Underwater polarization image restoration method based on physical and data dual drive

Abstract

The invention discloses an underwater polarization image restoration method based on physical and data dual drive, and belongs to the technical field of image processing, and the method comprises the following steps: S1, constructing an underwater polarization data set through an underwater imaging experiment platform; s2, establishing a target light model according to the Stokes vector of the underwater polarization image; and S3, a PCUNet network is constructed, and training is carried out. According to the underwater polarization image restoration method based on the physical and data dual drive, aiming at the problem of image noise amplification in the calculation process, the target light model is further simplified through the constraint model and parameter combination.

Description

Technical Field

[0001] The present invention relates to the field of image processing technology, and in particular to an underwater polarization image restoration method based on dual physics and data driving. Background Art

[0002] Underwater optical imaging technology is a core technology supporting marine resource development and ecological protection, providing fundamental support for deep-sea exploration, environmental monitoring, and equipment development. However, due to the light absorption characteristics of water and the scattering effect of suspended matter, underwater images commonly suffer from color casts and blurring. Traditional imaging methods struggle to accurately extract information, directly impacting underwater target identification and operational efficiency.

[0003] Underwater polarization imaging provides multi-dimensional polarization information. By leveraging the polarization property differences between background scattered light and target reflected light in water, it can effectively suppress backscatter interference and enhance the contrast of underwater targets, providing an effective way to improve the performance of underwater vision tasks. Existing methods improve underwater imaging by introducing the angle of polarization (AoP). However, due to the influence of water scattering, AoP images are subject to significant noise interference, resulting in limited image restoration quality. At the same time, the increased model complexity restricts practical applications. Therefore, to achieve reliable underwater polarization imaging, it is necessary to focus on solving key issues such as parameter estimation stability under scattering interference, improving computational efficiency, and effectively extracting polarization information. The effective solution of these issues will directly affect the practical application of this technology in scenarios such as underwater detection. Summary of the Invention

[0004] The purpose of the present invention is to provide an underwater polarization image restoration method based on dual physics and data driving to solve the problems mentioned in the background technology.

[0005] To achieve the above objectives, the present invention provides an underwater polarization image restoration method based on dual physics and data driving, comprising the following steps:

[0006] S1. Construct underwater polarization dataset through underwater imaging experimental platform;

[0007] S2, establishing a target light model based on the Stokes vector of the underwater polarization image;

[0008] S3. Build the PCUNet network and train it.

[0009] Preferably, the S1 includes:

[0010] S11. Build an underwater imaging experimental platform;

[0011] S12. Select targets with different polarization characteristics as experimental objects, prepare suspensions, and simulate water environments with different attenuation levels by adjusting the concentration of the suspensions;

[0012] S13. Use the underwater imaging experimental platform to obtain target polarization images at three angles of 0°, 60°, and 120° in water environments with different turbidity.

[0013] Preferably, the underwater imaging experimental platform includes an experimental container, an imaging module, a light source module, a water quality monitoring module and a target. The inner wall of the experimental container is covered with a black matte plate, a water body is arranged inside the experimental container, and the imaging module, light source module and target sample are arranged in the experimental container in sequence.

[0014] Preferably, the experimental container is a water tank; the imaging module uses a color polarization camera, which is packaged in a waterproof sealed cabin and installed in the experimental container; the light source module uses a polarization-modulated white light LED; and the water quality monitoring module uses a dual-beam ratio turbidity meter.

[0015] Preferably, the S2 includes:

[0016] S21. Establish a right triangle constraint model for polarization information based on the mathematical definitions of the degree of polarization DoP and the angle of polarization AoP.

[0017] S22. Introduce the AoP information into the underwater polarization imaging model through the right triangle constraint model, obtain the equation containing the target light S and find the solution for S;

[0018] S23. Perform model transformation using the angle relationship and merge unknown parameters to obtain the final target light model.

[0019] Preferably, the S21 includes:

[0020] The polarization state of the underwater polarization image is represented by the Stokes vectors I, Q, and U. The formulas for the Stokes vectors I, Q, and U are as follows:

[0021]

[0022] Among them, I0, I 60 and I 120 Represents polarization angle images of 0°, 60°, and 120° respectively;

[0023] The expressions for the polarization angle θ and the degree of polarization P obtained by the Stokes vector are as follows:

[0024]

[0025] Where AoP represents the angle of polarization and DoP represents the degree of polarization.

