Explainability adverse environment image enhancement method, system, device and medium

Abstract

The application discloses an interpretable harsh environment image enhancement method, system, device and medium, and relates to the technical field of image processing.The method comprises the following steps: inputting a harsh environment image; separating the harsh environment image into a background layer and a noise layer; denoising the noise layer to obtain a denoised layer; and obtaining an enhanced harsh environment image according to the background layer and the denoised layer.The application introduces a neural network unfolding technology, and unfolds a traditional optimization iterative algorithm into a deep neural network, wherein each layer of the neural network can be regarded as a solving process of the optimization iterative algorithm, so that a deep neural network operator with certain interpretability is constructed, and an interpretable harsh environment image enhancement algorithm is realized through combination of the interpretability operator.

Description

Technical Field

[0001] This invention relates to the field of image processing technology, and more specifically to an interpretable method, system, device, and medium for enhancing images in harsh environments. Background Technology

[0002] In smart cities and intelligent transportation, on-site information monitoring and identification are essential. However, existing surveillance and video image recognition systems typically assume that the input video images have clear visual features, and the training data used for these algorithms is generally high-quality video or images. But in real-world applications, many environments present harsh conditions, such as low light or fog. Recognition algorithms trained on clear images or videos will face decreased accuracy or even failure in these challenging environments. Therefore, we need to enhance images used in harsh environments.

[0003] Existing image enhancement algorithms for harsh environments are mainly based on training deep neural network models. However, the mechanisms of deep neural networks are still not fully understood. How to construct the neural network structure and set the parameters can generally only be done through trial and error, and the trained image enhancement algorithms lack interpretability. Summary of the Invention

[0004] To address the shortcomings of existing harsh environment image enhancement algorithms, which rely on deep neural network models and lack interpretability, this invention introduces neural network unfolding technology. This technology unfolds traditional optimization iterative algorithms into deep neural networks, where each layer of the neural network can be viewed as a solution process of the optimization iterative algorithm. This constructs a deep neural network operator with a certain degree of interpretability. By combining interpretable operators, an interpretable harsh environment image enhancement algorithm is achieved.

[0005] To achieve the above objectives, the present invention provides the following technical solution:

[0006] In a first aspect, the present invention provides an interpretable image enhancement method for harsh environments, comprising:

[0007] Input images of harsh environments;

[0008] Separate the background layer and noise layer from the harsh environment image;

[0009] The noise layer is then denoised to obtain a denoised layer.

[0010] An enhanced image of the harsh environment is obtained based on the background layer and the denoising layer.

[0011] Secondly, the present invention provides an interpretable image enhancement system for harsh environments, comprising:

[0012] A layer decomposition module is used to receive an input harsh environment image and separate the harsh environment image into a background layer and a noise layer;

[0013] A multi-environment harsh environment image enhancement module is used to denoise the noise layer to obtain a denoised layer, and obtain an enhanced harsh environment image based on the background layer and the denoised layer.

[0014] Thirdly, the present invention provides an electronic device comprising a processor and a memory, wherein the memory stores at least one instruction, at least one program, a code set, or an instruction set, wherein the at least one instruction, the at least one program, the code set, or the instruction set is loaded and executed by the processor to implement the interpretable harsh environment image enhancement method described above.

[0015] Fourthly, the present invention provides a computer-readable storage medium storing at least one instruction, at least one program, a code set, or an instruction set, wherein the at least one instruction, the at least one program, the code set, or the instruction set is loaded and executed by a processor to implement the interpretable harsh environment image enhancement method described above.

[0016] Compared with the prior art, the advantages of this invention are as follows:

[0017] The method of this invention introduces neural network unfolding technology, which unfolds the traditional optimization iterative algorithm into a deep neural network. Each layer of the neural network can be regarded as a solution process of the optimization iterative algorithm, thereby constructing a deep neural network operator with a certain degree of interpretability. By combining the interpretable operators, an interpretable image enhancement algorithm for harsh environments is realized.

