A neural network-based underwater degraded image polarization recovery method

Abstract

The application provides a neural network-based underwater degraded image polarization recovery method and system, and relates to the field of underwater image restoration. The method first collects polarized images in a real underwater environment through a polarized camera to form a training data set. Then, a deep learning despeckling network is designed. The network is based on a polarized image degradation model and includes multiple parts, such as a channel attention mechanism module and a dense residual module, to enhance the ability to capture multi-scale features and improve the performance and training efficiency of the network. Finally, a joint loss function is designed by combining polarization and intensity information to guide the network training process and ensure that the restored image is optimal in terms of polarization and intensity.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of underwater image restoration, and particularly relates to a polarized recovery method for underwater degraded images based on a neural network. BACKGROUND

[0002] Underwater image restoration technology plays a crucial role in many key fields, especially in complex underwater operations such as underwater rescue, marine resource exploration, and underwater survey. However, the scattering and absorption of light by suspended particles in water can weaken the light signal received by imaging devices, resulting in severe degradation of underwater image quality, such as insufficient contrast, unclear edges, and shortened viewing distance. In the face of the challenges of "unclear vision and short viewing distance" caused by turbid water and scattering, we urgently need to adopt innovative optical information capture technology and advanced image processing methods to solve the image degradation problem in turbid water environments. This includes restoring target images from underwater scenes, eliminating the effects of backscattered light, and improving the contrast and clarity of underwater images.

[0003] In recent years, various methods have been proposed to improve underwater imaging quality. For example, histogram stretching, image filtering, Retinex imaging, dark channel prior-based method, and underwater polarization imaging. Compared to other methods, underwater polarization imaging has the advantage of accurately estimating the difference and connection between target information light and backscattered light, which can achieve the purpose of suppressing scattered light and improving imaging clarity, making it stand out among a variety of underwater imaging methods.

[0004] Currently, various polarization-based underwater image restoration methods have been proposed, among which the most classic model is the polarization imaging model proposed by Schechner et al. This method extracts polarization parameters from multiple polarization images to restore clear underwater images, improving the quality of underwater images and proving the superiority of using polarization technology underwater. However, the traditional underwater polarization imaging physical model is a simplification or idealization to some extent. For example, Treibitz et al. consider the polarization degree of backscattered light as a single value, and estimate the polarization degree of backscattered light using the background of the local area. However, the polarization degree of backscattered light varies across the image plane, and considering it as a fixed single value may have a large deviation from the actual situation. Hu et al. estimate the polarization degree of backscattered light based on parameter fitting, and use a polynomial to fit the polarization degree of backscattered light in the background area to estimate the polarization degree of the target area. However, the target area polarization degree estimated by the backscattered light polarization degree fitting plane still has a large error, and the target information light interpretation is still not accurate, the target information light extraction is still not accurate, and the image clarity still needs to be further improved. SUMMARY

[0005] To solve the above technical problems, the application provides a polarimetric recovery method and system for underwater degraded images based on a neural network.

[0006] The first aspect of the application discloses a polarimetric recovery method for underwater degraded images based on a neural network; the method comprises:

[0007] In step S1, polarized images in a real water environment are collected and preprocessed to form a training data set; the sample labels of each data sample in the training data set include intensity images, polarization degree images and polarization angle images;

[0008] In step S2, a despeckling network is constructed based on a polarized image degradation model; the despeckling network comprises shallow feature extraction modules, channel attention mechanism modules, dense residual modules, global feature fusion modules and reconstruction modules connected in sequence;

[0009] In step S3, the samples in the training data set are input into the despeckling network in sequence to obtain enhanced results;

[0010] In step S4, based on the enhanced results and the sample labels, the joint loss function value of polarization and intensity is calculated, and then the gradient of the despeckling network is calculated by back propagation according to the joint loss function, and then the optimized despeckling network is obtained;

[0011] In step S5, the target image is input into the optimized despeckling network to obtain the recovered image.

[0012] In the step S1, all polarized images are cropped horizontally and vertically to obtain a training data set containing pairs of underwater polarized images.

