corresponding to the first step of the iteration; the denoising module Corresponds to the second step of the iteration.

[0026] Furthermore, the network structure of the denoising module is as follows:

[0027] The first convolution module has a convolution kernel size of 3×3, a stride of 1, a padding of 1, an input channel of t×C, and an output channel of 64.

[0028] The second convolution module has a convolution kernel size of 3×3, a stride of 2, a padding of 1, and an output channel of 128.

[0029] Deconvolution module, the convolution kernel size of the deconvolution module is 3×3;

[0030] The third convolution module has a convolution kernel size of 3×3, a stride of 1, a padding of 1, an input channel of 128, and an output channel of C.

[0031] Four residual channel attention modules with 64 input and output channels are connected in sequence between the first convolution module and the second convolution module; four residual channel attention modules with 128 input and output channels are connected in sequence between the second convolution module and the deconvolution module; four residual channel attention modules with 64 input and output channels are connected in sequence between the deconvolution module and the third convolution module; wherein the residual channel attention module is composed of a residual block and a channel self-attention mechanism module.

[0032] Furthermore, the determining whether the deconvolution accuracy of the migration model meets a preset condition includes:

[0033] The transfer model is used to deconvolute the verification data in the target domain, and the PSNR of the deconvolution result and the corresponding clear image are calculated and averaged. When the average PSNR is greater than or equal to the preset value, it is determined that the deconvolution accuracy of the transfer model meets the preset conditions.

[0034] Furthermore, the calculation formula of PSNR is as follows:

[0035]

[0036]

[0037] Among them, X represents a clear image, represents the restored image, MAX represents the maximum value of the pixel, m and n represent the length and width of the image respectively, and i and j represent the horizontal and vertical coordinates of the image pixels respectively.

[0038] Another technical solution adopted by the present invention is:

[0039] An unsupervised domain adaptive image non-blind deconvolution device, comprising:

[0040] at least one processor;

[0041] at least one memory for storing at least one program;

[0042] When the at least one program is executed by the at least one processor, the at least one processor implements the above method.

[0043] Another technical solution adopted by the present invention is:

[0044] A computer-readable storage medium stores a program executable by a processor, wherein the program executable by the processor is used to execute the method described above when executed by the processor.

[0045] The beneficial effect of the present invention is that the present invention migrates the pre-trained model without accessing clear images in the source domain data and the target domain data, thereby improving the non-blind deconvolution performance of the model in the target domain. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the embodiments of the present invention or the drawings of related technical solutions in the prior art are introduced below. It should be understood that the drawings introduced below are only for the convenience of clearly describing some embodiments of the technical solutions of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative work.

[0047] Figure 1 is a flow chart of an unsupervised domain adaptive non-blind image deconvolution method in an embodiment of the present invention;

[0048] Figure 2 is a network structure design diagram based on optimization expansion in an embodiment of the present invention;

[0049] Figure 3 is a network design diagram of a denoising module in an embodiment of the present invention;

[0050] Figure 4 is a schematic diagram of source domain data and target domain data in an embodiment of the present invention;

[0051] Figure 5It is the visual effect of non-blind image deconvolution on microscope imaging in an embodiment of the present invention;

[0052] Figure 6 This is the visual effect of non-blind image deconvolution on remote sensing imaging in an embodiment of the present invention. DETAILED DESCRIPTION

[0053] The embodiments of the present invention are described in detail below, and examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and are not to be construed as limitations of the present invention. For the step numbers in the following embodiments, they are only provided for the convenience of explanation, and the order between the steps is not limited in any way, and the execution order of each step in the embodiment can be adaptively adjusted according to the understanding of those skilled in the art.

[0054] In the description of the present invention, it should be understood that descriptions involving orientations, such as up, down, front, back, left, right, etc., and orientations or positional relationships indicated are based on the orientations or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be understood as a limitation on the present invention.

[0055] In the description of the present invention, "several" means one or more, "more" means more than two, "greater than", "less than", "exceed" etc. are understood as not including the number itself, and "above", "below", "within" etc. are understood as including the number itself. If there is a description of "first" or "second", it is only used for the purpose of distinguishing the technical features, and cannot be understood as indicating or implying the relative importance or implicitly indicating the number of the indicated technical features or implicitly indicating the order of the indicated technical features.