[0018] The method of this invention is geared towards image enhancement in harsh environments. It utilizes a layer separation operator based on a neural network unfolding method to enable the construction and parameter setting of an interpretable deep neural network structure.

[0019] The method of this invention can enhance images in harsh environments such as low light, fog, and rain, and has versatility in solving various image enhancement problems in harsh environments. It has broad application prospects in smart cities and smart transportation. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 This is a flowchart of an interpretable harsh environment image enhancement method according to an embodiment of the present invention;

[0022] Figure 2 This is an algorithm structure diagram of the interpretable harsh environment image enhancement method in this embodiment of the invention;

[0023] Figure 3 This is a schematic diagram of the structure of an electronic device implementing the method in the embodiments of the invention. Detailed Implementation

[0024] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0025] Example:

[0026] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, in the embodiments of this invention are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0027] In the description of this invention, "a plurality of" means at least two, such as two, three, etc., unless otherwise explicitly specified. Furthermore, unless otherwise explicitly specified and limited, the terms "installed," "connected," and "linked" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0028] The word “exemplary” as used below means “serving as an example, embodiment, or illustration.” Any embodiment illustrated as an “exemplary” need not be construed as superior to or better than other embodiments.

[0029] This invention primarily targets harsh environmental conditions such as low light, fog, and rain. By combining a neural network unfolding method, it separates noise layers that affect image quality and performs denoising processing, thereby obtaining an interpretable image enhancement algorithm for harsh environments.

[0030] See Figure 1 and Figure 2 , Figure 1 A flowchart illustrating an interpretable image enhancement method for harsh environments based on neural network expansion is presented in an embodiment of the present invention. Figure 2 The overall algorithm structure diagram of the interpretable harsh environment image enhancement method based on neural network expansion in this embodiment of the invention is shown.

[0031] An interpretable method for image enhancement in harsh environments, comprising:

[0032] Input images of harsh environments;

[0033] Separate the background layer and noise layer from the harsh environment image;

[0034] The noise layer is then denoised to obtain a denoised layer.

[0035] An enhanced image of the harsh environment is obtained based on the background layer and the denoising layer.

[0036] Optionally, after obtaining the enhanced harsh environment image, the method further includes:

[0037] Feature similarity is obtained based on the enhanced harsh environment image and the reference image;

[0038] The feature similarity is used as a quality evaluation score;

[0039] The quality evaluation score is added to the loss function to construct an objective function with image quality evaluation characteristics;

[0040] The specific parameters in the image enhancement process are adjusted using the objective function.

[0041] Optionally, separating the background layer and the noise layer specifically includes:

[0042] A layer decomposition module that generates a neural network-based layer decomposition operator;

[0043] The layer decomposition module, using the neural network-based decomposition operator, decomposes the harsh environment image I into a background layer B, which is unaffected by the harsh environment, and a noise layer R, which is affected by the harsh environment.

[0044] Where, B, R = UnrollingDecom(I)

[0045] UnrollingDecom() represents the layer decomposition module based on the neural network unfolding layer decomposition operator.

[0046] Optional,

[0047] The image of the harsh environment can be represented as:

[0048] I = B + R

[0049] Where I represents the image of a harsh environment, B represents the background layer, and R represents the noise layer;

[0050] Layer decomposition is performed using a layer decomposition operator based on neural network expansion, with the following constraint form:

[0051]

[0052] Where, λ B w B (J) represents the constraint terms related to the background layer B, λ R w R (K) represents the constraint terms associated with the noise layer R, λ B and λ R These are the coefficients of the constraint terms for the background layer and the noise layer, respectively.

[0053] The Lagrange augmented form of the above constraints:

[0054]

[0055] Where, μ B μ R α and β represent the constraint coefficients of the Lagrange augmented form, and J and K represent the intermediate variables of the background layer B and the noise layer R during the solution process, respectively.

[0056] Solving using the ADMM method, we can obtain...