[0026] According to the definitions of θ and P, we can deduce:

[0027] (PI) 2 =Q 2+U 2 ;

[0028]

[0029] Based on the properties of the right triangle, the right triangle constraint model of polarization information is constructed as follows:

[0030]

[0031] Preferably, the S22 includes:

[0032] We obtain two equations containing the degree of polarization DoP and angle of polarization AoP of background light and target light:

[0033]

[0034] Among them, Q B and U B Stokes vector representing background light, P B and θ B Represent the polarization degree and polarization angle of the background light, Q S and U S The Stokes vector of the target light, P S and θ S Indicates the polarization degree and polarization angle of the target light;

[0035] According to the above formula, we can get:

[0036]

[0037] Preferably, the S23 includes:

[0038] The model of target light S is simplified by using the sum angle formula of trigonometric functions as follows:

[0039]

[0040] Among them, P S ,θ B and θ S are all unknown parameters, θ I represents the polarization angle of the underwater image I;

[0041] Combine all unknown parameters in the target light model into one parameter S dif , so that the model is further simplified to:

[0042] S=PIS dif ;

[0043] S dif The expression is as follows:

[0044]

[0045] Preferably, the S3 includes:

[0046] S31. Construct a PCUNet network based on the U-shaped ConvNeXt architecture. The PCUNet network includes an encoder, an intermediate layer, and a decoder.

[0047] S32. During the model optimization process, a composite loss function containing multiple loss functions is used to update parameters; the loss function combination includes L1 loss, SSIM loss, and perceptual loss;

[0048] S33. Use the underwater polarization dataset to train the network.

[0049] Preferably, in S32, the formula of the L1 loss function is expressed as follows:

[0050]

[0051] Among them, L1 represents L1 loss, N is the number of pixels, is the result of network output calculation, S is the corresponding label image in the clear water environment;

[0052] The formula of the SSIM loss function is as follows:

[0053]

[0054] in, express and SSIM value between S, L SSIM represents the SSIM loss, and μ S for and the mean of S, and σ S for and the standard deviation of S, for and the covariance of S, C1 and C2 are small constants introduced for stability;

[0055] The formula of the perceptual loss function is as follows:

[0056]

[0057] Among them, L Perceptual represents the perceptual loss, φ() represents the output from the starting layer to the conv2_1 layer in VGG16;

[0058] The L1 loss, SSIM loss and perceptual loss are weighted together to obtain a composite loss function expressed as follows:

[0059] Ltotal =L1+λ1L SSIM +λ2L Perceptual ;

[0060] Among them, L total Represents the total loss of the composite type, λ1 and λ2 represent the weight ratio of SSIM loss and perceptual loss in the total loss respectively, λ1=L1 / L SSIM ,λ2=L1 / L Perceptual .

[0061] Therefore, the present invention adopts the above-mentioned underwater polarization image restoration method based on dual physics and data driving, which has the following beneficial effects:

[0062] (1) To address the problem of polarization information utilization, a target light model based on polarization right triangle constraints was proposed. Through theoretical derivation, the AoP information was effectively integrated into the underwater polarization imaging model, while simplifying the model parameters to reduce the computational complexity. To improve the image restoration quality, a PCUNet network structure was designed, which improved the utilization of underwater polarization information through feature extraction and reconstruction mechanisms. Combined with the guidance of the physical model, it achieved the task of clarifying underwater polarization images.

[0063] (2) The PCUNet network not only achieves effective extraction of polarization features, but also further enhances image quality through feature recombination.

[0064] (3) In order to make full use of polarization information, the present invention establishes a constraint model of a right triangle of polarization information based on the mathematical definitions of DoP and AoP, combines the Stokes vector with DoP and AoP through geometric properties, establishes a new underwater polarization imaging model, and obtains the solution of the target light; in order to address the problem of image noise amplification during the calculation process, the target light model is further simplified by merging the constraint model and parameters.

[0065] (4) Polarization imaging technology can capture multi-dimensional polarization information features during underwater imaging, providing effective support for target feature reconstruction. The present invention constructs a PCUNet neural network architecture, through which deep extraction and reconstruction optimization of image features are achieved. To improve the physical interpretability of the algorithm model, the established target light model and polarization feature parameters are integrated into the PCUNet framework to form a polarization image restoration method driven by both physical data and effectively improve the imaging quality of targets in underwater turbid environments.