[0013] In the step S2, the shallow feature extraction module contains two convolution layers, and the size of the convolution kernel is 3x3, which is used to extract the shallow feature information of the polarized image to obtain a feature map.

[0014] In the step S2, the channel attention mechanism module uses a global average pooling layer to reduce the feature map to a dimension of 1x1xC, where C represents the number of channels of the feature map; then, weight learning is performed on each channel feature to adjust the weight of each feature in the feature map.

[0015] In the step S2, the dense residual module comprises a plurality of residual modules, each of which extracts features through a series of convolution layers and activation layers, wherein the output of each convolution layer is spliced with the input of the subsequent layer to realize dense connection of the features; at the same time, each residual module internally contains a residual connection, which directly adds the input to the output to promote the reuse of features and the accumulation of information.

[0016] The global feature fusion module firstly splices the outputs of all residual blocks in the channel dimension through a connection layer to combine the feature maps of different levels together, and then the spliced feature maps pass through two consecutive convolution layers to further extract and fuse the features.

[0017] The reconstruction module uses up-sampling to reconstruct a high-resolution image of the target sample size.

[0018] In step S4, the joint loss function L is:

[0019] L=L cont +λ×L polar

[0020] Wherein, λ represents an empirical value; L cont represents an intensity loss, L cont =||I HR -I HR-gt ||2, I HR represents an underwater intensity image enhanced by the de-scattering network, I HR-gt represents an intensity image in a clear water environment, and ||.||2 represents an L2 norm; L polar represents a polarization loss, which is composed of a degree of polarization loss and a polarization angle loss, i.e. polar =||DOLP-DOLP gt ||2+α||AOP-AOP gt ||2, DOLP and DOLP gt respectively represent a degree of polarization image enhanced by the de-scattering network and a degree of polarization image in clear water, and AOP and AOP gt respectively represent a polarization angle image enhanced by the de-scattering network and a polarization angle image in clear water, and α is a weight factor.

[0021] The second aspect of the present application discloses a neural network-based underwater degraded image polarization recovery system; the system comprises:

[0022] A first processing module is configured to collect polarization images in a real water environment and form a training data set after preprocessing; intensity images, degree of polarization images and polarization angle images;

[0023] A second processing module is configured to construct a de-scattering network based on a polarization image degradation model; wherein the de-scattering network comprises a shallow feature extraction module, a channel attention mechanism module, a dense residual module, a global feature fusion module and a reconstruction module connected in sequence;

[0024] A third processing module is configured to input samples in the training data set into the de-scattering network in sequence to obtain an enhanced result.

[0025] The fourth processing module is configured to calculate a joint loss function value of polarization and intensity based on the enhancement result and the sample label, and further calculate the gradient of the de-scattering network according to the joint loss function, and further obtain the optimized de-scattering network.

[0026] The fifth processing module is configured to input the target image into the optimized de-scattering network to obtain a recovered image.

[0027] The third aspect of the present application discloses an electronic device. The electronic device comprises a memory and a processor, the memory stores a computer program, and the processor executes the computer program to realize the steps of any one of the underwater degraded image polarization recovery methods based on the neural network in the first aspect of the present application.

[0028] The fourth aspect of the present application discloses a computer readable storage medium. The computer readable storage medium stores a computer program, and the computer program is executed by a processor to realize the steps of any one of the underwater degraded image polarization recovery methods based on the neural network in the first aspect of the present application.

[0029] In summary, the scheme proposed by the present application has the following technical effects:

[0030] (1) The present application introduces a degradation model based on a polarization image and a dense residual network. Since this technical means combines the in-depth understanding of the physical imaging model and the feature extraction capability of deep learning, it solves the problems of inaccurate image restoration and insufficient clarity that may occur when the traditional underwater polarization imaging method is used to process complex underwater environments, and achieves effective recovery of underwater targets and significant improvement of image quality in turbid water environments.

[0031] (2) The present application introduces a channel attention mechanism module in the network. Since this technical means can enable the network to identify and emphasize key areas in the image and assign different attention weights to different areas, it solves the problem that the traditional network may not be able to effectively distinguish between targets and backgrounds when processing polarization images, and achieves more accurate positioning of underwater targets and enhancement of image details, thereby improving the overall quality of underwater image restoration.

