Low-illumination power image enhancement method and system, and storage medium

Abstract

The invention provides a low-illumination power image enhancement method, which comprises the following steps of: performing Retinex decomposition on a low-illumination image by adopting an improved UNet network, and estimating an illumination component; performing gamma correction on the estimated illumination component to obtain a corrected illumination image; and performing element-by-element division operation on the original image and the corrected illumination component to obtain an enhanced reflection component as an output image, thereby realizing low-illumination image enhancement. According to the method, the brightness, contrast and color balance of the image can be automatically adjusted, the visual effect of the low-illumination image is effectively improved, and the purpose of performing low-illumination enhancement only by inputting the low-illumination power image without external data set training is achieved; the challenges of detail loss, color distortion and the like possibly faced by a traditional image enhancement algorithm when a complex illumination condition is processed are overcome, and the efficiency and the quality of image processing are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the field of power inspection image enhancement, and in particular to a low-light power image enhancement method, system and storage medium. Background Art

[0002] According to the public information of relevant research literature and authorized patents at home and abroad, the current implementation scheme of low-light image enhancement technology mainly relies on traditional image processing methods and deep learning technology that has emerged in recent years. Traditional low-light image enhancement algorithms are mainly divided into two categories: distribution mapping-based methods and model optimization-based methods. Among them, the methods based on distributed mapping include histogram equalization, gamma correction, and tone mapping, which directly adjust the pixel values ​​of the image to improve the brightness, contrast and color distribution of the image; while the methods based on model optimization focus on understanding the intrinsic characteristics and degradation process of the image, so as to achieve low-light image enhancement. Common methods include Retinex algorithm and dark channel prior. Although traditional methods can improve the brightness and contrast of images to a certain extent, they often have problems such as loss of details and color distortion when dealing with complex low-light scenes. The introduction of deep learning technology has brought new solutions to low-light image enhancement. These methods can learn the mapping relationship of images from low illumination to normal illumination through training a large number of low-light and normal illumination image pairs, thereby achieving effective enhancement of low-light images.

[0003] However, existing deep learning-based low-light image enhancement methods still face challenges in processing speed, model generalization ability, and detail retention of the enhancement effect. Especially in practical application scenarios such as power inspection, due to the complex image content, variable lighting conditions, and lack of paired data sets, existing methods are difficult to achieve efficient and stable image enhancement while ensuring the enhancement effect. Therefore, there is still a lack of an effective low-light image enhancement technology for power scenes, which restricts the understanding and analysis of power scene images. Summary of the invention

[0004] In view of the problems existing in the prior art, a low-light power image enhancement method, system and storage medium are provided. The image decomposition strategy of Retinex theory is used in combination with deep learning technologies such as convolutional neural networks to generate potential illumination components and reflection components of the image, and then low-light image enhancement is achieved by adjusting the estimated illumination components.

[0005] The first aspect of the present invention provides a low-light power image enhancement method, comprising:

[0006] The improved UNet network is used to perform Retinex decomposition on low-light images to estimate the illumination component;

[0007] Performing gamma correction on the estimated illumination component to obtain a corrected illumination component;

[0008] The original image is divided element-by-element by the corrected illumination component to obtain the enhanced reflection component as the output image, thus achieving low-light image enhancement.

[0009] As a preferred solution, the improved UNet network is used to perform Retinex decomposition on the low-light image to estimate the light component, which specifically includes:

[0010] Gaussian white noise images of two low-light images are obtained based on random sampling, which are used to obtain the illumination component and the reflection component respectively;

[0011] Improved UNet networks for obtaining illumination components and reflection components are constructed respectively, and two Gaussian white noise images are input into the corresponding improved UNet networks respectively;

[0012] The network parameters are iteratively optimized by minimizing the reconstruction error to estimate and generate the potential illumination component and reflection component of the image. The reflection component is used to form a reconstructed image with the illumination component, and the reconstruction loss is obtained by combining with the original image, which is used as the reconstruction error to constrain the generation of the illumination component.

[0013] As a preferred solution, the number of improved UNet network layers used to obtain the illumination component is less than the number of improved UNet network layers used to obtain the reflection component.

