Image defogging enhancement and intelligent perception collaborative optimization method based on stream matching

Abstract

The invention provides an image defogging enhancement and intelligent perception collaborative optimization method based on stream matching, and the method comprises the steps: obtaining a foggy input image, and constructing an ordinary differential equation for defining image transformation; the input image is input into a fog perception vector field, the fog perception vector field comprises an atmospheric scattering purifier and a defogging perception color lookup table, and the atmospheric scattering purifier is used for extracting multi-scale features from the foggy input image to generate clearer output; the defogging perception color lookup table adaptively adjusts the image color through a nonlinear color transformation mechanism; and solving the ordinary differential equation through an RK4 solver, iteratively updating the foggy input image at each time step, and ensuring that the input image can be stably converted into a clear image from an initial foggy state. The technical problems that in the prior art, the existing method is forced to reduce the model capacity due to the calculation efficiency constraint in the ultra-high-definition image defogging, so that the fog concentration estimation is incomplete, and the perception quality and the calculation efficiency cannot be considered at the same time are solved.

Description

Technical Field

[0001] The present invention relates to the technical field of high-definition image defogging, and in particular to a method for collaborative optimization of image defogging enhancement and intelligent perception based on flow matching. Background Art

[0002] With the widespread adoption of 4K imaging devices, existing dehazing methods face a dual dilemma regarding resolution adaptability. Traditional algorithms based on physical models (such as atmospheric scattering models) are prone to color distortion and artifacts under complex atmospheric conditions, making them difficult to meet the stringent texture detail requirements of ultra-high-definition (UHD) images. While data-driven deep learning methods can improve dehazing quality, their parameter count and computational complexity increase exponentially with resolution, resulting in GPU memory usage exceeding 12GB and frame rates below 30fps when processing a single 4K image, severely limiting their practical deployment value. Furthermore, existing methods are often forced to reduce model capacity for UHD image dehazing due to computational efficiency constraints, resulting in incomplete fog concentration estimation and, in turn, reduced accuracy in downstream visual tasks such as object detection and semantic segmentation. Therefore, developing an UHD dehazing framework that balances perceptual quality and computational efficiency has become a pressing challenge in the field of computer vision.

[0003] Currently, existing UHD dehazing methods, such as 4Kdehazing, 4K-Haze, and LapDehazeNet, primarily rely on traditional convolutional neural networks, which have large model parameters. For example, 4KDehazing requires 34.55M trainable parameters, which significantly increases computational complexity and inference time. Although existing general-purpose UHD image restoration models, such as UHDformer and UHDDIP, have smaller parameter counts (0.34M and 0.81M, respectively), they rely on additional prior information, such as normal maps, and their dehazing performance is inferior to models specifically designed for this task. Summary of the Invention

[0004] The purpose of the present invention is to overcome the above-mentioned technical deficiencies and provide a method for collaborative optimization of image dehazing enhancement and intelligent perception based on flow matching, so as to solve the technical problem that existing methods in the prior art are often forced to reduce model capacity due to computational efficiency constraints in ultra-high-definition image dehazing, resulting in incomplete estimation of fog concentration, thereby reducing the accuracy of downstream visual tasks such as target detection and semantic segmentation, and failing to take into account both perception quality and computational efficiency.

[0005] To achieve the above technical objectives, the technical solution of the present invention provides a method for collaborative optimization of image defogging enhancement and intelligent perception based on flow matching, comprising the steps of: Obtaining a foggy input image and constructing an ordinary differential equation defining a transformation of the image, wherein the ordinary differential equation is defined by a fog-aware vector field; Inputting the input image into a fog-aware vector field, the fog-aware vector field comprising: an atmospheric scattering purifier and a defogging-aware color lookup table, wherein the atmospheric scattering purifier is configured to extract multi-scale features from the foggy input image to generate a clearer output, and the defogging-aware color lookup table adaptively adjusts the image color through a nonlinear color transformation mechanism; The ordinary differential equation is solved by the RK4 solver, and the foggy input image is iteratively updated at each time step to ensure that the input image can be smoothly and stably transformed from an initial foggy state to a clear image.

