A method and system for enhancing severe weather images based on prior physical knowledge

Abstract

This invention discloses a method and system for enhancing severe weather images based on physical prior knowledge, belonging to the field of image enhancement technology. The method involves inputting the image to be enhanced into a trained image enhancement model and outputting a semantically segmented image. The training process of the image enhancement model is as follows: acquiring images under different driving conditions in clear and severe weather to construct a training dataset; using clear weather images as reference images, learning the distribution differences between clear weather and severe weather images based on a sharpening module, separating the weather component from the severe weather image, and generating a sharpened image; introducing physical prior knowledge into the image enhancement module to perform image enhancement processing on the sharpened image to obtain the enhanced image; and performing semantic segmentation on the enhanced image to obtain a semantically segmented image with detailed features. This method and system for enhancing severe weather images significantly improves the clarity and quality of rain and snow images.

Description

Technical Field

[0001] This invention relates to the field of image enhancement technology, and in particular to a method and system for enhancing severe weather images based on prior physical knowledge. Background Technology

[0002] In the field of autonomous driving, the reliability of vision sensors under various weather conditions is crucial. However, rain and snow can significantly degrade image quality. Images captured by vision sensors are often affected by raindrops, snowflakes, and fog, causing shifts in image sharpness, brightness, contrast, and tone. These weather factors not only affect the overall visual quality of the images but also impact the accuracy of subsequent computer vision tasks, especially those directly related to safety, such as semantic segmentation, object detection, and obstacle recognition.

[0003] Traditional image enhancement techniques for processing rain and snow images typically employ simple filtering, denoising, or brightness adjustment. However, these methods often fail to fundamentally remove the features introduced by rain and snow and struggle to achieve adaptive enhancement under varying weather conditions. Current research is increasingly exploring the use of deep learning techniques, combined with prior physical knowledge, to enhance images in rain and snow conditions. Specifically, by designing specialized loss functions, models can perform self-supervised learning by referencing clear weather images without paired datasets, generating clearer, interference-free images. Nevertheless, these methods still have room for improvement in terms of efficiency and accuracy. Summary of the Invention

[0004] Based on the technical problems existing in the background technology, this invention proposes a method and system for enhancing severe weather images based on physical prior knowledge, which significantly improves the clarity and quality of rain and snow images, enhances the visual perception ability of autonomous driving systems under rain and snow conditions, and ensures driving safety.

[0005] The present invention proposes a method and system for enhancing severe weather images based on prior physical knowledge, which inputs the image to be enhanced into a trained image enhancement model and outputs a semantic segmentation image;

[0006] The training process for the image enhancement model is as follows:

[0007] Acquire images under different driving conditions in sunny and inclement weather, and construct a training dataset;

[0008] Using clear weather images as reference images, the distribution differences between clear weather images and severe weather images are learned based on the sharpening module. Weather components are separated from severe weather images to generate sharpened images.

[0009] In the image enhancement module, physical priors are introduced to perform image enhancement processing on the sharpened image to obtain the enhanced image;

[0010] Semantic segmentation is performed on the enhanced image to obtain a semantically segmented image with detailed features;

[0011] Construct a total loss function and adjust the parameters of the image enhancement model.

[0012] Furthermore, the sharpening module includes six sequentially connected convolutional layers;

[0013] The first three convolutional layers are used to process severe weather images and downsample layer by layer to reduce the size of severe weather images and extract low-level features of severe weather images to obtain the first image;

[0014] The last three convolutional layers gradually upsample through deconvolution operations to restore the first image to its original resolution and ensure that low-level features in the first image are preserved, thereby generating a sharpened image;

[0015] Batch regularization is applied after each convolutional layer to stabilize the feature distribution, and the ReLU activation function is used to enhance the non-linear expressive power of the sharpening module.

[0016] Furthermore, the image enhancement module includes six sequentially connected convolutional layers;

[0017] Each convolutional layer progressively doubles the number of channels and reduces the spatial size to half that of the previous convolutional layer to extract multi-scale feature information from the sharpened image step by step. The output of the last convolutional layer of the image enhancement module generates the parameters required for image enhancement; image enhancement is then performed on the extracted sharpened image based on the required parameters.

[0018] Furthermore, a physical prior is introduced into the image enhancement module to perform image enhancement processing on the sharpened image, specifically as follows:

[0019] A model was established based on atmospheric scattering theory to describe the scattering process of light in the atmosphere under severe weather conditions, and to obtain the impact of haze and suspended water molecules on the sharpened image.