[0014] As a preferred solution, in the improved UNet network, a residual branch consisting of a convolution layer and a batch normalization layer is embedded in each layer of the double convolution structure of the encoder. The residual branch connects the input and output ends of the double convolution structure, so that the input feature image can be directly passed to the output, while ensuring that the detail information in the input image is retained and transmitted at multiple levels.

[0015] As a preferred solution, in the improved UNet network, the CBAM attention mechanism is used to realize the jump connection between the feature map in the encoder path and the upsampled feature map in the decoder path.

[0016] As a preferred solution, the reconstruction error includes reconstruction loss, illumination consistency loss, illumination smoothness loss and reflection loss; wherein,

[0017] The image reconstruction loss L1 is:

[0018]

[0019] Among them, G I (z1) represents the generated illumination component, G R(z2) represents the generated reflection component, which ensures the similarity between the original image and the reconstructed image by minimizing the mean square error;

[0020] The lighting consistency loss L2 is:

[0021] L2=||G I (z1)-I0||1

[0022]

[0023] Among them, S c (p) represents the value of the pixel p in the image in the three RGB channels, I0(p) is the maximum value of the pixel p in the image in the three RGB channels, and it is used as the initial illumination value of the pixel. The initial illumination value of each pixel in the image is combined into the initial illumination map I0 of the image. The loss function L2 is used to ensure that the illumination component generated in each iteration is structurally consistent with the initial illumination, thereby maintaining the consistency of illumination distribution during the image enhancement process;

[0024] The reflection loss L3 is:

[0025]

[0026] in, represents the first-order gradient operator,

[0027] The illumination smoothing loss L4 is:

[0028]

[0029] in, represents horizontal and vertical gradients; W is the weight factor;

[0030] The total reconstruction error loss function L that guides the iterative training of the model is:

[0031]

[0032] Among them, λ1, λ2, and λ3 are weight parameters for balancing various losses.

[0033] As a preferred solution, performing gamma correction on the estimated illumination component to obtain the corrected illumination component specifically includes:

[0034] I1=I γ

[0035] Among them, I1 represents the illumination component after gamma correction, γ represents the correction factor, and I is the illumination component output by the improved UNet network.

[0036] As a preferred solution, the original image is divided by the corrected illumination component by an element-by-element operation to obtain an enhanced reflection component as an output image, which specifically includes:

[0037] S 11 =S0 c . / I1,c∈{R,G,B}

[0038] Among them, S1 is the enhanced image, S0 c represents the input low-light image.

[0039] A second aspect of the present invention provides a system, including a memory and a processor, wherein the memory stores a computer program that can be loaded by the processor and executes the low-light power image enhancement method as described in the first aspect.

[0040] A third aspect of the present invention provides a computer-readable storage medium having computer program instructions stored thereon, which, when executed by a processor, are used to implement a process corresponding to the low-light power image enhancement method described in the first aspect.

[0041] Compared with the prior art, the beneficial effects of adopting the above technical solution are:

[0042] (1) Deep learning technology is embedded in the Retinex image enhancement framework. The illumination and reflection components required by the Retinex framework are captured based on image generation technology. No external data is required and only a single low-light power image is needed to achieve image enhancement, which breaks through the traditional method's dependence on data sets.

[0043] (2) An improved UNet network was proposed, which introduced the convolutional block attention mechanism (CBAM) module and residual structure. The illumination component and reflection component were obtained based on the improved model. At the same time, the number of network layers was flexibly adjusted, which could effectively estimate better illumination component and reflection component, and significantly improve the detail clarity and color richness of the enhanced image. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 This is a schematic diagram of the low-light power image enhancement method proposed by the present invention.

[0045] Figure 2 Schematic diagram of the improved UNet network structure proposed in one embodiment of the present invention.

[0046] Figure 3 This is a residual branch double convolution structure proposed in an embodiment of the present invention.

[0047] Figure 4 FIG. 4 is a schematic diagram of the structure of a CBAM module in an embodiment of the present invention. DETAILED DESCRIPTION

[0048] Embodiments of the present application 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 modules or modules with 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 application, and cannot be construed as limitations on the present application. On the contrary, the embodiments of the present application include all changes, modifications and equivalents that fall within the spirit and connotation of the appended claims.