[0057]

[0058] Differentiating the above equation and setting the derivative to 0, we get...

[0059]

[0060] An iterative solution is performed, updating the output image u, intermediate variable Z, and constraint coefficients μ and α sequentially in each iteration. The iterative calculation expression can be obtained as follows:

[0061]

[0062] Among them, J i and R iLet J and K represent the intermediate variables J and K respectively in the i-th iteration, and μ represent the intermediate variables J and K respectively. B i and μ R i Let μ and μ represent the constraint coefficients in the i-th iteration, respectively. B and μ R α i and β i B represents the constraint coefficients α and β in the i-th iteration, respectively; i+1 R i+1 J i+1 K i +1 μ B i+1 μ R i+1 , αi +1 and β i+1 Let B represent the background layer and noise layer R, the intermediate variables J and K, and the constraint coefficient μ in the (i+1)th iteration, respectively. B μ R α and β, and P() and Q() are the shrink functions respectively.

[0063] Optionally, obtaining the enhanced harsh environment image specifically includes:

[0064] A module for generating image enhancement in harsh environments;

[0065] The noise layer R is denoised using the multi-environment harsh image enhancement module to obtain a denoised layer R. en , where R en =MultiEnv(R)

[0066] MultiEnv() is the multi-environment harsh image enhancement module.

[0067] Optionally, obtaining the enhanced harsh environment image specifically includes:

[0068] The background layer B and the noise reduction layer R en Summation is performed to obtain the enhanced harsh environment image I. en :

[0069] I en =B+R en .

[0070] Optionally, obtaining feature similarity specifically includes:

[0071] Generate an image quality assessment module based on neural network expansion;

[0072] The enhanced harsh environment image I is input into the neural network-based image quality assessment module. en and the reference image I ref ;

[0073] The enhanced harsh environment image I is processed by the layer decomposition operator based on neural network expansion. en and the reference image I ref Layer decomposition and feature extraction are performed to obtain the relevant layer features F. en and F ref ,

[0074] F en =DecomFea(I en )

[0075] F ref =DecomFea(I ref )

[0076] DecomFea() represents the unit in the image quality assessment module that performs layer decomposition and feature extraction;

[0077] According to the layer feature F en and F ref Obtain the feature similarity D;

[0078] D = |F en -F ref | 2 .

[0079] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. In one embodiment, the interpretable harsh environment image enhancement method based on neural network expansion may include the following steps:

[0080] Step 1: Input a harsh environment image, and use the harsh environment layer decomposition module based on neural network unfolding layer decomposition operator to decompose the harsh environment image, and decompose it into a background layer that is not affected by the harsh environment and a noise layer that is affected by the harsh environment.

[0081] Step 2: Use the neural network multi-environment harsh environment image enhancement module to denoise the noise layer to obtain a denoised layer. Combine the denoised layer and the background layer to obtain the enhanced harsh environment image.

[0082] Step 3: Input the enhanced harsh environment image and the reference image into the image quality evaluation module based on neural network expansion. Use the neural network expansion layer decomposition operator to separate and extract the corresponding features of the enhanced image and the reference image. Calculate the image quality of the enhanced image based on feature similarity. Construct an objective function with image quality evaluation characteristics to guide the parameter tuning of the harsh environment image enhancement algorithm.

[0083] In the above embodiments, low-light images, foggy images, and rainy images can all be mathematically represented as the impact of noise layers on the original image content:

[0084] I = B + R

[0085] Where I represents an image affected by adverse environments such as low light, fog, and rain, B represents the background layer of the image, and R is the noise layer caused by the adverse environment.

[0086] Therefore, low-light enhancement, dehazing, and deraining algorithms can all be mathematically described as a process of image layer decomposition and noise layer removal. This layer decomposition process can be implemented using a deep neural network, and thus an interpretable layer decomposition operator can be constructed by performing neural network unrolling techniques on this layer decomposition network.