[0006] Compared with the prior art, the present invention has the following beneficial effects:

[0007] Flow matching provides an efficient solution. By directly learning the mapping relationship between data distributions, flow matching avoids the complex iterative process in traditional generative models, significantly reducing computational complexity and video memory usage. When processing 4K images, the flow matching method of this application reduces the number of parameters to 7.9M and the video memory requirement to 3.2GB, significantly improving computational efficiency. In addition, flow matching can efficiently model the complex distribution of high-resolution images without relying on additional prior information, thereby achieving better detail preservation and color restoration capabilities in dehazing tasks. Compared with UHDformer and UHDDIP, flow matching improves the PSNR indicator by about 2dB, achieving a better balance between performance and efficiency. Therefore, this application attempts to introduce this technology to solve the challenges of ultra-high-definition dehazing.

[0008] Dehazing based on flow matching differs from progressive and diffusion dehazing strategies. Progressive dehazing methods restore a haze-free image through a phased optimization process. These methods typically utilize multi-scale feature extraction, iterative optimization, or layered processing strategies to gradually refine the dehazing results, effectively mitigating the problems of detail loss or over-dehazing that can occur during a single dehazing process. However, progressive dehazing methods typically rely on discretization iterations, gradually optimizing the dehazing results through preset discretization steps. This fixed path can lead the algorithm to become trapped in local optima, making it difficult to find a global optimal solution, especially in complex scenes or under variable haze conditions. Furthermore, progressive optimization based on physical priors lacks the ability to adaptively model complex haze distributions and is prone to color shift in scenes with non-uniform fog or multi-scale fog clusters. Diffusion model dehazing methods, by simulating the natural diffusion process of data distribution, can optimize dehazing results in a more flexible and global manner. However, their back-diffusion process requires hundreds of iterative computations, resulting in inference times of up to tens of seconds for a single 4K image. Furthermore, the noise-sensitive nature of diffusion models can easily cause content and color distortion in low-texture areas. Although some studies have proposed accelerated sampling strategies, existing solutions still face significant bottlenecks in balancing model lightweighting with fidelity.

[0009] This application proposes a method for ultra-high-definition image dehazing based on flow matching and a dehazing-aware color lookup table (Haze-LUT), named 4KDehazeFlow. Unlike traditional dehazing methods, flow matching models the dehazing process as solving an ordinary differential equation (ODE) for a continuous vector field. By directly learning the mapping from hazy to clear images, it avoids the local optimum problem caused by discretization iterations. Furthermore, this application designs a learnable three-dimensional lookup table module that utilizes a data-driven adaptive nonlinear color transformation mechanism to achieve refined dehazing effects while ensuring efficient computation. Specifically, this application constructs a general flow matching dehazing framework whose core components include an initial distribution, a target distribution, a vector field, and an ODE solver. First, the hazy image is defined as the initial distribution, and the clear image as the target distribution. The goal is to gradually map the initial distribution to the target distribution through flow matching. Second, during the flow matching process, the vector field defines the transformation path for image dehazing. To further optimize the dehazing effect, this application introduces an improved CNN network in the vector field to extract multi-scale features of the image, and combines it with a color-aware dehazing lookup table (Haze-LUT) module to adaptively adjust the color and brightness of the image through a nonlinear color transformation mechanism, thereby improving the accuracy and visual quality of the dehazing effect. Finally, this application adopts a fourth-order Runge-Kutta (RK4) differential equation solver and uses an accurate step-by-step iterative method to stably solve the dehazing flow field. The RK4 method effectively balances computational accuracy and stability by calculating the weighted average of four stages within each time step, ensuring the precision and reliability of the dehazing process. In the inference stage, based on the learned vector field and the dehazing-aware color lookup table (Haze-LUT), the input data is mapped to the target distribution by inversely solving the ordinary differential equation, thereby achieving accurate image dehazing.

[0010] According to some embodiments of the present invention, the defogging process is controlled by the ordinary differential equation: ; in represents the time step, It's time images, is Parameterized fog-aware vector field, is the initial time step, For the input image, by solving the ordinary differential equation, the input image will gradually change from a foggy state to a clear state along a probability path.

[0011] According to some embodiments of the present invention, the process of processing the input image by the fog perception vector field is expressed as follows: ; in, represents the output of the fog perception vector field, is the current input image, is the output image obtained by the atmospheric scattering purifier, It is the result of color adjustment through the dehazing-aware color lookup table. is a scaling factor that controls the intensity of the color transform.

[0012] According to some embodiments of the present invention, an atmospheric scattering model is designed in the atmospheric scattering purifier, which is expressed as: ; in, It is a clear image. is the input foggy image, is a learnable scaling parameter, is a learnable bias term whose initial value is set to 1.