[0049] In order to improve the clarity and detail of power inspection images under low-light conditions, an embodiment of the present invention proposes a low-light power image enhancement method, which aims to improve the quality of images captured during power inspections through intelligent algorithms, especially in low-light environments, to ensure that key elements in the image (such as switchgear, circuit breakers, leads, and equipment fault points, etc.) are clearly presented. The image decomposition strategy based on the Retinex theory is used, and combined with deep learning technologies such as convolutional neural networks, potential illumination components and reflection components of the image are generated, and then low-light image enhancement is achieved by adjusting the estimated illumination components. The method can automatically adjust the brightness, contrast, and color balance of the image, effectively improve the visual effect of the low-light image, and achieve the purpose of low-light enhancement without external data set training, only by inputting low-light power images. In addition, this method optimizes the image decomposition strategy based on the Retinex theory, which not only overcomes the challenges of detail loss and color distortion that traditional image enhancement algorithms may face when dealing with complex lighting conditions, but also significantly improves the efficiency and quality of image processing, providing an efficient and reliable image preprocessing solution for intelligent analysis of power inspection images. Please refer to Figure 1 , the specific plan is as follows:

[0050] Step 1: Use the improved UNet network to perform Retinex decomposition on the low-light image and estimate the illumination component.

[0051] The traditional Retinex theory decomposes an image into two core components, the illumination component and the reflection component, and processes and recombine these two components to obtain the image enhancement effect. In this embodiment, the traditional Retinex model is combined with a deep learning algorithm, and the decomposition problem of the illumination component and the reflection component is converted into a generation problem. The improved UNet network is used to estimate and generate the potential illumination component and reflection component of the image based on the Gaussian white noise images z1 and z2.

[0052] Specifically, the improved UNet network used incorporates the Convolutional Block Attention Module (CBAM) and the residual structure UNet network, which is committed to generating more realistic illumination components and reflection components. At the same time, since the desired illumination component is smooth and the reflection component has rich details, we adjusted the number of layers of the illumination component and reflection component generation network. It should be noted that the improved UNet network for generating reflection components and generating illumination components in this embodiment are different in structure. The number of layers of the designed reflection component generation network is deeper than that of the illumination component generation network, so as to more effectively capture and learn the feature representations of these two components, thereby making the generated illumination components and reflection components more accurate. The specific structure of the improved UNet network in this embodiment is as follows: Figure 2 As shown, it consists of an encoder on the left and a decoder on the right. Each layer of the encoder contains a double convolution structure, and the corresponding decoding layer contains a corresponding deconvolution structure. This embodiment improves the double convolution structure, and the rest of the structure is a conventional design, which will not be described here.

[0053] In particular, in the UNet network of the present application, a residual branch is added to each layer of the double convolution structure of the encoder to form a residual branch double convolution structure, such as Figure 3 The residual branch consists of a convolution layer and a batch normalization layer, which are connected to the input and output of the double convolution structure respectively. This design enables the input feature image to be directly transmitted to the output, while ensuring that the detail information in the input image is retained and transmitted at multiple levels, thereby generating feature-rich illumination components and reflection components, laying the foundation for obtaining more natural and detailed enhanced images.

[0054] Before concatenating the feature map in the encoder path with the upsampled feature map in the decoder path, the present invention applies the CBAM attention mechanism to the feature map of the skip connection, and its structure is as follows: Figure 4As shown. The CBAM module includes a channel attention module and a spatial attention module. The channel attention module first performs average pooling and maximum pooling operations on the feature map, and then sends it to the multi-layer perceptron for processing to generate a channel attention map; the spatial attention module performs maximum pooling and average pooling again based on the feature map refined by the channel attention module, and finally uses the convolution layer to generate the spatial attention map. When generating the illumination component, CBAM dynamically adjusts the weight of each channel through the channel attention module, which can more effectively enhance the brightness of the dark area and avoid overexposure in the bright area, which is crucial for restoring the realism of the image under low light conditions; when generating the reflection component, CBAM uses the spatial attention module to more accurately locate the key areas in the image, which helps to highlight details such as the boundaries of objects, while reducing the impact of noise in the shadow area on the image, making the enhanced image clearer and more natural.