[0056] In the description of the present invention, unless otherwise clearly defined, terms such as setting, installing, connecting, etc. should be understood in a broad sense, and technicians in the relevant technical field can reasonably determine the specific meanings of the above terms in the present invention based on the specific content of the technical solution.

[0057] In order to solve the problem that non-blind deconvolution methods based on supervised deep learning in the fields of medical imaging and astronomical imaging cannot be trained, the unsupervised adaptive loss function proposed in the present invention uses the result of the model pre-trained in the source domain acting on the target domain data as a pseudo-label, and then regenerates a new degraded image to obtain paired data for training. The model pre-trained in the source domain can be migrated when there is no need to access the clear images in the source domain data and the target domain data, which can greatly increase the performance of the model on the target domain data. In addition, the effectiveness of the unsupervised adaptive loss function proposed in the present invention is verified through simulation experiments of microscope imaging deblurring, remote sensing imaging deblurring and Gaussian deblurring.

[0058] like Figure 1 As shown, the present embodiment provides an unsupervised domain adaptive non-blind image deconvolution method, which specifically includes the following steps:

[0059] S1. Determine the source domain data and the target domain data.

[0060] The source domain data includes a blurred image, a blur kernel and its corresponding clear image, while the target domain data only includes a blurred image and a blur kernel.

[0061] In some specific embodiments, natural images and motion blur kernels are selected as source domain data. A large amount of clear natural image data is first obtained by shooting natural scenes with a camera, and then a motion blur kernel is synthesized by randomly simulating the camera motion trajectory. The blur kernel is applied to the clear image, and Gaussian noise is added to generate a blurred image. Thus, a large number of image pairs are obtained for training the network.

[0062] Microscope images and remote sensing images are respectively used as target domain data, where the target domain data only contains blurred images and blur kernels.

[0063] Examples of source domain data and target domain data are as follows: Figure 4 As shown in , the image and blur kernel of the target domain are very different from those of the source domain. Figure 4 (a) is the source domain data, including blurred image, blur kernel and clear image; Figure 4 (b) is the target domain data, microscope imaging: including blurred image and blur kernel; Figure 4 (c) is the target domain data, remote sensing imaging: including blurred image and blur kernel.

[0064] S2. Build a non-blind deconvolution model.

[0065] Deep learning has achieved great success in image processing. For many image restoration tasks, the restoration process can be learned directly through the data pairs of degraded and clear images. However, for the task of non-blind image deconvolution, simple end-to-end models are not suitable. The reason is that the types of blur kernels vary, and blur kernels can have arbitrary sizes and shapes. It is difficult to fill the entire space where the blur kernel is located to generate blurred images during training. Secondly, since the blur kernel is known, a simple end-to-end model cannot fully utilize the model knowledge. For this reason, an embodiment of the present invention adopts a network structure design based on optimized expansion (Deep Unrolling). The Deep Unrolling framework expands the traditional optimization process, splits the non-blind image deconvolution problem into two steps: image inversion and noise suppression, and finally obtains a clear image estimate without noise through alternating iterations of the inversion module and the noise suppression module.

[0066] The specific non-blind deconvolution network model is based on a semi-quadratic splitting scheme to expand the regularization model, which is represented by the following expression:

[0067]

[0068] in represents some image prior {W p} p is a set of 3×3 wavelet high-pass filters. The iterative scheme based on semi-quadratic splitting is as follows: for t=1,…T,

[0069]

[0070]

[0071] where {λ t} t are hyperparameters, simply set to λ0 = 0.5% and λ t = 10%, when t>0. Therefore, the above non-blind deconvolution network model is constructed by alternately stacking the inversion module and the denoising module. Figure 2 shown.

[0072] Inversion Module Corresponding to the first step of the iteration, this is a quadratic problem with an analytical solution. The embodiment of the present invention uses a method based on fast Fourier transform (FFT) to calculate the analytical solution, namely:

[0073]

[0074] Where F stands for Fast Fourier Transform, (·) * represents the conjugation operation of complex numbers, represents bit-by-bit division, and ⊙ represents bit-by-bit multiplication.