[0013] According to some embodiments of the present invention, a convolutional neural network (CNN) is used to estimate directly from the input , the dehazed image can be obtained by: ; in, is the current input image, is the output of the enhanced convolutional neural network. The convolutional neural network (CNN) consists of three downsampling encoder modules, a spatial attention module, and three upsampling decoder modules: ; in, is the current input image, represents the downsampling encoder module, Refers to the attention mechanism, Table upsampling decoder module.

[0014] According to some embodiments of the present invention, a three-dimensional lookup table represents a three-dimensional lattice, consisting of elements, denoted as ,in is the number of quantization levels for each color channel; Each element Corresponds to RGB color index and its transformed output RGB color , the accuracy of this transformation is determined by The transformation of the dehazing-aware color lookup table is achieved through two key steps: lookup and trilinear interpolation.

[0015] According to some embodiments of the present invention, for an input RGB color , the coordinates in the lookup table is calculated as follows: ; in, Is the maximum color value; after determining the position, the output RGB color It is obtained by trilinear interpolation from 8 adjacent lattice vertices; finally, the dehazing-aware color lookup table process is expressed as: ; in, is the input image, represents the trilinear interpolation function, is a dehaze-aware color lookup table containing the sampled output RGB values.

[0016] In a second aspect, the present application provides an ultra-high-definition image defogging system based on flow matching, comprising: Data acquisition module, used to obtain foggy input images; An equation building module, constructing an ordinary differential equation defining an image transformation, wherein the ordinary differential equation is defined by a fog perception vector field; A fog-aware vector field, comprising: an atmospheric scattering purifier and a defogging-aware color lookup table, wherein the atmospheric scattering purifier is used to extract multi-scale features from a foggy input image to generate a clearer output; and the defogging-aware color lookup table adaptively adjusts the image color through a nonlinear color transformation mechanism; The image defogging module solves the ordinary differential equation using the RK4 solver and iteratively updates the foggy input image at each time step to ensure that the input image can be smoothly and stably transformed from the initial foggy state to a clear image.

[0017] In a third aspect, the present application provides a computer-readable storage medium storing computer-executable instructions, which are used to enable a computer to execute the image dehazing enhancement and intelligent perception collaborative optimization method based on flow matching as described in any one of the first aspects.

[0018] Additional aspects and advantages of the present invention will be set forth in part in the description which follows and, in part, will be obvious from the description which follows, or may be learned by practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] The above and / or additional aspects and advantages of the present invention will become apparent and easily understood from the description of the embodiments in conjunction with the following drawings, wherein the abstract drawing is identical to one of the drawings in the specification: Figure 1 A flow chart of a method for collaborative optimization of image dehazing enhancement and intelligent perception based on flow matching provided by one embodiment of the present invention; Figure 2 A flowchart of a method for collaborative optimization of image dehazing enhancement and intelligent perception based on flow matching provided by another embodiment of the present invention; Figure 3 A vector field diagram of a flow matching-based image dehazing enhancement and intelligent perception collaborative optimization method provided by another embodiment of the present invention. DETAILED DESCRIPTION

[0020] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0021] It should be noted that although the system diagrams illustrate functional module divisions and the flowcharts illustrate logical sequences, in certain circumstances, the steps shown or described may be performed in a sequence that differs from the module divisions in the system or the sequence in the flowcharts. The terms "first," "second," and so on, in the specification, claims, and drawings, are used to distinguish similar items and are not necessarily used to describe a specific sequence or precedence.

[0022] Reference Figures 1 to 3 , Figure 1 A flow chart of a method for collaborative optimization of image dehazing enhancement and intelligent perception based on flow matching provided by one embodiment of the present invention; Figure 2 A flowchart of a method for collaborative optimization of image dehazing enhancement and intelligent perception based on flow matching provided by another embodiment of the present invention; Figure 3 A vector field diagram of a method for collaborative optimization of image defogging enhancement and intelligent perception based on flow matching provided by another embodiment of the present invention. The method for collaborative optimization of image defogging enhancement and intelligent perception based on flow matching includes but is not limited to the following steps: Step S110, obtaining a foggy input image and constructing an ordinary differential equation defining an image transformation, where the ordinary differential equation is defined by a fog perception vector field; Step S110: Input the input image into a fog-aware vector field. The fog-aware vector field includes an atmospheric scattering purifier and a defogging-aware color lookup table. The atmospheric scattering purifier is used to extract multi-scale features from the foggy input image to generate a clearer output. The defogging-aware color lookup table adaptively adjusts the image color through a nonlinear color transformation mechanism. In step S110 , the ordinary differential equation is solved by the RK4 solver, and the foggy input image is iteratively updated at each time step, so that the input image is smoothly and stably transformed from the initial foggy state to a clear image.