[0055] In the design of the present invention, the reason why the number of layers of the generation networks of the illumination component and the reflection component are different is based on the complexity of the image features they process respectively. The illumination component network is mainly responsible for constructing a smooth illumination map, so its structural design is relatively simple. In contrast, the reflection component network needs to finely capture the inherent properties and texture details in the image, and needs to have more network layers to achieve deeper feature extraction and enhancement. In one embodiment, the network structure for obtaining the illumination component can be set to 4 layers, and the number of channels in each layer is designed to be [8, 16, 32, 64], while the network structure for obtaining the reflection component can be set to 6 layers, and the number of channels in each layer is designed to be [8, 16, 32, 64, 128, 256]. This differentiated network design enables the model to more effectively separate and optimize the illumination and reflection components, thereby improving the overall quality and detail performance of the image.

[0056] It should be noted that in this embodiment, the illumination component and reflection component obtained by the improved UNet network do not need to rely on external data sets to obtain knowledge priors, but learn to generate real illumination components and reflection components by minimizing the reconstruction error. The reflection component obtained by the improved UNet network is mainly used to form a reconstructed image with the illumination component, and combined with the original image to obtain the reconstruction loss, which is used as the reconstruction error to constrain the generation of the illumination component. In this embodiment, the reconstruction error consists of reconstruction loss, illumination consistency loss, illumination smoothness loss and reflection loss.

[0057] The definition of image reconstruction loss L1 is shown in formula (1):

[0058]

[0059] In the formula, G I (z1) represents the generated illumination component, G R(z2) represents the generated reflection component, which ensures the similarity between the original image and the reconstructed image by minimizing the mean squared error (MSE).

[0060] The illumination consistency loss L2 is shown in formula (2):

[0061] L2=‖G I (z1)-I0‖1 (2)

[0062]

[0063] Among them, S c (p) represents the value of pixel p in the image in the three RGB channels, I0(p) is the maximum value of pixel p in the image in the three RGB channels, and it is used as the initial illumination value of the pixel. The initial illumination value of each pixel in the image is combined into the initial illumination map I0 of the image. The loss function L2 ensures that the illumination component generated in each iteration is structurally consistent with the initial illumination, thereby maintaining the consistency of illumination distribution during the image enhancement process.

[0064] The definition of reflection loss L3 is shown in formula (4):

[0065]

[0066] In the formula, Represents the first-order gradient operator. In order to make the generated reflection component close to the ideal noise-free reflection component, we use the total variation loss (TV) as a constraint. This loss function suppresses noise by minimizing the gradient change of the reflection component generated by the model, ensuring that the generated reflection component is smooth and close to the real reflection characteristics.

[0067] The definition of illumination smoothing loss L4 is shown in formula (5):

[0068]

[0069] In the formula, Represents the horizontal and vertical gradients. The ideal illumination component is piecewise smooth and only changes at significant edges. TV loss helps maintain the smoothness of the illumination component and reduce the impact of noise on the image, but over-smoothing may blur the details and clear edges of the image. In order to balance this, a weight factor is introduced to adjust the impact of TV loss. The calculation of W is shown in formula (6).

[0070] Therefore, the total reconstruction error loss function L that guides the iterative training of the model is defined as shown in formula (7):

[0071]

[0072] Where λ1, λ2, and λ3 are weight parameters for balancing the losses.

[0073] Step 2: Perform gamma correction on the estimated illumination component to obtain a corrected illumination component.

[0074] The obtained illumination component estimation can only represent the illumination distribution of the input low-light image, and additional gamma correction is required to obtain an enhanced illumination component image. The gamma correction formula is shown in formula (8):

[0075] I1=I γ (8)

[0076] Among them, I1 represents the illumination component after gamma correction, γ represents the correction factor, and I is the illumination component output by the improved UNet network.

[0077] Step 3: Perform element-by-element division operation on the original image and the corrected illumination component to obtain the enhanced reflection component as the output image, thereby achieving low-light image enhancement.

[0078] For the corrected illumination component I1, the enhanced reflected component is calculated by the following formula:

[0079] S1=S0 c . / I1,c∈{R,G,B} (9)

[0080] Where S1 is the enhanced reflection component, that is, the enhanced image, S0 c represents the input low-light image.