[0032] (3) The present application constructs a joint loss function of polarization and intensity. Since this technical means not only focuses on the image reconstruction quality at the pixel level, but also considers the reconstruction effect of polarization information at the feature level, it solves the problem of polarization information loss or inaccurate reconstruction that may occur in the process of underwater image restoration by existing methods, and achieves effective balance between intensity information and polarization information, thereby improving the polarization recovery quality of underwater degraded images. BRIEF DESCRIPTION OF DRAWINGS

[0033] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0034] Figure 1 This is a flowchart of a neural network-based polarization restoration method for water degradation images according to an embodiment of the present invention;

[0035] Figure 2 A schematic diagram of an experiment involving the capture of polarization datasets according to an embodiment of the present invention;

[0036] Figure 3 This is a schematic diagram of the descattering network structure according to an embodiment of the present invention;

[0037] Figure 4 This is a schematic diagram of the channel attention mechanism structure according to an embodiment of the present invention;

[0038] Figure 5 This is a schematic diagram of the dense residual module structure according to an embodiment of the present invention;

[0039] Figure 6 This is a structural diagram of an electronic device according to an embodiment of the present invention. Detailed Implementation

[0040] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, 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 the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0041] It is understood that the terms "first," "second," etc., used herein may be used to describe various elements, but these elements are not limited by these terms. These terms are only used to distinguish one element from another. For example, without departing from the scope of this application, a first image may be referred to as a second image, and similarly, a second image may be referred to as a first image. Both the first image and the second image are images, but they are not the same image.

[0042] To address the shortcomings of existing technologies in terms of target interpretation accuracy and image restoration quality, this invention employs a deep learning framework, leveraging a large-scale dataset for in-depth feature mining of the target. Through a multi-layered network structure, it achieves layer-by-layer abstraction from basic to complex features, utilizing these high-level features to analyze new input images, thereby realizing high-resolution underwater imaging.

[0043] According to an embodiment of the present invention, in a first aspect, a method for polarization restoration of water degradation images based on neural networks is provided; see also Figure 1 This method first acquires polarized images in a real underwater environment using a polarization camera to form a training dataset. Then, a deep learning descattering network is designed. This network, based on a polarization image degradation model, employs multiple components, including a channel attention mechanism module and a dense residual module, to enhance the capture of multi-scale features and improve network performance and training efficiency. Finally, a joint loss function is designed by combining polarization and intensity information to guide the network training process, ensuring that the reconstructed image achieves optimal performance in both polarization and intensity.

[0044] The method includes:

[0045] Step S1: Acquire polarization images from a real water environment and preprocess them to form a training dataset. The sample labels for each data sample in the training dataset include intensity image, degree of polarization image, and angle of polarization image. The intensity image, degree of polarization image, and angle of polarization image can be obtained from the polarization images at various angles through simple calculations.

[0046] Obtaining realistic underwater image datasets is challenging; therefore, most current underwater image datasets are generated through simulation. These datasets are typically based on classic underwater scattering models, artificially converting clear images into simulated turbid underwater images by setting relevant parameters. While this method is simple to operate, the resulting simulation datasets often have limited quality and significantly differ from actual underwater environments, affecting their practicality. To overcome the limitations of simulation methods, this invention obtains a corresponding real underwater image dataset comprising the following four steps, illustrated in the experimental diagram below. Figure 2 As shown.

[0047] The target object is placed securely in a specific position, and the polarization camera, LED light source, and polarizer are adjusted to the appropriate angles to ensure that their positions remain fixed.

[0048] Pour clean water into the tank and fix the target objects to be tested on the base in sequence for photography; pour clean water into the tank and fix the target objects on the base in a predetermined order for photography in a clear water environment.

[0049] Add 1 ml of skim milk to the water tank while stirring constantly to mix the water evenly. Then readjust and fix the position of the target object and take the picture again.

[0050] Continue adding milk and repeat step 3).