[0023] In one embodiment, a flow matching-based image defogging enhancement and intelligent perception collaborative optimization method includes the steps of: obtaining a foggy input image, constructing an ordinary differential equation that defines the image transformation, where the ordinary differential equation is defined by a fog-perceived vector field; inputting the input image into the fog-perceived vector field, where the fog-perceived vector field includes: an atmospheric scattering purifier and a defogging-perceived color lookup table, where the atmospheric scattering purifier is used to extract multi-scale features from the foggy input image to generate a clearer output, and the defogging-perceived color lookup table adaptively adjusts the image color through a nonlinear color transformation mechanism; solving the ordinary differential equation through an RK4 solver, and iteratively updating the foggy input image at each time step to ensure that the input image can smoothly and stably transform from an initial foggy state to a clear image.

[0024] The atmospheric scattering cleaner extracts multi-scale features from foggy input images to produce clearer output. Multi-scale feature extraction captures image information at different resolutions and levels of detail, helping to comprehensively remove the effects of fog. For example, fog in distant and near scenes has different features at different scales. The cleaner accurately analyzes and processes these differences, better restoring the image's original clarity. The dehazed image retains overall contours while displaying rich details. The dehaz-aware color lookup table adaptively adjusts image color using a nonlinear color transformation mechanism. During the dehazing process, fog often causes image color distortion and reduced contrast. This lookup table adaptively adjusts color based on the actual dehazing condition, avoiding the color disharmony and unnaturalness that can occur with traditional fixed color adjustment methods. This results in more realistic and vivid colors in the dehazed image, consistent with human visual perception and improving image quality.

[0025] The RK4 solver is used to solve ordinary differential equations and iteratively update the foggy input image at each time step. This approach ensures a smooth and stable transition from the initial foggy state to a clear image. This approach avoids issues such as image abrupt changes and local over-processing that can occur during the dehazing process, resulting in a more uniform and natural dehazing effect. The overall image transformation is gradual, and the resulting clear image is of reliable quality. In practical applications, this approach delivers stable and high-quality dehazing performance for both static ultra-high-definition images and continuous video frames (when processing video frame by frame).

[0026] This application proposes a method for ultra-high-definition image dehazing based on flow matching and a dehazing-aware color lookup table (Haze-LUT), named 4KDehazeFlow. Unlike traditional dehazing methods, flow matching models the dehazing process as solving an ordinary differential equation (ODE) for a continuous vector field. By directly learning the mapping from hazy to clear images, it avoids the local optimum problem caused by discretization iterations. Furthermore, this application designs a learnable three-dimensional lookup table module that utilizes a data-driven adaptive nonlinear color transformation mechanism to achieve refined dehazing effects while ensuring efficient computation. Specifically, this application constructs a general flow matching dehazing framework whose core components include an initial distribution, a target distribution, a vector field, and an ODE solver. First, the hazy image is defined as the initial distribution, and the clear image as the target distribution. The goal is to gradually map the initial distribution to the target distribution through flow matching. Second, during the flow matching process, the vector field defines the transformation path for image dehazing. To further optimize the dehazing effect, this application introduces an improved CNN network in the vector field to extract multi-scale features of the image, and combines it with a color-aware dehazing lookup table (Haze-LUT) module to adaptively adjust the color and brightness of the image through a nonlinear color transformation mechanism, thereby improving the accuracy and visual quality of the dehazing effect. Finally, this application adopts a fourth-order Runge-Kutta (RK4) differential equation solver and uses an accurate step-by-step iterative method to stably solve the dehazing flow field. The RK4 method effectively balances computational accuracy and stability by calculating the weighted average of four stages within each time step, ensuring the precision and reliability of the dehazing process. In the inference stage, based on the learned vector field and the dehazing-aware color lookup table (Haze-LUT), the input data is mapped to the target distribution by inversely solving the ordinary differential equation, thereby achieving accurate image dehazing.