[0020] Gamma correction technology is used to adjust the brightness and contrast of the sharpened image to compensate for the reduction in brightness and contrast shift caused by severe weather, resulting in an enhanced image.

[0021] Furthermore, the atmospheric scattering theory model is specifically established as follows:

[0022] I(x)=J(x)t(x)+A(1-t(x));

[0023] Where I(x) is the pixel value of the inclement weather image at position x, J(x) is the pixel value of the real image of the scene at position x, A represents the background brightness, and t(x) is the transmittance, defined as the light transmission rate from the scene to the camera.

[0024] Furthermore, gamma correction technology is employed to adjust the brightness and contrast of the sharpened image, specifically:

[0025]

[0026] Among them, I in It is the pixel value of the input severe weather image, I out γ is the pixel value of the enhanced image, and γ is the gamma value. When γ < 1, it is used to enhance dark details, and when γ > 1, it is used to compress bright details.

[0027] Furthermore, the total loss function includes sharpening loss, exposure loss, and semantic segmentation loss.

[0028] Furthermore, the sharpening loss includes prior loss and structural similarity index loss;

[0029] The prior loss L ref for:

[0030]

[0031] The structural similarity index loss L ssim for:

[0032]

[0033] The exposure loss L exp for:

[0034]

[0035] The semantic segmentation loss L seg for:

[0036]

[0037] Among them, C(I) ref The image shown is the sharpened image output by the sharpening module. The feature map is all zeros, ∥·∥2 is the L2 norm, SSIM is the structural index, and C in For the sharpened image output by the sharpening module, I in For severe weather images, I ref Here, is the reference image, s is a constant used to measure the similarity between the reference image and the severe weather image, ||·||1 is the L1 norm, and M is the number of local regions divided in the enhanced image. It is an average pooling layer used to calculate the exposure level of a local area, E in For the enhanced image, N is the total number of pixels in the severe weather image, and C is the total number of categories. It is the one-hot encoding of the true value of the k-th category, where GT is the true value. ω is the probability predicted by the image enhancement model that pixel n belongs to category k, where n∈[1,N], k∈[1,C], and ω k This represents the weighting parameter.

[0038] A severe weather image enhancement system based on prior physical knowledge inputs rain / snow images into a trained image enhancement model and outputs semantically segmented images.

[0039] The training process of an image enhancement model includes a training set construction module, a sharpening module, an image enhancement module, a semantic segmentation module, and a loss construction module.

[0040] The training set construction module is used to acquire images under different driving conditions in sunny and inclement weather to build a training dataset;

[0041] The sharpening module uses a clear sky image as a reference image to learn the distribution differences between clear sky images and severe weather images, separates the weather components from the severe weather image, and generates a sharpened image.

[0042] The image enhancement module is used to enhance a sharpened image by introducing physical priors, resulting in an enhanced image.

[0043] The semantic segmentation module is used to perform semantic segmentation on the enhanced image to obtain a semantically segmented image with detailed features;

[0044] The loss construction module is used to construct the total loss function and adjust the parameters of the image enhancement model.

[0045] Furthermore, the sharpening module includes six sequentially connected convolutional layers;

[0046] The first three convolutional layers are used to process severe weather images and downsample layer by layer to reduce the size of severe weather images and extract low-level features of severe weather images to obtain the first image;

[0047] The last three convolutional layers gradually upsample through deconvolution operations to restore the first image to its original resolution and ensure that low-level features in the first image are preserved, thereby generating a sharpened image;

[0048] Batch regularization is applied after each convolutional layer to stabilize the feature distribution, and the ReLU activation function is used to enhance the non-linear expressive power of the sharpening module.

[0049] The advantages of the severe weather image enhancement method and system based on prior physical knowledge provided by this invention are as follows: By combining prior knowledge and a self-supervised learning strategy, and designing multiple loss functions such as sharpening, enhancement, and semantic segmentation, the system achieves efficient removal of interference, enhancement of sharpness, and improvement of semantic segmentation in rain and snow weather images. This system can separate weather components from images, generate high-quality images that meet practical application needs under different weather conditions, and optimize visual perception through adaptive training, ultimately improving the safety and reliability of autonomous driving systems in severe weather conditions such as rain and snow. Attached Figure Description

[0050] Figure 1 This is a schematic diagram of the process of the present invention. Detailed Implementation

[0051] The technical solution of the present invention will now be described in detail through specific embodiments. Many specific details are set forth in the following description to provide a thorough understanding of the invention. However, the present invention can be implemented in many other ways different from those described herein, and those skilled in the art can make similar modifications without departing from the spirit of the invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0052] like Figure 1 As shown, the present invention proposes a method and system for enhancing severe weather images based on prior physical knowledge. The image to be enhanced is input into a trained image enhancement model, and the semantic segmentation image is output.