[0066] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] Figure 1 This is a flow chart of an underwater polarization image restoration method based on dual physics and data driving according to an embodiment of the present invention;

[0068] Figure 2 A schematic diagram of a polarization information triangle constraint according to an embodiment of the present invention;

[0069] Figure 3 This is a PCUNet network structure diagram of an embodiment of the present invention;

[0070] Figure 4 4 is a contrast diagram of the experimental results of an embodiment of the present invention. DETAILED DESCRIPTION

[0071] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings 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. The components of the embodiments of the present invention generally described and shown in the drawings herein can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the claimed invention, but merely represents selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0072] It should be noted that similar reference numerals and letters denote similar items in the following drawings, and therefore, once an item is defined in one drawing, it does not need to be further defined or explained in subsequent drawings.

[0073] Example

[0074] like Figure 1 As shown, the present invention provides an underwater polarization image restoration method based on dual physics and data driving, comprising the following steps:

[0075] S1. Build an underwater polarization dataset.

[0076] Specifically, step S1 includes:

[0077] S11. Build an underwater imaging experimental platform. The underwater imaging experimental platform includes an experimental container, an imaging module, a light source module, a water quality monitoring module, and a target. The inner wall of the experimental container is covered with a black matte plate to eliminate external stray light interference. Water is set inside the experimental container, and the imaging module, light source module, and target are arranged in the experimental container in sequence. In this embodiment, the experimental container adopts a water tank with a size of 100cm×50cm×50cm. The imaging module uses a color polarization camera with a resolution of 2592×1944. The device is encapsulated in a waterproof sealed cabin and installed in the experimental container. It can simultaneously collect jpg format images with three polarization angles of 0°, 60°, and 120°. The light source module uses a polarization-modulated white light LED with a color temperature of 6500-7500K. The water quality monitoring module is equipped with a dual-beam ratio turbidity meter that complies with the HJ1075-2019 standard to record the turbidity data of the water body under the environment of adding different amounts of turbid liquid.

[0078] S12. We selected targets with different polarization properties as experimental subjects and mixed milk and sodium copper chlorophyllin to prepare suspensions. By adjusting the concentration of the mixture, we simulated water environments with varying attenuation levels. Polarization properties include polarization, color, and texture.

[0079] S13. Use the underwater polarization imaging platform to obtain target polarization images at three angles of 0°, 60°, and 120° in water environments with different turbidity.

[0080] S2. Establish a target light model.

[0081] Specifically, step S2 includes:

[0082] S21. Based on the mathematical definitions of the degree of polarization DoP and the angle of polarization AoP, a right-angle triangle constraint model of polarization information is established.

[0083] Specifically, step S21 includes:

[0084] The polarization state of an underwater polarization image can be represented by the Stokes vectors I, Q, and U:

[0085]

[0086] Among them, I0, I 60 and I 120 These represent polarization angle images of 0°, 60°, and 120°, respectively.

[0087] The polarization angle θ and the degree of polarization P can be calculated from the Stokes vector using the following formula:

[0088]

[0089] According to the definitions of θ and P, we can deduce:

[0090] (PI) 2 =Q 2 +U 2 ;

[0091]

[0092] From the definition of a right triangle, we know that the Stokes vectors, θ and P satisfy the constraints of a right triangle with sides Q and U, hypotenuse PI, and an angle of 2θ between Q and PI, as follows: Figure 2 shown.

[0093] According to the properties of the right triangle, the Stokes vector, θ, and P satisfy the following relationship, that is, the right triangle constraint model of polarization information is as follows:

[0094]

[0095] S22. Using a right-angled triangle constraint model, the AoP information is introduced into the underwater polarization imaging model, and an equation containing the target light S is obtained and the solution of S is found.

[0096] In this embodiment, step S22 includes:

[0097] We obtain two equations containing the degree of polarization DoP and angle of polarization AoP of background light and target light:

[0098]

[0099] Among them, Q B and U B Stokes vector representing background light, P B and θ B Represent the polarization degree and polarization angle of the background light, Q S and U S The Stokes vector of the target light, P S and θ S Indicates the polarization degree and polarization angle of the target light;

[0100] According to the above formula, we can get:

[0101]

[0102] S23. Perform model transformation using the angle relationship and merge unknown parameters to obtain the final target light model.