[0087] In the above embodiments, the layer decomposition using a layer decomposition operator based on neural network expansion is based on an optimization iterative method, with the following constraint form:

[0088]

[0089] Where I represents an image affected by adverse environmental conditions such as low light, fog, and rain, B represents the background layer of the image, R is the noise layer caused by the adverse environment, and λ B w B (J) represents the constraint terms related to the background layer B, λ R w R (K) represents the constraint terms associated with the noise layer R, λ B and λ R These are the coefficients for the constraint terms of the background layer and the noise layer, respectively.

[0090] To solve the above constraints, write the Lagrange augmented form of the constraints:

[0091]

[0092] Where, μ B μ R α and β represent the constraint coefficients of the Lagrange augmented form, and J and K represent the intermediate variables of the background layer B and the noise layer R during the solution process, respectively.

[0093] Solving using the ADMM method, we can obtain...

[0094]

[0095] Differentiating the above equation and setting the derivative to 0, we get...

[0096]

[0097] An iterative solution is performed, updating the output image u, intermediate variable Z, and constraint coefficients μ and α sequentially in each iteration. The iterative calculation expression can be obtained as follows:

[0098]

[0099] Among them, J i and R i Let J and K represent the intermediate variables J and K respectively in the i-th iteration, and μ represent the intermediate variables J and K respectively. B i and μ R i Let μ and μ represent the constraint coefficients in the i-th iteration, respectively. B and μ R α i and β i Let α and β represent the constraint coefficients in the i-th iteration, respectively. i+1 R i+1 J i+1 K i +1 μ B i+1 μ R i+1 α i+1 and β i+1 Let B represent the background layer and noise layer R, the intermediate variables J and K, and the constraint coefficient μ in the (i+1)th iteration, respectively. B μ R α and β, P() and Q() are shrink functions respectively.

[0100] In one embodiment, to achieve better processing results, functions P() and Q() can each use a fully convolutional neural network to calculate the output results.

[0101] In the neural network unfolding structure, each layer of the layer decomposition operator is one of the above iterative calculation expressions. In this way, layer decomposition can be regarded as an iterative multi-scale image filtering process, which realizes the interpretation of layer decomposition based on deep neural networks.

[0102] Therefore, based on the above derivation process, a harsh environment layer decomposition module based on a neural network unfolding layer separation operator can be obtained. The harsh environment layer decomposition module takes the harsh environment image I as input and decomposes it to obtain a background layer B unaffected by the harsh environment and a noise layer R affected by the harsh environment.

[0103] B, R = UnrollingDecom(I)

[0104] Wherein, UnrollingDecom() represents the layer decomposition module based on the neural network unfolding layer decomposition operator.

[0105] Then, after obtaining the noise layer R through the harsh environment layer decomposition module, the noise layer R is input into the multi-environment image enhancement module to obtain the enhanced denoised layer R. en :

[0106] R en =MultiEnv(R)

[0107] MultiEnv() is a multi-environment image enhancement module. To increase the versatility of MultiEnv() in different harsh environments and make it adaptable to various harsh environments, we utilize a multi-scale neural network model that can process a variety of different features.

[0108] Finally, the denoised layer R is obtained through the MultiEnv() module for multi-environment image enhancement. en Then, combine the background layer B and the noise reduction layer R. en Summation yields the enhanced image I for harsh environments. en :

[0109] I en =B+R en

[0110] Therefore, in order to improve the enhancement effect of our model on images in harsh environments, we use the layer decomposition operator based on neural network expansion mentioned above to construct an image quality evaluation module for the enhanced image, which is used to guide the image enhancement algorithm for harsh environments.

[0111] In some embodiments, the image quality evaluation module further inputs the enhanced image I. en and reference image I ref First, the layer decomposition operator based on neural network expansion mentioned above is used to enhance the image I. en and reference image I ref Perform layer decomposition and feature extraction to obtain relevant layer features F en and F ref :

[0112] F en =DecomFea(I en )

[0113] F ref =DecomFea(I ref )

[0114] Here, DecomFea() represents the module in the image quality assessment module that performs layer decomposition and feature extraction.