[0053] This embodiment improves the visual clarity and semantic segmentation performance of images under complex weather conditions. The method relies on prior physical knowledge and achieves component separation and sharpening through self-supervised learning. This embodiment uses rain and snow as a type of severe weather for illustration. The training process of the image enhancement model is as follows:

[0054] Step 1: Acquire images under different driving conditions in sunny and inclement weather, and construct a training dataset;

[0055] Data Acquisition: Visual sensors were used to acquire driving environment image data under clear and rainy / snowy weather conditions. Clear Weather Image I ref This image serves as a reference, free from rain and snow interference, and is used for self-supervised learning in subsequent image enhancement models.

[0056] Step 2: Using a clear sky image as a reference image, the distribution differences between the clear sky image and the severe weather image are learned based on the sharpening module. The weather component is separated from the severe weather image to generate a sharpened image.

[0057] A sharpened image is generated using a reference image. The sharpening module learns the distribution differences between clear weather images and inclement weather images, thereby removing interference from the input rain and snow images under self-supervision. The distribution differences specifically include (a1) to (a3):

[0058] (a1) Learn the distribution differences of rain / snow weather components through the clarity module.

[0059] Specifically, by comparing images of sunny and rainy / snowy weather, the network learns the distribution characteristics of weather components such as raindrops, watermarks, and snowflakes in the images. Using these distribution differences, the network identifies and separates the rain / snow components in the images, thereby generating clear images free from weather interference.

[0060] (a2) Learn the distribution differences in brightness and hue of rain / snow weather components through the sharpening module;

[0061] Rain / snow weather can cause an overall decrease in image brightness and a shift in tone. By learning from the differences in distribution, the system can identify these characteristic shifts in brightness and tone, thereby restoring the normal brightness and tone of the image after removing the weather component, making the generated image closer to the visual effect under sunny conditions.

[0062] (a3) Learn the distribution differences of local contrast of rain / snow weather components through the sharpening module;

[0063] Rain / snow components affect the contrast of local image regions, especially in areas covered by raindrops or snowflakes where contrast is lower. The model learns the differences in the distribution of local contrast to restore local contrast in a sharpened image, thereby improving the image's visual clarity.

[0064] In this embodiment, the sharpening module consists of six sequentially connected convolutional layers. The first three convolutional layers process the severe weather image and downsample it layer by layer to reduce the image size and extract low-level features, resulting in a first image. The last three convolutional layers perform deconvolution operations to upsample the image step by step, restoring it to its original resolution while ensuring that low-level features in the first image are preserved, thereby generating a sharpened image.

[0065] In these six convolutional layers, a ResNetBlock structure is used to enhance the expressive power of the sharpening module. The residual connection between two convolutional layers enables the effective propagation and enhancement of feature information, ensuring the preservation of image details and efficient processing.

[0066] In the sharpening module, batch regularization is applied after each convolutional layer to stabilize the feature distribution, and the ReLU activation function is used to enhance the nonlinear expressive power of the network, thereby improving the system's processing efficiency and stability under complex weather conditions.

[0067] The declarative module uses 7×7 convolutional kernels in the first and last convolutional layers to expand the receptive field, thereby enhancing the expressiveness of initial feature extraction and final image reconstruction. The remaining layers use 3×3 convolutional kernels to ensure computational efficiency and accuracy of feature extraction. The final output of the declarative module is a three-channel RGB image, i.e., a clarified image after removing rain / snow interference, which is used for subsequent semantic segmentation or other visual tasks.

[0068] Step 3: Introduce physical priors into the image enhancement module to perform image enhancement processing on the sharpened image to obtain the enhanced image;

[0069] Based on physical priors, this paper proposes an image enhancement model to make rain and snow image enhancement systems more accurate and produce higher-quality images when processing images in severe weather. Specifically, it establishes a model based on atmospheric scattering theory to describe the scattering process of light in the atmosphere under rain and snow conditions. This model is based on the scattering and absorption effects of light propagating in the atmosphere and is mainly used to estimate image degradation caused by haze, rain, and snow, effectively reducing blurring caused by light scattering, thereby improving image contrast and clarity.