[0051] To ensure the network's generalization ability across different turbidity levels and target materials, this step involves acquiring multiple sets of underwater target polarization images and preprocessing them to construct a dataset. The dataset is randomly divided into training, validation, and test sets in a ratio of 8:1:1. Furthermore, to expand the dataset size, all images are cropped horizontally and vertically, with a side length of 64 pixels and a stride of 32 pixels. Finally, a large-scale dataset containing pairs of underwater polarization images is obtained.

[0052] Step S2: Construct a descattering network based on the polarization image degradation model; wherein, the descattering network includes a shallow feature extraction module, a channel attention mechanism module, a dense residual module, a global feature fusion module, and a reconstruction module connected in sequence;

[0053] This paper proposes a neural network-based polarization restoration method for water degradation images. The network structure consists of five parts: a shallow feature extraction module (SFE), a channel attention mechanism module (CAM), a residual module (RM), a global feature fusion module (GFF), a reconstruction module (RM), and a joint intensity and polarization loss function. The descattering network structure is shown in Figure 3. The network used in this invention has four input channels, consisting of polarization images at 0°, 45°, 90°, and 135°, with an image size of H×W×4, where H and W represent the height and width of the input image, respectively.

[0054] Shallow Feature Extraction Module: This step designs a shallow feature extractor to extract shallow feature information from the polarization image. Specifically, the shallow feature extraction module contains two convolutional layers with 3×3 kernels. Smaller kernels reduce the number of model parameters, thereby reducing model complexity and accelerating training. Simultaneously, smaller kernels can extract more detailed features, and the two 3×3 convolutional layers increase the network depth, thus improving the model's ability to learn complex features.

[0055] Channel Attention Mechanism Module: This module first utilizes Global Average Pooling (GAP) to reduce the feature map to a dimension of 1×1×C, where C represents the number of channels in the feature map. Then, it sequentially deploys two 1×1 convolutional operations to learn weights for the features of each channel. See details below. Figure 4The first 1×1 convolution adjusts the number of kernels to facilitate the integration of features from different channels and reduce the feature dimension, thereby alleviating the computational burden on the model. Next, the second 1×1 convolution restores the feature dimension, ensuring that the number of weights precisely corresponds to the number of feature channels. Finally, these coefficients are multiplied by their corresponding original feature channels to complete the feature weight adjustment.

[0056] This step adds dense residual modules after the channel attention mechanism module, and extracts features through a series of convolutional and activation layers. The output of each convolutional layer is concatenated with the input of subsequent layers, achieving dense feature connections. Simultaneously, each RDB contains residual connections, directly adding the input to the output, thus promoting feature reuse and information accumulation, and mitigating the gradient vanishing problem in deep network training. This design not only enhances the network's ability to capture multi-scale features but also improves network performance and training efficiency through the combination of residual learning and dense connections. Specific details are as follows... Figure 5 As shown.

[0057] Global Feature Fusion Module: This step utilizes the global feature fusion module to fuse feature information from different levels, enhancing the network's ability to reconstruct details in polarization images, especially when processing low-resolution polarization images. Located after the dense residual module, the global feature fusion module further fuses features from different residual blocks, enabling the network to capture more global information and thus improve image reconstruction quality. Specifically, the global feature fusion module first concatenates the outputs of all residual blocks along the channel dimension through a connection layer, merging feature maps from different levels so that the network can consider information at different scales simultaneously. The concatenated feature map then passes through two consecutive convolutional layers, both using the ReLU activation function, to further extract and fuse features. This process not only enhances the expressive power of the features but also helps the network learn more complex feature combinations.

[0058] Reconstruction Module: This step uses an upsampling module to reconstruct a high-resolution image of the target image size. Specifically, the feature map is first further processed through a convolutional layer (followed by a ReLU activation function). This helps extract richer feature representations before upsampling. Upsampling is then achieved by rearranging the pixels of the feature map while maintaining its spatial consistency. After upsampling, the feature map is processed through a final convolutional layer to generate the final high-resolution polarization image.

[0059] Step S3: Input the samples in the training dataset into the descattering network in sequence to obtain the enhancement results;

[0060] Step S4: Based on the enhancement results and sample labels, calculate the joint loss function value of polarization and intensity, and then perform backpropagation to calculate the gradient of the descattering network according to the joint loss function, thereby obtaining the optimized descattering network.