[0115] Obtain layer features F en and F ref Then, calculate the feature similarity D:

[0116] D = |F en -F ref | 2

[0117] After obtaining the feature similarity D, the feature similarity D is used as the quality evaluation score S.

[0118] After obtaining the quality evaluation score S, the quality evaluation score S is added to the loss function to construct an objective function with image quality evaluation characteristics to guide the parameter tuning of the harsh environment image enhancement algorithm, thereby realizing interpretable harsh environment image enhancement.

[0119] Based on the same inventive concept, this invention also provides an interpretable harsh environment image enhancement system, which includes: a layer decomposition module and a multi-environment harsh image enhancement module. Specifically, the layer decomposition module is used to receive an input harsh environment image and separate a background layer and a noise layer from the harsh environment image; the multi-environment harsh image enhancement module is used to denoise the noise layer to obtain a denoised layer, and obtain an enhanced harsh environment image based on the background layer and the denoised layer.

[0120] Since this system corresponds to the interpretable harsh environment image enhancement method of the present invention, and the principle of solving the problem in this system is similar to that of the method, the implementation of this system can refer to the implementation process of the above method embodiments, and the repeated parts will not be described again.

[0121] See Figure 3 Based on the same inventive concept, embodiments of the present invention also provide an electronic device, the electronic device including a processor and a memory, the memory storing at least one instruction, at least one program, code set or instruction set, the at least one instruction, the at least one program, the code set or instruction set being loaded and executed by the processor to realize the interpretable harsh environment image enhancement method as described above.

[0122] It is understood that the memory may include random access memory (RAM) or read-only memory. Optionally, the memory may include non-transitory computer-readable storage medium. The memory can be used to store instructions, programs, code, code sets, or instruction sets. The memory may include a stored program area and a stored data area, wherein the stored program area may store instructions for implementing an operating system, instructions for at least one function, instructions for implementing the various method embodiments described above, etc.; the stored data area may store data created according to the use of the server, etc.

[0123] A processor may include one or more processing cores. The processor connects to various parts of the server via various interfaces and lines, performing various server functions and processing data by running or executing instructions, programs, code sets, or instruction sets stored in memory, and by accessing data stored in memory. Optionally, the processor may be implemented using at least one of the following hardware forms: Digital Signal Processing (DSP), Field-Programmable Gate Array (FPGA), and Programmable Logic Array (PLA). The processor may integrate one or more of the following: a Central Processing Unit (CPU) and a modem. The CPU primarily handles the operating system and applications; the modem handles wireless communication. It is understood that the modem may also be implemented as a separate chip without being integrated into the processor.

[0124] Since this electronic device is the electronic device corresponding to the interpretable harsh environment image enhancement method of the present invention, and the principle of the electronic device in solving the problem is similar to that of the method, the implementation of this electronic device can refer to the implementation process of the above method embodiments, and the repeated parts will not be described again.

[0125] Based on the same inventive concept, embodiments of the present invention also provide a computer-readable storage medium storing at least one instruction, at least one program, code set, or instruction set, wherein the at least one instruction, the at least one program, the code set, or the instruction set is loaded and executed by a processor to implement the interpretable harsh environment image enhancement method described above.

[0126] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, including read-only memory (ROM), random access memory (RAM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), one-time programmable read-only memory (OTPROM), electrically-Erasable Programmable Read-Only Memory (EEPROM), compact disc read-only memory (CD-ROM) or other optical disc storage, disk storage, magnetic tape storage, or any other computer-readable medium capable of carrying or storing data.

[0127] Since this storage medium is the storage medium corresponding to the interpretable harsh environment image enhancement method of the present invention, and the principle of the storage medium in solving the problem is similar to that of the method, the implementation of this storage medium can refer to the implementation process of the above method embodiments, and the repeated parts will not be described again.