[0027] Among them, the vector field It consists of an atmospheric scattering purifier and a dehazing-aware color lookup table, which are used to extract multi-scale features and perform adaptive color correction. The dehazing process is controlled by an ordinary differential equation: ; in represents the time step, It's time images, is Parameterized fog-aware vector field, is the initial time step, For the input image, by solving the ordinary differential equation, the input image will gradually change from a foggy state to a clear state along a probabilistic path.

[0028] During the flow matching process, the vector field defines the transformation path for image dehazing, modeling the dynamic changes of image distribution from foggy to clear over time. To achieve this goal, an atmospheric scattering purifier was developed, inspired by AODNet, which directly generates clear images through a lightweight convolutional neural network (CNN). This approach avoids separate transmittance and atmospheric light estimation, can capture multi-scale features, and improve the dehazing effect. However, since flow matching can perform the dehazing function in multiple steps, it may cause image color loss. To further improve color accuracy, we propose a dehazing-aware color lookup table, which is a three-dimensional lookup table (LUT) that adaptively adjusts the image color through a nonlinear color transformation mechanism. The processing process of the fog-aware vector field on the input image is expressed as: ; in, represents the output of the fog perception vector field, is the current input image, is the output image obtained by the atmospheric scattering purifier, It is the result of color adjustment through the dehazing-aware color lookup table. is a scaling factor that controls the intensity of the color transform.

[0029] Atmospheric scattering purifier: The atmospheric scattering cleaner in 4KDehazeFlow plays a key role in extracting multi-scale features from hazy input images to produce clearer output. This convolutional neural network directly generates a clear image without explicitly estimating transmittance maps or atmospheric light, providing a lightweight and efficient dehazing solution. The atmospheric scattering cleaner is designed with an atmospheric scattering model expressed as: ; in, It is a clear image. is the input foggy image, is a learnable scaling parameter, is a learnable bias term with an initial value of 1.

[0030] Convolutional neural networks (CNNs) are used to estimate directly from the input , the dehazed image can be obtained by: ; in, is the current input image, is the output of the enhanced convolutional neural network, which consists of three downsampling encoder modules, a spatial attention module, and three upsampling decoder modules: ; in, is the current input image, represents the downsampling encoder module, Refers to the attention mechanism, Table upsampling decoder module. The encoder extracts hierarchical features while gradually reducing the spatial dimensionality through max pooling, convolution, adaptive normalization, and Gaussian Error Linear Unit (GELU) activation. The spatial attention module highlights important regions, while the decoder optimizes features through bilinear upsampling and skip connections to recover fine details. The network uses convolutional kernels of size 1×1 or 3×3.

[0031] Dehaze-aware color lookup table: Although atmospheric scattering purifiers can effectively remove haze, they often cause color loss or distortion in the process. To alleviate these problems, we propose a dehazing-aware color lookup table, a three-dimensional lookup table (LUT) that adaptively adjusts image colors through a nonlinear color transformation mechanism. A three-dimensional lookup table represents a three-dimensional lattice composed of elements, denoted as ,in is the number of quantization levels for each color channel. Each element Corresponds to RGB color index and its transformed output RGB color , the accuracy of this transformation is determined by In practice, it is usually set to 33. The transformation of the dehazing-aware color lookup table is achieved through two key steps: lookup and trilinear interpolation. , the coordinates in the lookup table is calculated as follows: ; in, Is the maximum color value. After determining the position, the output RGB color It is obtained by trilinear interpolation from 8 adjacent lattice vertices. Finally, the dehazing-aware color lookup table process is expressed as: ; in, is the input image, represents the trilinear interpolation function, is a dehaze-aware color lookup table containing the sampled output RGB values.

[0032] Improved results: Comprehensive experiments are conducted on ultra-high-definition (UHD) and non-UHD datasets to evaluate the performance of the proposed method. The proposed method is compared with seven state-of-the-art image dehazing techniques, including Dehamer, C2PNet, MB-TaylorFormer, PTTD, 4KDehazing, UHDformer, and UHDDIP.

[0033] This application uses four widely recognized metrics to evaluate performance: PSNR, SSIM, LPIPS, and NIQE. PSNR and SSIM focus on assessing the fidelity and structural integrity of the reconstructed image, while LPIPS assesses perceptual quality by measuring the similarity of deep features, providing an assessment more consistent with human visual quality. NIQE is a no-reference image quality evaluation metric that measures the degree of image distortion by counting the statistical features of natural scenes in the image. It can evaluate the naturalness and authenticity of the dehazing results without a reference image.