[0070] The atmospheric scattering theory model is specifically established as follows:

[0071] I(x)=J(x)t(x)+A(1-y(x));

[0072] Where I(x) is the pixel value of the inclement weather image at position x, J(x) is the pixel value of the real image of the scene at position x, A represents the background brightness, and t(x) is the transmittance, defined as the light transmission rate from the scene to the camera.

[0073] By employing gamma correction technology, the brightness and contrast of the image are adjusted to compensate for the reduction in brightness and contrast shift caused by rain and snow.

[0074]

[0075] Among them, I in I is the pixel value of the input severe weather image (typically in the range of 0 to 1). out γ is the pixel value of the enhanced image, and γ is the gamma value. When γ < 1, it is used to enhance dark details, and when γ > 1, it is used to compress bright details.

[0076] By adjusting the gamma value of the sharpened image, the visual effect of the sharpened image is improved, making the image details clearer, while enhancing the expressiveness of dark areas and ensuring visual consistency under different lighting conditions.

[0077] Specifically, a parameter prediction module is implemented using a convolutional neural network (CNN). This module is designed to be resolution-independent, differentiable, and easy to understand, in order to ensure the universality and stability of the image enhancement model under different input image resolutions.

[0078] The image enhancement module consists of six sequentially connected convolutional layers. Each convolutional layer doubles the number of channels and reduces the spatial size to half that of the previous convolutional layer to extract multi-scale feature information of the sharpened image step by step. The output of the last convolutional layer of the image enhancement module generates the parameters required for image enhancement. Based on the required parameters, the sharpened image is enhanced.

[0079] For the i-th convolutional layer, its operation is implemented using the following formula:

[0080] X i =IN(σ(W) i X i-1 +b i ));

[0081] Among them, W i b i X is the weight of the image enhancement module. i Let σ represent the output feature of the i-th layer, where σ is the LeakyReLU activation function used to enhance the nonlinearity of feature representation, and IN represents instance normalization, used to reduce distribution bias between batches, thereby improving the robustness of the model under different weather conditions.

[0082] Through the above structural design, the image enhancement module uses convolutional layers to gradually extract features related to weather components during the forward propagation process and maps them to specific image enhancement parameters, thereby achieving a significant improvement in visual effects under rain / snow weather conditions.

[0083] Step 4: Perform semantic segmentation on the enhanced image to obtain a semantically segmented image with detailed features;

[0084] Downstream Tasks: In autonomous driving systems, the primary goal of image enhancement is to provide clearer and more accurate visual data for downstream detection and perception tasks, thereby improving the system's robustness and safety in complex environments. Through image enhancement, the system can recover the tone and details of objects, enabling object attribute recognition models to more accurately analyze and judge the detailed features of objects. This helps autonomous driving systems make more accurate decisions in complex scenarios, such as making avoidance or overtaking decisions based on vehicle type and direction.

[0085] Step 5: Construct the total loss function and adjust the parameters of the image enhancement model.

[0086] The overall loss function is designed to enable system training and parameter updates, ensuring the system output meets expectations and improving the performance of downstream tasks. The overall loss function specifically includes sharpening loss, exposure loss, and semantic segmentation loss.

[0087] Sharpening Loss: This loss term measures the difference between the generated sharpened image and the target sharpened image, ensuring that rain / snow components are effectively removed. By minimizing the sharpening loss, the image enhancement model can improve the removal of rain and snow interference during training, thereby increasing image sharpness.

[0088] Exposure Loss: This loss term adjusts the overall brightness level of the image to correct brightness deviations caused by rain or snow. By optimizing the exposure loss, the image enhancement model can dynamically adjust the image brightness, resulting in an enhanced image with appropriate brightness that better reflects the visual characteristics of sunny conditions.

[0089] Semantic segmentation loss: This loss term ensures the enhanced image performs well in semantic segmentation tasks. By introducing semantic segmentation loss, the image enhancement model preserves semantic information in the image during the sharpening process, thereby maintaining object boundaries and semantic features in the processed image and improving detection and recognition accuracy in intelligent driving systems.

[0090] By utilizing reference image I taken in clear weather ref Define the prior loss function, the prior loss L. ref for:

[0091]

[0092] By learning the prior loss term, the image enhancement model learns the distribution differences of features in sunny and rainy / snowy weather to improve its ability to identify and separate severe weather.