[0061] Since directly calculating the loss between the output and label does not accurately match the polarization image to the ground truth image, this step combines the intensity loss and polarization loss as the final loss function for network training. The intensity loss is expressed as: (The value is calculated by adding the difference between the label and output.)

[0062] L cont =||I HR -I HR-gt ||2

[0063] Among them, I HR This indicates an underwater image enhanced by the network described in this patent, I HR-gt This is the intensity image under clear water conditions, where ||.||2 is the L2 norm.

[0064] Polarization loss consists of polarization degree loss and polarization angle loss, and is calculated as follows:

[0065] L polar =||DOLP-DOLP gt ||2+α||AOP-AOP gt ||2

[0066] Among them, DOLP and DOLP gt These represent the polarization images enhanced by the proposed network and the polarization images in clear water, respectively. AOP and AOP are also shown. gt These represent the polarization angle images enhanced by the proposed network and the polarization angle images in clear water, respectively. α is a weighting factor used to balance the parameters. Therefore, the final loss of the network is expressed as follows:

[0067] L = L cont +λ×L polar

[0068] Here, λ is an empirical value that balances the intensity loss and polarization sensing loss and constrains the two losses to be of the same order of magnitude.

[0069] Step S5: Input the target image into the optimized descattering network to obtain the restored image.

[0070] A second aspect of this invention discloses a polarization restoration system for water degradation images based on a neural network; the system includes:

[0071] The first processing module is configured to acquire polarization images in a real water environment and preprocess them to form a training dataset; the sample labels of each data sample in the training dataset include intensity image, degree of polarization image and polarization angle image;

[0072] The second processing module is configured to construct a descattering network based on a polarization image degradation model; wherein the descattering network includes a shallow feature extraction module, a channel attention mechanism module, a dense residual module, a global feature fusion module, and a reconstruction module connected in sequence.

[0073] The third processing module is configured to sequentially input samples from the training dataset into the descattering network to obtain the enhancement results;

[0074] The fourth processing module is configured to calculate the joint loss function value of polarization and intensity based on the enhancement results and sample labels, and then perform backpropagation to calculate the gradient of the descattering network based on the joint loss function, thereby obtaining the optimized descattering network.

[0075] The fifth processing module is configured to input the target image into the optimized descattering network to obtain the restored image.

[0076] A third aspect of this invention discloses an electronic device. The electronic device includes a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, it implements the steps of the polarization restoration method for water degradation images based on a neural network, as described in any of the first aspects of this disclosure.

[0077] Figure 6 This is a structural diagram of an electronic device according to an embodiment of the present invention, such as... Figure 6 As shown, the electronic device includes a processor, memory, communication interface, display screen, and input device connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, Near Field Communication (NFC), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input device can be a touch layer covering the display screen, buttons, a trackball, or a touchpad mounted on the device's casing, or an external keyboard, touchpad, or mouse.

[0078] Those skilled in the art will understand that Figure 6The structure shown is merely a structural diagram of the part related to the technical solution of this disclosure and does not constitute a limitation on the electronic device to which the solution of this application is applied. The specific electronic device may include more or fewer components than shown in the figure, or combine certain components, or have different component arrangements.

[0079] A fourth aspect of this invention discloses a computer-readable storage medium. The computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps in a neural network-based polarization restoration method for water degradation images according to any one of the first aspects of this disclosure.