[0103] In this embodiment, step S23 includes:

[0104] In order to reduce excessive calculation errors and simplify the model, the final target light S model can be obtained by using the sum angle formula of trigonometric functions, in which the unknown parameters include P S ,θ B and θ S .

[0105]

[0106] Among them, θ I represents the polarization angle of the underwater image I;

[0107] The model that introduces polarization angle information still has multiple unknown parameters even after computational simplification. All unknown parameters in the model are combined into one parameter S dif , so that the model is further simplified to:

[0108] S=PIS dif ;

[0109] S dif The expression is as follows:

[0110]

[0111] It can be seen that S dif The numerator contains the light intensity and background light polarization angle information, and the denominator contains the intermediate physical quantity of the target light polarization information and the background light polarization angle.

[0112] S3. Build the PCUNet network based on the U-shaped ConvNeXt architecture and train it.

[0113] In this embodiment, step S3 includes:

[0114] S31. Given the powerful capabilities of neural networks in feature extraction and feature recombination, construct Figure 3 The U-shaped PCUNet network shown, based on the ConvNeXt v2 Block, consists of an encoder, intermediate layers, and a decoder. The encoder input receives a 9-channel feature tensor as initial data, while the decoder inputs a 3-channel feature tensor corresponding to the polarization information PI, ultimately outputting a 3-channel underwater restored image.

[0115] The overall architecture of the PCUNet network is based on the U-Net and ConvUNeXt bottleneck structures. The encoder consists of multiple downsampling blocks, with the basic convolutional module being the ConvNeXt v2Block. The first layer has a Down(4x)-CB×N structure, and subsequent downsampling blocks use a LN-Down(2x)-CB×N structure to extract multi-scale features from the image. N is the number of CB blocks, set to [3, 3, 9, 3] in the encoder stage. The connection between the encoder and decoder uses an intermediate layer consisting of a Conv1(1×1)-CB×N layer, with a single CB block. The decoder consists of multiple upsampling blocks. The first upsampling block uses a Conv(1×1)-LN-Up(2x)-Conv(1×1)-CB×N structure, and subsequent upsampling blocks use a LN-Up(2 / 4x)-Conv(1×1)-CB×N structure. The number of CB blocks in the decoder stage is always one. The output layer at the end of the network, corresponding to the first encoder layer, is upsampled by a factor of 4. While receiving the output features of the previous layer, this layer integrates the initial input data and the parameters of the polarization imaging physical model through residual connections, enhancing feature reconstruction capabilities. To improve network performance, a CGA module, which incorporates spatial attention, channel attention, and pixel attention mechanisms, is embedded in the skip connections. Skip connections effectively improve network stability and consistency, while the CGA module further enhances the network's ability to capture diverse features, thereby improving restoration results.

[0116] The network's first layer input is a 9-channel tensor consisting of color polarization images at three angles: 0°, 60°, and 120°. At the output layer, polarization information (PI) is integrated into the network's processing using skip connections, ultimately guiding the network to output a 3-channel color restored image. Table 1 shows the changes in the input and output of the tensors for each module within the PCUNet network on the main branch. C × H × W represents the number of channels, height, and width of the tensor.

[0117] Table 1 Network module input and output shape change table

[0118]

[0119] S32. To improve the network's predictive capabilities, a composite loss function consisting of multiple loss functions is used to update parameters during model optimization, thereby guiding network learning. The loss function combination includes L1 loss, SSIM loss, and perceptual loss.

[0120] Compared with MSE loss, L1 loss can avoid falling into local optimality, and at the same time, it has less penalty for large error values ​​during gradient descent, thereby improving the stability and effect of optimization. The formula of L1 loss function is as follows:

[0121]

[0122] Among them, L1 represents L1 loss, also known as mean absolute error, N is the number of pixels, is the result of network output calculation, S is the corresponding label image in the clear water environment, and is also the target light image.