[0093] The Structural Similarity Index (SSIM) is used to evaluate the sharpened image C output by the sharpening module. in With input severe weather image I in and reference image I ref Similarity, structural similarity index loss L ssim for:

[0094]

[0095] To ensure the enhanced image E in Brightness and contrast meet expectations; however, exposure loss is introduced to quantify the deviation between the exposure level of a local area and the ideal exposure level. The exposure loss L... exp for:

[0096]

[0097] Weighted cross-entropy loss is used to quantify the difference between the model's predicted segmentation map and the true label; semantic segmentation loss L... seg for:

[0098]

[0099] Among them, C(I) ref The image shown is the sharpened image output by the sharpening module. The feature map is all zeros, ∥·∥2 is the L2 norm, SSIM is the structural index, and C in For the sharpened image output by the sharpening module, I in For severe weather images, I ref Here, s is a reference image, a constant set according to the source of the reference image to measure the similarity between the reference image and the severe weather image, ∥·∥1 is the first norm, and M is the number of local regions divided in the enhanced image. It is an average pooling layer used to calculate the exposure level of a local region, where the size of the local region is set to 16×16 pixels, E in For the enhanced image, N is the total number of pixels in the severe weather image, and C is the total number of categories. It is the one-hot encoding of the true value of the k-th category, where GT is the true value. ω is the probability predicted by the image enhancement model that pixel n belongs to category k, where n∈[1,N], k∈[1,C], and ω k This represents the weighting parameter.

[0100] The semantic segmentation loss term ensures that the augmented image retains boundary and semantic information in the semantic segmentation task.

[0101] This embodiment combines prior knowledge and a self-supervised learning strategy, and by designing multiple loss functions such as sharpening, enhancement, and semantic segmentation, it achieves the goals of efficiently removing interference, enhancing sharpness, and improving semantic segmentation performance in rain and snow weather images. The system can separate weather components from images, generate high-quality images that meet practical application needs under different weather conditions, and optimize visual perception through adaptive training, ultimately improving the safety and reliability of autonomous driving systems in adverse weather conditions such as rain and snow.

[0102] That is, this embodiment has the following advantages:

[0103] (1) Introducing prior physical knowledge to enhance image enhancement: This embodiment employs prior physical knowledge such as atmospheric scattering theory and gamma correction, enabling the system to more accurately estimate the impact of haze and light scattering on image clarity when processing rain and snow weather images, thereby effectively reducing image blurring and contrast loss caused by these factors. Compared to traditional image denoising or filtering methods, this embodiment not only restores the brightness and contrast of the image but also retains more image details, thus significantly improving the visual quality of the image.

[0104] (2) Enhancing semantic segmentation performance and improving autonomous driving safety: This embodiment combines semantic segmentation loss, so that image enhancement not only improves visual clarity, but also ensures the image's performance in semantic segmentation tasks. In autonomous driving systems, the accuracy of object detection and semantic segmentation is directly related to driving safety. The model in this embodiment can enhance the semantic information of images under adverse weather conditions, thereby improving the accuracy of target detection and recognition in autonomous driving systems and enhancing the reliability and safety of autonomous driving systems in rainy and snowy weather.

[0105] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for enhancing severe weather images based on prior physical knowledge, characterized in that, The image to be enhanced is input into the trained image enhancement model, which outputs a semantic segmentation image. The training process for the image enhancement model is as follows: Acquire images under different driving conditions in sunny and inclement weather, and construct a training dataset; Using clear weather images as reference images, the distribution differences between clear weather images and severe weather images are learned based on the sharpening module. Weather components are separated from severe weather images to generate sharpened images. A physical prior is introduced in the image enhancement module to perform image enhancement processing on the sharpened image to obtain the enhanced image. The image enhancement module includes six sequentially connected convolutional layers. Each convolutional layer progressively doubles the number of channels and reduces the spatial size to half that of the previous convolutional layer to extract multi-scale feature information of the sharpened image step by step. The last convolutional layer of the image enhancement module outputs the parameters required for image enhancement. Based on the required parameters, image enhancement is performed on the sharpened image. Semantic segmentation is performed on the enhanced image to obtain a semantically segmented image with detailed features; Construct a total loss function and adjust the parameters of the image enhancement model.