[0128] In some possible implementations, various aspects of the methods of the embodiments of the present invention can also be implemented as a program product comprising program code that, when run on a computer device, causes the computer device to perform the steps of the interpretable harsh environment image enhancement method according to various exemplary embodiments of the present application described above. The executable computer program code or "code" for performing the various embodiments can be written in high-level programming languages ​​such as C, C++, C#, Smalltalk, Java, JavaScript, Visual Basic, Structured Query Language (e.g., Transact-SQL), Perl, or in various other programming languages.

[0129] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0130] The above embodiments are merely illustrative of the technical concept and features of the present invention, and are intended to enable those skilled in the art to understand the content of the present invention and implement it accordingly. They should not be construed as limiting the scope of protection of the present invention. All equivalent changes or modifications made based on the essence of the content of the present invention should be covered within the scope of protection of the present invention.

Claims

1. An interpretable method for enhancing images in harsh environments, characterized in that, include: Input images of harsh environments; The harsh environment image is separated into a background layer and a noise layer, wherein separating the background layer and the noise layer includes: generating a layer decomposition module based on a neural network unfolding layer decomposition operator; using the layer decomposition module based on the neural network unfolding layer decomposition operator to decompose the harsh environment image I, decomposing the background layer B which is not affected by the harsh environment and the noise layer R which is affected by the harsh environment; The image of the harsh environment is represented as follows: I = B + R Where I represents the image of a harsh environment, B represents the background layer, and R represents the noise layer; Layer decomposition is performed using a layer decomposition operator based on neural network expansion, with the following constraint form: Where, λ B w B (J) represents the constraint terms related to the background layer B, λ R w R (K) represents the constraint terms associated with the noise layer R, λ B and λ R These are the coefficients of the constraint terms for the background layer and the noise layer, respectively. The Lagrange augmented form of the above constraints: Where, μ B μ R α and β represent the constraint coefficients of the Lagrange augmented form, and J and K represent the intermediate variables of the background layer B and the noise layer R during the solution process, respectively. Solving using the ADMM method, we can obtain... Differentiating the above equation and setting the derivative to 0, we get... By performing iterative solutions, in each iteration, the output image u, the intermediate variable Z, and the constraint coefficients μ and α are updated sequentially; the iterative calculation expression can be obtained as follows: Among them, J i and R i Let J and K represent the intermediate variables J and K respectively in the i-th iteration, and μ represent the intermediate variables J and K respectively. B i and μ R i Let μ and μ represent the constraint coefficients in the i-th iteration, respectively. B and μ R α i and β i B represents the constraint coefficients α and β in the i-th iteration, respectively; i+1 R i+1 J i+1 K i+1 μ B i+1 μ R i+1 α i+1 and β i+1 Let B represent the background layer and noise layer R, the intermediate variables J and K, and the constraint coefficient μ in the (i+1)th iteration, respectively. B μ R α and β, P() and Q() are the shrink functions respectively; The noise layer is then denoised to obtain a denoised layer. An enhanced image of the harsh environment is obtained based on the background layer and the denoising layer.

2. The interpretable harsh environment image enhancement method according to claim 1, characterized in that, After obtaining the enhanced image of the harsh environment, the process further includes: Feature similarity is obtained based on the enhanced harsh environment image and the reference image; The feature similarity is used as a quality evaluation score; The quality evaluation score is added to the loss function to construct an objective function with image quality evaluation characteristics; The specific parameters in the image enhancement process are adjusted using the objective function.

3. The interpretable harsh environment image enhancement method according to claim 1, characterized in that, The background layer B and the noise layer R affected by the harsh environment are specifically as follows: B,R = UnrollingDecom(I) UnrollingDecom() represents the layer decomposition module based on the neural network unfolding layer decomposition operator.

4. The interpretable harsh environment image enhancement method according to claim 1, characterized in that, The process of obtaining the enhanced image of the harsh environment specifically includes: A module for generating image enhancement in harsh environments; The noise layer R is denoised using the multi-environment harsh image enhancement module to obtain a denoised layer R. en , Among them, R en =MultiEnv(R) MultiEnv() is the multi-environment harsh image enhancement module.