[0123] SSIM loss can effectively capture the structural information of the image, focusing on the preservation of image brightness, contrast, and structure. Introducing SSIM loss helps improve the visual quality of the restored image and ensure the perceptual consistency between the restored result and the real image. The formula of the SSIM loss function is expressed as follows:

[0124]

[0125] in, express The SSIM value between L and S, SSIM is a structural similarity index, which is a commonly used and classic index to measure the similarity between two images. SSIM represents the SSIM loss, and μ S for and the mean of S, and σ S for and the standard deviation of S, for and the covariance of S, C1 and C2 are small constants introduced for stability.

[0126] To further improve the perceptual quality of images, we introduced a perceptual loss. This loss function uses the output of the conv2_1 layer in the VGG16 network as a shallow feature extractor and calculates the L1 loss between the restored image and the reference image in the shallow feature space. This approach allows the network to focus more on preserving high-level semantic information and detailed features of the image. The perceptual loss function is expressed as follows:

[0127]

[0128] Among them, L Perceptual represents the perceptual loss, and φ() represents the output from the starting layer to the conv2_1 layer in VGG16.

[0129] Ultimately, the network training uses a weighted composite loss that combines the advantages of L1 loss, SSIM loss, and perceptual loss. The weight parameters adjust the contribution of each loss to the optimization objective, thus achieving a good balance between restoration accuracy and perceptual quality. The resulting composite loss function is expressed as follows:

[0130] L total=L1+λ1L SSIM +λ2L Perceptual ;

[0131] Among them, L total Represents the total composite loss, λ1 and λ2 represent the weight ratio of SSIM loss and perceptual loss in the total loss, λ1=L1 / L SSIM ,λ2=L1 / L Perceptual ,Due to the different scales among L1 loss, SSIM loss and perceptual loss, this dynamic adjustment method is adopted to align the scales of SSIM loss and perceptual loss to the same level as L1 loss to ensure the stability during training and the effectiveness of the loss function.

[0132] S33. Use the underwater polarization dataset to train the network.

[0133] The network training set consists of 160 polarization image samples. Each image was preprocessed and cropped to a fixed size of 512×512, with a cropping step of 512 pixels. The network model uses a multi-channel input structure, with a 9-channel tensor consisting of three polarization angle images as input and a 3-channel tensor as output after processing. Training was performed on an Nvidia RTXA6000 graphics card using the PyTorch framework for 200 epochs. The optimizer used Adamw, with an initial learning rate of 0.0005 and a cosine annealing learning rate schedule for learning rate adjustment. The batch size was set to 16.

[0134] Reference Figure 4 In order to measure the effectiveness of the present invention, the underwater polarization dataset is used to evaluate the performance of PCUNet with the image enhancement method CLAHE, the imaging model-based method UDCP, two underwater polarization imaging methods APD and UDPLC, and two deep learning methods Semi-UIR and CWR in terms of SSIM, PSNR, PCQI, LPIPS, UIQM and UCIQE indicators. The results are shown in Table 2:

[0135] Table 2 Experimental results evaluation indicators

[0136]

[0137] Comparative experimental results with other methods show that the present invention not only outperforms other methods in polarization information processing performance, but also demonstrates competitive advantages in edge detail preservation and contrast enhancement, providing a new methodological path and potential technical development direction for underwater polarization imaging technology.

[0138] Therefore, the present invention adopts the above-mentioned underwater polarization image restoration method based on dual physics and data driving, establishes a polarization information right-angle triangle constraint model through the physical definition of polarization information, introduces AoP into the underwater polarization imaging model, and combines DoP and AoP to realize the physical modeling of target light.

[0139] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit the same. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that they can still modify or replace the technical solutions of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for underwater polarization image restoration based on dual physics and data driving, characterized in that: The following steps are involved: S1. Construct underwater polarization dataset through underwater imaging experimental platform; S2, establishing a target light model based on the Stokes vector of the underwater polarization image; S3. Build the PCUNet network and train it.

2. The underwater polarization image restoration method based on dual physics and data driving according to claim 1 is characterized in that: Said S1 comprises: S11. Build an underwater imaging experimental platform; S12. Select targets with different polarization characteristics as experimental objects, prepare suspensions, and simulate water environments with different attenuation levels by adjusting the concentration of the suspensions; S13. Use the underwater imaging experimental platform to obtain target polarization images at three angles of 0°, 60°, and 120° in water environments with different turbidity.