2. The severe weather image enhancement method based on prior physical knowledge according to claim 1, characterized in that, The sharpening module includes six sequentially connected convolutional layers; The first three convolutional layers are used to process severe weather images and downsample layer by layer to reduce the size of severe weather images and extract low-level features of severe weather images to obtain the first image; The last three convolutional layers gradually upsample through deconvolution operations to restore the first image to its original resolution and ensure that low-level features in the first image are preserved, thereby generating a sharpened image; Batch regularization is applied after each convolutional layer to stabilize the feature distribution, and the ReLU activation function is used to enhance the non-linear expressive power of the sharpening module.

3. The severe weather image enhancement method based on prior physical knowledge according to claim 1, characterized in that, Introducing physical priors into the image enhancement module, image enhancement processing is performed on the sharpened image, specifically as follows: A model was established based on atmospheric scattering theory to describe the scattering process of light in the atmosphere under severe weather conditions, and to obtain the impact of haze and suspended water molecules on the sharpened image. Gamma correction technology is used to adjust the brightness and contrast of the sharpened image to compensate for the reduction in brightness and contrast shift caused by severe weather, resulting in an enhanced image.

4. The severe weather image enhancement method based on prior physical knowledge according to claim 3, characterized in that, The atmospheric scattering theory model is specifically established as follows: ； in, Image of severe weather at location pixel values, For the real image of the scene in location pixel values, Indicates background brightness. Transmittance is defined as the rate at which light travels from the scene to the camera.

5. The severe weather image enhancement method based on prior physical knowledge according to claim 3, characterized in that, Gamma correction technology is used to adjust the brightness and contrast of the sharpened image, specifically: ； in, These are the pixel values ​​of the input severe weather image. These are the pixel values ​​of the enhanced image. It is the gamma value, when <1 is used to enhance shadow details. A value >1 is used to compress highlight details.

6. The severe weather image enhancement method based on prior physical knowledge according to claim 1, characterized in that, The total loss function includes sharpening loss, exposure loss, and semantic segmentation loss.

7. The severe weather image enhancement method based on prior physical knowledge according to claim 6, characterized in that, The sharpening loss includes prior loss and structural similarity index loss; The prior loss for: ； The structural similarity index loss for: ； The exposure loss for: ； The semantic segmentation loss for: ； in, The image is a sharpened image output by the sharpening module. A feature map consisting entirely of zeros. It is a norm 2. It is a structural index. The sharpened image output by the sharpening module. Images of severe weather. For reference image, To define a constant for measuring the similarity between reference images and severe weather images, For a norm, To increase the number of local regions in the enhanced image, It is an average pooling layer used to calculate the exposure level of local areas. To enhance the image, This represents the total number of pixels in the severe weather image. The total number of categories, It is the first One-hot encoding of the true values ​​of each category For the true value, The pixels predicted by the image enhancement model Category The probability, , , This represents the weighting parameter.

8. A severe weather image enhancement system based on prior physical knowledge, characterized in that, Input rain / snow images into a trained image augmentation model and output semantic segmentation images; The training process of an image enhancement model includes a training set construction module, a sharpening module, an image enhancement module, a semantic segmentation module, and a loss construction module. The training set construction module is used to acquire images under different driving conditions in sunny and inclement weather to build a training dataset; The sharpening module uses a clear sky image as a reference image to learn the distribution differences between clear sky images and severe weather images, separates the weather components from the severe weather image, and generates a sharpened image. The image enhancement module is used to enhance a sharpened image by introducing physical priors, resulting in an enhanced image. The image enhancement module includes six sequentially connected convolutional layers. Each convolutional layer progressively doubles the number of channels and reduces the spatial size to half that of the previous layer to extract multi-scale feature information from the sharpened image step by step. The last convolutional layer of the image enhancement module outputs the parameters required for image enhancement. Based on these parameters, image enhancement is performed on the sharpened image. The semantic segmentation module is used to perform semantic segmentation on the enhanced image to obtain a semantically segmented image with detailed features; The loss construction module is used to construct the total loss function and adjust the parameters of the image enhancement model.

9. The severe weather image enhancement system based on prior physical knowledge according to claim 8, characterized in that, The sharpening module includes six sequentially connected convolutional layers; The first three convolutional layers are used to process severe weather images and downsample layer by layer to reduce the size of severe weather images and extract low-level features of severe weather images to obtain the first image; The last three convolutional layers gradually upsample through deconvolution operations to restore the first image to its original resolution and ensure that low-level features in the first image are preserved, thereby generating a sharpened image; Batch regularization is applied after each convolutional layer to stabilize the feature distribution, and the ReLU activation function is used to enhance the non-linear expressive power of the sharpening module.

Images