[0080] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein, and such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A neural network-based underwater degraded image polarization recovery method, characterized in that, The method comprises: Step S1, collecting polarization images and intensity images in a real water environment, and forming a training data set after pre-processing; the sample labels of each data sample in the training data set include intensity images, degree of polarization images and polarization angle images; Step S2, constructing a despeckling network based on a polarization image degradation model; wherein the despeckling network comprises shallow feature extraction modules, channel attention mechanism modules, dense residual modules, global feature fusion modules and reconstruction modules connected in sequence; Step S3, inputting the samples in the training data set into the despeckling network in sequence to obtain enhanced results; Step S4, calculating a joint loss function value of polarization and intensity based on the enhanced results and the sample labels, and then calculating the gradient of the despeckling network according to the joint loss function for back propagation, and then obtaining an optimized despeckling network; In step S4, the joint loss function L is: , wherein, denotes the empirical value; L cont denotes the intensity loss, , denotes the underwater intensity image enhanced by the despeckling network, denotes the intensity image in clear water, and ||.||2denotes the L2norm; L polar denotes the polarization loss, which consists of the polarization degree loss and the polarization angle loss, i.e. , DOLP and DOLP gt denote the polarization degree image enhanced by the despeckling network and the polarization degree image in clear water, respectively, AOP and AOP gt denote the polarization angle image enhanced by the despeckling network and the polarization angle image in clear water, respectively, and a is a weight factor; Step S5, inputting a target image into the optimized despeckling network to obtain a restored image.

2. The method of claim 1, wherein, In the step S1, all polarization images are cropped horizontally and vertically to obtain a training data set containing pairs of underwater polarization images.

3. The method of claim 2, wherein, In the step S2, the shallow feature extraction module contains two convolution layers with a kernel size of 3x3, which are used to extract shallow feature information of the polarization images to obtain feature maps.

4. The method of claim 3, wherein, In the step S2, the channel attention mechanism module uses a global average pooling layer to reduce the feature map to a dimension of 1x1xC, where C represents the number of channels of the feature map; then, weight learning is performed on each channel feature to adjust the weights of each feature in the feature map.

5. The method of claim 4, wherein, In the step S2, the dense residual module includes multiple residual modules, each of which extracts features through a series of convolution layers and activation layers, wherein the output of each convolution layer is spliced with the input of the subsequent layer to realize dense connection of features; at the same time, each residual module internally includes a residual connection, which directly adds the input to the output to facilitate feature reuse and information accumulation.

6. The method of claim 5, wherein, The global feature fusion module first splices the outputs of all residual blocks in the channel dimension through a connection layer to merge feature maps of different levels together, and then passes the spliced feature map through two consecutive convolution layers to further extract and fuse features; The reconstruction module uses up-sampling to reconstruct a high-resolution image of the target sample size.

7. A neural network-based underwater degraded image polarization recovery system, characterized in that, The system comprises: A first processing module configured to collect polarization images in a real water environment, and form a training data set after pre-processing; the sample labels of each data sample in the training data set include intensity images, degree of polarization images and polarization angle images; A second processing module configured to construct a despeckling network based on a polarization image degradation model; wherein the despeckling network comprises shallow feature extraction modules, channel attention mechanism modules, dense residual modules, global feature fusion modules and reconstruction modules connected in sequence; A third processing module configured to input the samples in the training data set into the despeckling network in sequence to obtain enhanced results; The fourth processing module is configured to calculate a joint loss function value of polarization and intensity based on the enhancement result and the sample label, to calculate the gradient of the de-scattering network according to the joint loss function, and to obtain the optimized de-scattering network; and the joint loss function is L is: where, denotes the empirical value; L cont denotes the intensity loss, , denotes the underwater intensity image enhanced by the despeckling network, denotes the intensity image in clear water, ||.||2denotes the L2norm; L polar denotes the polarization loss, which consists of the polarization degree loss and the polarization angle loss, i.e. , DOLP and DOLP gt denote the polarization degree image enhanced by the despeckling network and the polarization degree image in clear water, respectively, AOP and AOP gt denote the polarization angle image enhanced by the despeckling network and the polarization angle image in clear water, respectively, and a is a weight factor; A fifth processing module is configured to input the target image into the optimized de-scattering network to obtain a restored image.

8. An electronic device, comprising: The electronic device includes a memory and a processor, the memory stores a computer program, and the processor implements the steps of the neural network-based underwater degraded image polarization restoration method in any one of claims 1 to 6 when executing the computer program.

9. A computer-readable storage medium, characterized in that, The computer readable storage medium stores a computer program, and the computer program is executed by the processor to implement the steps of the neural network-based underwater degraded image polarization restoration method in any one of claims 1 to 6.

Images