[0075] Denoising module Corresponding to the second step of the iteration, a U-Net is used to Remove artifacts and noise.

[0076] As an optional implementation, the denoising module is composed of U-Net as follows Figure 3 As shown. UNet is constructed as follows: the first layer has a convolution kernel size of 3×3, a stride of 1, a padding of 1, an input channel of t×C, and an output channel of 64, followed by 4 consecutive residual channel attention blocks (RCAB). Next, it passes through a convolution kernel size of 3×3, a padding of 1, a stride of 2, and an output channel of 128. The purpose is to reduce the spatial size of the feature map. Next, it is followed by 4 residual channel attention modules with input and output channels of 128. Then it passes through a deconvolution module with a convolution kernel size of 3×3, the purpose of which is to increase the spatial size of the feature map. Then it is followed by 4 residual channel attention modules with input and output channels of 64. Finally, it passes through a convolution kernel size of 3×3, a stride of 1, a padding of 1, an input channel of 128, and an output channel of C. The above-mentioned residual channel attention module is composed of a residual block and a channel self-attention mechanism module, as shown Figure 3 In the figure, Conv refers to the convolution operation with a kernel size of 3×3, a stride of 1, and a padding of 1. GAP refers to Global Average Pooling, Linear refers to full connection, and ⊙ refers to bitwise multiplication. Refers to bitwise addition.

[0077] S3. Perform supervised training on the model on the data in the source domain to obtain a pre-trained model.

[0078] Using the mean square error (MSE) loss L MSE The model is pre-trained as the loss function, where

[0079]

[0080] in is the output of the model, and X is the clear image.

[0081] During pre-training, the Adam optimizer was used, and the two momentum hyperparameters β1 = 0.9, β2 = 0.999. The initial learning rate was 1×10 -4 , train for 250 epochs.

[0082] S4. Apply the pre-trained model to the target domain data and use unsupervised domain adaptive loss to train the model to obtain the migration model.

[0083] Among them, the domain adaptation loss is:

[0084]

[0085] Where M represents a non-blind deconvolution model, whose input includes blurred images and blur kernel Γ1 and Γ2 represent data enhancement methods, including image rotation and folding. represents the image noise, Represents a convolution operation. The essence of the unsupervised domain adaptation function is to use the output of the model in the target domain as a pseudo-label, and then regenerate the blurred image according to the degradation model on the target domain to train the model.

[0086] S5. Determine whether the deconvolution accuracy of the non-blind deconvolution model meets a preset condition.

[0087] In some embodiments, the model is used to deconvolve the verification data in the target domain, and the PSNR of the deconvolution result and the corresponding clear image are calculated and averaged. When the average value of the PSNR is greater than or equal to a third-party preset value, it is determined that the deconvolution accuracy of the model meets the preset conditions, wherein,

[0088]

[0089]

[0090] X stands for clear image, Represents the restored image, MAX represents the maximum value of the pixel, if 8 bits are used, it is 255, m, n represent the length and width of the image respectively.

[0091] S6. Apply the migrated model to the target domain data and output the deconvolution result of the test image.

[0092] In an optional example, the effect of the transferred model on microscopic imaging and remote sensing imaging deconvolution is tested respectively. The visual effect of microscopic image deconvolution is as follows: Figure 5 As shown, Figure 5 (a) is a blurred image. Figure 5 (b) is the image output before model migration. Figure 5 (c) is the output image after model migration. Figure 5 (d) is a clear image. It can be seen that Figure 5 (b) contains a lot of artifacts and noise, Figure 5(c) The noise and artifacts of the output image are effectively suppressed, and the details of the image can be seen more clearly. Figure 6 is the result of remote sensing imaging deconvolution, Figure 6 (a) is a blurred image. Figure 6 (b) is the image output before model migration. Figure 6 (c) is the output image after model migration. Figure 6 (d) is a clear image. The same conclusion can be drawn by comparing the visual effects. Before the model is subjected to domain adaptation, the output of the network will contain a lot of noise and artifacts. After the model is subjected to domain adaptation, the noise and artifacts can be eliminated well. In terms of numerical results, by using the unsupervised adaptive loss proposed in the present invention, the PSNR of the model on the test set of microscope image deconvolution is improved by 0.38 to 0.58 dB. The higher the noise intensity, the more obvious the improvement effect. In remote sensing image deconvolution, the PSNR is improved by 0.05 to 0.64 dB. The higher the noise intensity, the more obvious the improvement effect.