3. The underwater polarization image restoration method based on dual physics and data driving according to claim 2, characterized in that: The underwater imaging experimental platform includes an experimental container, an imaging module, a light source module, a water quality monitoring module and a target. The inner wall of the experimental container is covered with a black matte plate. There is water inside the experimental container. The imaging module, light source module and target sample are arranged in the experimental container in sequence.

4. The underwater polarization image restoration method based on dual physics and data drive according to claim 3, characterized in that: The experimental container is a water tank; the imaging module uses a color polarization camera, which is encapsulated in a waterproof sealed cabin and installed in the experimental container; the light source module uses a polarization-modulated white light LED; and the water quality monitoring module uses a dual-beam ratio turbidity meter.

5. The underwater polarization image restoration method based on dual physics and data driving according to claim 2 is characterized in that: The S2 includes: S21. Establish a right triangle constraint model for polarization information based on the mathematical definitions of the degree of polarization DoP and the angle of polarization AoP. S22. Introduce the AoP information into the underwater polarization imaging model through the right triangle constraint model, obtain the equation containing the target light S and find the solution for S; S23. Perform model transformation using the angle relationship and merge unknown parameters to obtain the final target light model.

6. The underwater polarization image restoration method based on dual physics and data driving according to claim 5 is characterized in that: The S21 includes: The polarization state of the underwater polarization image is represented by the Stokes vectors I, Q, and U. The formulas for the Stokes vectors I, Q, and U are as follows: Among them, I0, I 60 and I 120 Represents polarization angle images of 0°, 60°, and 120° respectively; The expressions for the polarization angle θ and the degree of polarization P obtained by the Stokes vector are as follows: Where AoP represents the angle of polarization and DoP represents the degree of polarization. According to the definitions of θ and P, we can deduce: (PI) 2 =Q 2 +U 2 ; Based on the properties of the right triangle, the right triangle constraint model of polarization information is constructed as follows:

7. The underwater polarization image restoration method based on dual physics and data driving according to claim 6, characterized in that: The S22 includes: We obtain two equations containing the degree of polarization DoP and angle of polarization AoP of background light and target light: Among them, Q B and U B Stokes vector representing background light, P B and θ B Represent the polarization degree and polarization angle of the background light, Q S and u S The Stokes vector of the target light, P S and θ S Indicates the polarization degree and polarization angle of the target light; According to the above formula, we can get:

8. The underwater polarization image restoration method based on dual physics and data driving according to claim 7, characterized in that: The S23 includes: The model of target light S is simplified by using the sum angle formula of trigonometric functions as follows: Among them, P S ,θ B and θ S are all unknown parameters, θ I represents the polarization angle of the underwater image I; Combine all unknown parameters in the target light model into one parameter S dif , so that the model is further simplified to: S=PIS dif ; S dif The expression is as follows:

9. The underwater polarization image restoration method based on dual physics and data driving according to claim 8, characterized in that: The S3 includes: S31. Construct a PCUNet network based on the U-shaped ConvNeXt architecture. The PCUNet network includes an encoder, an intermediate layer, and a decoder. S32. During the model optimization process, a composite loss function containing multiple loss functions is used to update parameters; the loss function combination includes L1 loss, SSIM loss, and perceptual loss; S33. Use the underwater polarization dataset to train the network.

10. The underwater polarization image restoration method based on dual physics and data driving according to claim 9, characterized in that: In S32, the formula of the L1 loss function is expressed as follows: Where L1 represents L1 loss, N is the number of pixels, is the result of network output calculation, S is the corresponding label image in the clear water environment; The formula of the SSIM loss function is as follows: in, express and SSIM value between S, L SSIM represents the SSIM loss, and μ S for and the mean of S, and σ S for and the standard deviation of S, for and the covariance of S, C1 and C2 are small constants introduced for stability; The formula of the perceptual loss function is as follows: Among them, L Perceptual represents the perceptual loss, φ() represents the output from the starting layer to the conv2_1 layer in VGG16; The L1 loss, SSIM loss and perceptual loss are weighted together to obtain a composite loss function expressed as follows: L total =L1+λ1L SSIM +λ2L Perceptual ; Among them, L total Represents the total composite loss, λ1 and λ2 represent the weight ratio of SSIM loss and perceptual loss in the total loss, λ1=L1 / L SSIM ,λ2=L1 / L Perceptual .

Images