[0034] Among them, PSNR (Peak Signal-to-Noise Ratio) is a common indicator for measuring image quality. It is mainly used to evaluate the difference between the processed image and the original image. The larger the value, the better the image quality. The calculation formula is: ; ; Among them, MSE stands for mean square error, which is used to measure the difference between two images at the pixel level. and Respectively represent the generated fusion image and the corresponding gold standard image (ground truth) at position m and n represent the width and height of the image respectively, and MAX represents the maximum value of the image pixels.

[0035] SSIM (Structural Similarity Index Measure) is an indicator used to evaluate the quality of two images. It mainly measures the similarity of images by comparing their brightness, contrast, and structural information. The closer the value is to 1, the better the image quality. The calculation formula of SSIM is: ; ; ; in, is the fused image output by the model The average value of is with Corresponding gold standard The average value of It represents an image and The covariance of yes The variance of yes variance; is the dynamic range of pixel values, and Represents the preset hyperparameters, here =0.01, =0.03, and represents the smoothing parameter.

[0036] The Learned Perceptual Image Patch Similarity (LPIPS) metric is used to evaluate the perceptual similarity between two images. It is based on features extracted by a deep convolutional neural network, rather than simple pixel differences. Lower values ​​indicate better image quality. The calculation formula for LPIPS is: ; in, represents the fused image output by the model, express The corresponding gold standard image, and Represents the first Layer feature map. is an image In the Layer characteristics, is an image In the Characteristics of the layer. Represents the Euclidean distance (L2 norm) between feature maps. This is the feature map and The difference metric between the two images is used to measure the Similarity in layer features. Represents the weighted coefficient of each layer’s feature map. Different network layers have different effects on perceived similarity, and the weights Used to reflect this, it is usually learned through training. Indicates that all The results of the layers are summed up to comprehensively consider the feature differences at different levels.

[0037] NIQE (Natural Image Quality Evaluator) is a non-reference image quality evaluation metric used to measure the naturalness and distortion of an image. It constructs a quality model based on the statistical characteristics of natural images, rather than relying on a reference image for comparison. Lower values ​​indicate higher image quality.

[0038] The calculation formula of NIQE is: ; in, is the feature mean vector of the natural image (reference model), is the covariance matrix of the natural image (reference model), is the feature mean vector of the image to be evaluated, is the covariance matrix of the image to be evaluated, is the matrix transpose operation, is the inverse operation of the matrix, is the mean of the covariance matrix.

[0039] Table 1 Quantitative comparison on ultra-high-definition datasets

[0040] As can be seen from Table 1, the method 4KDehazeFlow of this application achieved the best performance on the 4KID dataset, achieving the highest PSNR and the lowest LPIPS. Among them, PSNR improved by 0.11 dB compared to the second-best method, while LPIPS decreased by 0.012, indicating that the dehazed image has improved in both reconstruction quality and perceptual quality. On the I-HAZE dataset, the method of this application achieved the best results in PSNR, SSIM, and LPIPS, further verifying its robustness and superiority on different datasets. In the O-HAZE dataset, 4KDehazeFlow performed best in the SSIM and LPIPS indicators, with SSIM being 0.1755 higher than the second-best method and LPIPS being 0.0473 lower, indicating that it has obvious advantages in maintaining image structural integrity and perceptual quality. Overall, on the three ultra-high-definition datasets, the method 4KDehazeFlow proposed in this application achieved the best results in terms of average PSNR, SSIM and LPIPS indicators, fully demonstrating that its dehazing effect in ultra-high-definition scenes is better than existing advanced methods.

[0041] Table 2 Quantitative comparison on SOTS dataset

[0042] In order to verify the dehazing effect of the method of this application in non-UHD scenes, this application further conducted experiments on the low-resolution dataset SOTS. The experimental results are shown in Table 2. 4KDehazeFlow achieved the best performance in both NIQE and LPIPS indicators, indicating that the dehazed images are superior to other methods in terms of naturalness and perceptual quality. In addition, in terms of the SSIM indicator, the method of this application also performed well, only slightly lower than the MB-TaylorFormer method by 0.0101, and still maintained a high level of structural fidelity. These results further demonstrate the robustness and superiority of 4KDehazeFlow in scenes with different resolutions. It is not only applicable to UHD datasets, but is also competitive on low-resolution datasets.