[0093] In summary, the method of this embodiment has the following advantages and beneficial effects compared with the prior art:

[0094] (1) Currently, there is a lack of unsupervised domain-adaptive non-blind image deconvolution methods, and the embodiments of the present invention fill the gap in this field.

[0095] (2) The unsupervised domain adaptation method proposed in the embodiment of the present invention performs domain migration on the pre-trained model without requiring source domain data and clear images of the target domain, thereby improving the non-blind deconvolution performance of the model in the target domain.

[0096] This embodiment also provides an unsupervised domain adaptive image non-blind deconvolution device, including:

[0097] at least one processor;

[0098] at least one memory for storing at least one program;

[0099] When the at least one program is executed by the at least one processor, the at least one processor implements Figure 1 The method shown.

[0100] An unsupervised domain adaptive non-blind image deconvolution device of this embodiment can execute an unsupervised domain adaptive non-blind image deconvolution method provided by the method embodiment of the present invention, can execute any combination of implementation steps of the method embodiment, and has the corresponding functions and beneficial effects of the method.

[0101] The present application also discloses a computer program product or a computer program, which includes a computer instruction stored in a computer-readable storage medium. A processor of a computer device can read the computer instruction from the computer-readable storage medium, and the processor executes the computer instruction, so that the computer device executes Figure 1 The method shown.

[0102] This embodiment also provides a storage medium, which stores instructions or programs that can execute an unsupervised domain adaptive image non-blind deconvolution method provided by an embodiment of the method of the present invention. When the instructions or programs are run, any combination of implementation steps of the method embodiment can be executed, and the corresponding functions and beneficial effects of the method can be obtained.

[0103] In some selectable embodiments, the function / operation mentioned in the block diagram may not occur in the order mentioned in the operation diagram. For example, depending on the function / operation involved, the two boxes shown in succession can actually be executed substantially simultaneously or the boxes can sometimes be executed in reverse order. In addition, the embodiment presented and described in the flow chart of the present invention is provided by way of example, for the purpose of providing a more comprehensive understanding of technology. The disclosed method is not limited to the operation and logic flow presented herein. Selectable embodiments are expected, wherein the order of various operations is changed and the sub-operation of a part for which is described as a larger operation is performed independently.

[0104] In addition, although the present invention is described in the context of functional modules, it should be understood that, unless otherwise specified, one or more of the functions and / or features described may be integrated into a single physical device and / or software module, or one or more functions and / or features may be implemented in separate physical devices or software modules. It is also understood that a detailed discussion of the actual implementation of each module is unnecessary for understanding the present invention. More specifically, in view of the properties, functions, and internal relationships of the various functional modules in the device disclosed herein, the actual implementation of the module will be understood within the conventional skills of the engineer. Therefore, those skilled in the art can implement the present invention set forth in the claims without excessive experimentation using ordinary techniques. It is also understood that the specific concepts disclosed are merely illustrative and are not intended to limit the scope of the present invention, which is determined by the full scope of the appended claims and their equivalents.

[0105] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc., which can store program codes.

[0106] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute instructions), or in conjunction with such instruction execution systems, devices or apparatuses. For the purposes of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in conjunction with such instruction execution systems, devices or apparatuses.

[0107] More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection with one or more wires (electronic device), a portable computer disk case (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be a paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, deciphering or, if necessary, processing in another suitable manner, and then stored in a computer memory.

[0108] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, a plurality of steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0109] In the above description of this specification, the description with reference to the terms "one embodiment / example", "another embodiment / example" or "certain embodiments / examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner.

[0110] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the claims and their equivalents.

[0111] The above is a specific description of the preferred implementation of the present invention, but the present invention is not limited to the above embodiments. Those skilled in the art may make various equivalent modifications or substitutions without violating the spirit of the present invention. These equivalent modifications or substitutions are all included in the scope defined by the claims of this application.