[0081] The present invention has the following advantages:

[0082] (1) Deep learning technology is embedded in the Retinex image enhancement framework. The illumination and reflection components required by the Retinex framework are captured based on image generation technology. No external data is required and only a single low-light power image is needed to achieve image enhancement, which breaks through the traditional method's dependence on data sets.

[0083] (2) An improved UNet network structure was proposed, which introduced the convolutional block attention mechanism (CBAM) module and residual structure. The illumination component and reflection component were obtained based on the improved model. The number of network layers was flexibly adjusted, which could effectively estimate better illumination component and reflection component, and significantly improve the detail clarity and color richness of the enhanced image.

[0084] An embodiment of the present invention further proposes a system, including a memory and a processor, wherein the memory stores a computer program corresponding to the low-light power image enhancement method described above that can be loaded and executed by the processor.

[0085] In particular, according to an embodiment of the present application, the process described above with reference to the flowchart can be implemented as a computer software program. For example, an embodiment of the present application includes a computer program product, which includes a computer program carried on a computer readable medium, and the computer program includes a program code for executing the method shown in the flowchart.

[0086] It should be noted that the computer-readable medium shown in the embodiment of the present application may be a computer-readable signal medium or a computer-readable storage medium or any combination of the above two. The computer-readable storage medium may be, for example, - but not limited to - an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM), a flash memory, an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present application, a computer-readable storage medium may be any tangible medium containing or storing a program, which may be used by an instruction execution system, device or device or used in combination with it. In the present application, a computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, wherein a computer-readable program code is carried. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. Computer-readable signal media may also be any computer-readable medium other than computer-readable storage media, which may send, propagate, or transmit programs for use by or in conjunction with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium may be transmitted using any appropriate medium, including but not limited to: wireless, wired, etc., or any suitable combination of the above.

[0087] The flowchart and block diagram in the accompanying drawings illustrate the possible architecture, functions and operations of the system, method and computer program product according to various embodiments of the present application. Wherein, each box in the flowchart or block diagram can represent a module, a program segment, or a part of the code, and the above-mentioned module, program segment, or a part of the code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram or flowchart, and the combination of boxes in the block diagram or flowchart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0088] The units involved in the embodiments described in this application may be implemented by software or hardware, and the units described may also be set in a processor. The names of these units do not constitute limitations on the units themselves in some cases.

[0089] As another aspect, the present application further provides a computer program product or a computer program, the computer program product or the computer program including computer instructions, the computer instructions being stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device executes the low-light power image enhancement method described in the above embodiment.

[0090] As another aspect, the present application also provides a computer-readable medium, which may be included in the electronic device described in the above embodiment; or may exist independently without being assembled into the electronic device. The above computer-readable medium carries one or more programs, and when the above one or more programs are executed by an electronic device, the electronic device implements the low-light power image enhancement method described in the above embodiment.

[0091] It should be noted that, although several modules or units of the equipment for action execution are mentioned in the above detailed description, this division is not mandatory. In fact, according to the embodiments of the present application, the features and functions of two or more modules or units described above can be embodied in one module or unit. On the contrary, the features and functions of one module or unit described above can be further divided into being embodied by multiple modules or units.

[0092] Through the description of the above implementation methods, it is easy for those skilled in the art to understand that the example implementation methods described here can be implemented by software or by combining software with necessary hardware. Therefore, the technical solution according to the implementation methods of the present application can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, a USB flash drive, a mobile hard disk, etc.) or on a network, and includes several instructions to enable a computing device (which can be a personal computer, a server, a touch terminal, or a network device, etc.) to execute the method according to the implementation methods of the present application.

[0093] For those skilled in the art, the specific meanings of the above terms in the present invention can be understood in specific situations; the drawings in the embodiments are used to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Generally, the components of the embodiments of the present invention described and shown in the drawings herein can be arranged and designed in various different configurations.

[0094] Although the embodiments of the present application have been shown and described above, it can be understood that the above embodiments are exemplary and cannot be understood as limitations on the present application. Ordinary technicians in the field can change, modify, replace and modify the above embodiments within the scope of the present application.