5. The interpretable harsh environment image enhancement method according to claim 1, characterized in that, The process of obtaining the enhanced image of the harsh environment specifically includes: The background layer B and the noise reduction layer R en Summation is performed to obtain the enhanced harsh environment image I. en : I en =B+R en 。 6. The interpretable harsh environment image enhancement method according to claim 2, characterized in that, The acquisition of feature similarity specifically includes: Generate an image quality assessment module based on neural network expansion; The enhanced harsh environment image I is input into the neural network-based image quality assessment module. en and the reference image I ref ; The enhanced harsh environment image I is processed by the layer decomposition operator based on neural network expansion. en and the reference image I ref Layer decomposition and feature extraction are performed to obtain the relevant layer features F. en and F ref , F en =DecomFea(I en ) F ref =DecomFea(I ref ) DecomFea() represents the unit in the image quality assessment module that performs layer decomposition and feature extraction; According to the layer feature F en and F ref Obtain the feature similarity D; D=|F en -F ref | 2 。 7. An interpretable image enhancement system for harsh environments, characterized in that, include: A layer decomposition module is used to receive an input harsh environment image and separate the harsh environment image into a background layer and a noise layer; The separation of the background layer and the noise layer includes: generating a layer decomposition module based on a neural network unfolding layer decomposition operator; using the layer decomposition module based on the neural network unfolding layer decomposition operator to decompose the harsh environment image I, decomposing the background layer B which is not affected by the harsh environment and the noise layer R which is affected by the harsh environment; The image of the harsh environment is represented as follows: I = B + R Where I represents the image of a harsh environment, B represents the background layer, and R represents the noise layer; Layer decomposition is performed using a layer decomposition operator based on neural network expansion, with the following constraint form: Where, λ B w B (J) represents the constraint terms related to the background layer B, λ R w R (K) represents the constraint terms associated with the noise layer R, λ B and λ R These are the coefficients of the constraint terms for the background layer and the noise layer, respectively. The Lagrange augmented form of the above constraints: Where, μ B μ R α and β represent the constraint coefficients of the Lagrange augmented form, and J and K represent the intermediate variables of the background layer B and the noise layer R during the solution process, respectively. Solving using the ADMM method, we can obtain... Differentiating the above equation and setting the derivative to 0, we get... By performing iterative solutions, in each iteration, the output image u, the intermediate variable Z, and the constraint coefficients μ and α are updated sequentially; the iterative calculation expression can be obtained as follows: Among them, J i and R i Let J and K represent the intermediate variables J and K respectively in the i-th iteration, and μ represent the intermediate variables J and K respectively. B i and μ R i Let μ and μ represent the constraint coefficients in the i-th iteration, respectively. B and μ R α i and β i B represents the constraint coefficients α and β in the i-th iteration, respectively; i+1 R i+1 J i+1 K i+1 μ B i+1 μ R i+1 α i+1 and β i+1 Let B represent the background layer and noise layer R, the intermediate variables J and K, and the constraint coefficient μ in the (i+1)th iteration, respectively. B μ R α and β, P() and Q() are the shrink functions respectively; A multi-environment harsh environment image enhancement module is used to denoise the noise layer to obtain a denoised layer, and obtain an enhanced harsh environment image based on the background layer and the denoised layer.

8. An electronic device, characterized in that, The electronic device includes a processor and a memory, wherein the memory stores at least one instruction, at least one program, a code set, or an instruction set, and the at least one instruction, the at least one program, the code set, or the instruction set is loaded and executed by the processor to implement the interpretable harsh environment image enhancement method as described in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, The storage medium stores at least one instruction, at least one program, code set, or instruction set, wherein the at least one instruction, the at least one program, the code set, or instruction set is loaded and executed by a processor to implement the interpretable harsh environment image enhancement method according to any one of claims 1 to 6.

Images