Claims

1. An unsupervised domain adaptive image non-blind deconvolution method, characterized in that: The following steps are involved: Determining source domain data and target domain data, wherein the source domain data includes a blurred image, a blur kernel, and a clear image, and the target domain data includes a blurred image and a blur kernel; Build a non-blind deconvolution model; The source domain data is used to perform supervised training on the non-blind deconvolution model to obtain a pre-trained model; Apply the obtained pre-trained model to the target domain data and use unsupervised adaptive loss to train the pre-trained model to obtain a migration model. Determine whether the deconvolution accuracy of the migration model meets the preset conditions. If so, stop training; otherwise, continue training. Apply the transfer model after transfer training to the target domain data and output the deconvolution result of the test image; The expression of the unsupervised adaptive loss is as follows: Among them, M represents the non-blind deconvolution model, and the input of the non-blind deconvolution model includes the blurred image and blur kernel Γ1 and Γ2 represent data enhancement methods, represents the image noise, represents the convolution operation, Represents the square of the Frobenius norm of the matrix; The non-blind deconvolution model is based on a semi-quadratic splitting scheme to expand the regularization model. The expression of the non-blind deconvolution model is as follows: in, represents some image prior, W p It is a set of wavelet high-pass filters, X represents a clear image, Y represents a blurred image, and K represents a blur kernel. represents the square of the Frobenius norm of the matrix, and p represents the number of the wavelet high-pass filter; The iterative scheme based on semi-quadratic splitting is as follows: for t=1,…T, where λ t is a hyperparameter, T is the total number of iterations, A p Represents a set of auxiliary variables.

2. The unsupervised domain adaptive image non-blind deconvolution method according to claim 1, characterized in that: The non-blind deconvolution model is constructed by alternating and iterating an inversion module and a denoising module; Among them, the inversion module solves the quadratic problem by using the fast Fourier transform method to obtain the inversion result, and the denoising module is composed of a UNet network.

3. The unsupervised domain adaptive image non-blind deconvolution method according to claim 2, characterized in that: The inversion module Corresponding to the first step of the iteration; the denoising module Corresponds to the second step of the iteration.

4. The unsupervised domain adaptive image non-blind deconvolution method according to claim 2, characterized in that: The network structure of the denoising module is as follows: The first convolution module has a convolution kernel size of 3×3, a stride of 1, a padding of 1, an input channel of t×C, and an output channel of 64. The second convolution module has a convolution kernel size of 3×3, a stride of 2, a padding of 1, and an output channel of 128. Deconvolution module, the convolution kernel size of the deconvolution module is 3×3; The third convolution module has a convolution kernel size of 3×3, a stride of 1, a padding of 1, an input channel of 128, and an output channel of C. Four residual channel attention modules with 64 input and output channels are connected in sequence between the first convolution module and the second convolution module; four residual channel attention modules with 128 input and output channels are connected in sequence between the second convolution module and the deconvolution module; four residual channel attention modules with 64 input and output channels are connected in sequence between the deconvolution module and the third convolution module; wherein the residual channel attention module is composed of a residual block and a channel self-attention mechanism module.

5. The unsupervised domain adaptive image non-blind deconvolution method according to claim 1, characterized in that: The step of judging whether the deconvolution accuracy of the migration model meets a preset condition includes: The transfer model is used to deconvolute the verification data in the target domain, and the PSNR of the deconvolution result and the corresponding clear image are calculated and averaged. When the average PSNR is greater than or equal to the preset value, it is determined that the deconvolution accuracy of the transfer model meets the preset conditions.

6. The unsupervised domain adaptive image non-blind deconvolution method according to claim 5, characterized in that: The calculation formula of PSNR is as follows: Among them, X represents a clear image, represents the restored image, MAX represents the maximum value of the pixel, m and n represent the length and width of the image respectively, and i and j represent the horizontal and vertical coordinates of the image pixels respectively.

7. An unsupervised domain adaptive image non-blind deconvolution device, characterized in that: include: at least one processor; at least one memory for storing at least one program; When the at least one program is executed by the at least one processor, the at least one processor implements the method according to any one of claims 1 to 6.

8. A computer-readable storage medium storing a program executable by a processor, characterized in that: The processor-executable program is used to perform the method according to any one of claims 1 to 6 when executed by the processor.

Images