Claims

1. A low-light power image enhancement method, characterized in that: include: The improved UNet network is used to perform Retinex decomposition on low-light images to estimate the illumination component; Performing gamma correction on the estimated illumination component to obtain a corrected illumination component; The original image is divided element-by-element by the corrected illumination component to obtain the enhanced reflection component as the output image, thus achieving low-light image enhancement.

2. The low-light power image enhancement method according to claim 1, characterized in that: The improved UNet network is used to perform Retinex decomposition on the low-light image and estimate the illumination component, specifically including: Gaussian white noise images of two low-light images are obtained based on random sampling, which are used to obtain the illumination component and the reflection component respectively; Improved UNet networks for obtaining illumination components and reflection components are constructed respectively, and two Gaussian white noise images are input into the corresponding improved UNet networks respectively; The network parameters are iteratively optimized by minimizing the reconstruction error to estimate and generate the potential illumination component and reflection component of the image. The reflection component is used to form a reconstructed image with the illumination component, and the reconstruction loss is obtained by combining with the original image, which is used as the reconstruction error to constrain the generation of the illumination component.

3. The low-light power image enhancement method according to claim 2, characterized in that: The number of improved UNet network layers for obtaining the illumination component is less than the number of improved UNet network layers for obtaining the reflection component.

4. The low-light power image enhancement method according to claim 2, characterized in that: In the improved UNet network, a residual branch consisting of a convolution layer and a batch normalization layer is embedded in each double convolution structure of the encoder. The residual branch connects the input and output ends of the double convolution structure to enable the input feature image to be directly transmitted to the output while ensuring that the detail information in the input image is retained and transmitted at multiple levels.

5. The low-light power image enhancement method according to claim 2, characterized in that: In the improved UNet network, the CBAM attention mechanism is used to realize the jump connection between the feature map in the encoder path and the upsampled feature map in the decoder path.

6. The low-light power image enhancement method according to claim 5, characterized in that: The reconstruction error includes reconstruction loss, illumination consistency loss, illumination smoothness loss and reflection loss; wherein, The image reconstruction loss L1 is: Among them, G I (z1) represents the generated illumination component, G R (z2) represents the generated reflection component, which ensures the similarity between the original image and the reconstructed image by minimizing the mean square error; The lighting consistency loss L2 is: <h2 style=";text-align:left;direction:ltr">L2=||G<h2 style=";text-align:left;direction:ltr"> I <h2 style=";text-align:left;direction:ltr"> (z1)-I0||1 Among them, S c (p) represents the value of the pixel p in the image in the three RGB channels, I0(p) is the maximum value of the pixel p in the image in the three RGB channels, and it is used as the initial illumination value of the pixel. The initial illumination value of each pixel in the image is combined into the initial illumination map I0 of the image. The loss function L2 is used to ensure that the illumination component generated in each iteration is structurally consistent with the initial illumination, thereby maintaining the consistency of illumination distribution during the image enhancement process; The reflection loss L3 is: in, represents the first-order gradient operator, The illumination smoothing loss L4 is: in, represents horizontal and vertical gradients; W is the weight factor; The total reconstruction error loss function L that guides the iterative training of the model is: Among them, λ1, λ2, and λ3 are weight parameters for balancing various losses.

7. The low-light power image enhancement method according to claim 1, characterized in that: The performing gamma correction on the estimated illumination component to obtain the corrected illumination component specifically includes: I1=I γ Among them, I1 represents the illumination component after gamma correction, γ represents the correction factor, and I is the illumination component output by the improved UNet network.

8. The low-light power image enhancement method according to claim 1, characterized in that: Perform an element-by-element division operation on the original image and the corrected illumination component to obtain the enhanced reflection component as the output image, which includes: S 11 =S0 c . / I1,c∈{R,G,B} Among them, S1 is the enhanced image, S0 c represents the input low-light image.

9. A system, characterized in that: The method comprises a memory and a processor, wherein the memory stores a computer program corresponding to the low-light power image enhancement method according to any one of claims 1 to 8 that can be loaded by the processor and executed.

10. A computer-readable storage medium having computer program instructions stored thereon, characterized in that: When the program instructions are executed by a processor, they are used to implement a process corresponding to the low-light power image enhancement method according to any one of claims 1 to 8.

Images