[0043] Table 3 Comparison of computational efficiency of ultra-high-definition methods

[0044] This application also evaluates model complexity. Table 3 compares the computational overhead of different ultra-high-definition (UHD) dehazing methods. Compared to UHDformer and UHDDIP, 4KDehazeFlow significantly reduces the number of multiplication-add operations (MACs) while maintaining competitive efficiency, effectively reducing computational cost. Notably, this application's method flexibly adjusts the number of flow matching steps to balance inference time and performance requirements. Therefore, 4KDehazeFlow achieves a more optimal trade-off between computational efficiency and dehazing quality.

[0045] In one embodiment, an ultra-high-definition image defogging system based on flow matching includes: a data acquisition module for acquiring a foggy input image; a program construction module for constructing an ordinary differential equation that defines an image transformation, wherein the ordinary differential equation is defined by a fog-perceiving vector field; the fog-perceiving vector field includes: an atmospheric scattering purifier and a defogging-perceiving color lookup table, wherein the atmospheric scattering purifier is used to extract multi-scale features from the foggy input image to generate a clearer output, and the defogging-perceiving color lookup table adaptively adjusts the image color through a nonlinear color transformation mechanism; an image defogging module for solving the ordinary differential equation through an RK4 solver, and iteratively updating the foggy input image at each time step to ensure that the input image can smoothly and stably transform from an initial foggy state to a clear image.

[0046] The memory, as a non-transient computer-readable storage medium, can be used to store non-transient software programs and non-transient computer executable programs. In addition, the memory may include a high-speed random access memory and may also include a non-transient memory, such as at least one disk storage device, a flash memory device, or other non-transient solid-state storage device. In some embodiments, the memory may optionally include a memory remotely located relative to the processor, and these remote memories may be connected to the processor via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0047] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate and may be located in one place or distributed across multiple network units. Some or all of the modules may be selected based on actual needs to achieve the objectives of this embodiment.

[0048] In addition, an embodiment of the present invention also provides a computer-readable storage medium, which stores computer-executable instructions. The computer-executable instructions are executed by a processor or controller, for example, by a processor in the above-mentioned terminal embodiment, so that the above-mentioned processor can execute the image dehazing enhancement and intelligent perception collaborative optimization method based on flow matching in the above-mentioned embodiment.

[0049] Those skilled in the art will appreciate that all or some of the steps and systems disclosed above can be implemented as software, firmware, hardware, or any suitable combination thereof. Some or all of the physical components may be implemented as software executed by a processor, such as a central processing unit, digital signal processor, or microprocessor, or as hardware, or as an integrated circuit, such as an application-specific integrated circuit. Such software may be distributed on computer-readable media, which may include computer storage media (or non-transitory media) and communication media (or transient media). As is well known to those skilled in the art, the term computer storage media includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information, such as computer-readable instructions, data structures, program modules, or other data. Computer storage media includes, but is not limited to, RAM, ROM, EEPROM, flash memory or other memory technologies, CD-ROM, digital versatile disks (DVDs) or other optical disk storage, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other medium that can be used to store the desired information and can be accessed by a computer. Furthermore, as is well known to those skilled in the art, communication media typically embodies computer-readable instructions, data structures, program modules, or other data in a modulated data signal such as a carrier wave or other transport mechanism, and may include any information delivery media.

[0050] The above is a specific description of the preferred implementation of the present invention, but the present invention is not limited to the above implementation. Those skilled in the art can make various equivalent modifications or substitutions without violating the spirit of the present invention. These equivalent modifications or substitutions are all included in the scope defined by the claims of the present invention.

[0051] The specific embodiments of the present invention described above do not limit the scope of protection of the present invention. Any other corresponding changes and modifications made based on the technical concept of the present invention should be included in the scope of protection of the claims of the present invention.

Claims

1. A collaborative optimization method for image dehazing enhancement and intelligent perception based on flow matching, characterized in that: Including steps: Obtaining a foggy input image and constructing an ordinary differential equation defining a transformation of the image, wherein the ordinary differential equation is defined by a fog-aware vector field; Inputting the input image into a fog-aware vector field, the fog-aware vector field comprising: an atmospheric scattering purifier and a defogging-aware color lookup table, wherein the atmospheric scattering purifier is configured to extract multi-scale features from the foggy input image to generate a clearer output, and the defogging-aware color lookup table adaptively adjusts the image color through a nonlinear color transformation mechanism; The ordinary differential equation is solved by an RK4 solver, and the foggy input image is iteratively updated at each time step, so that the input image is smoothly and stably transformed from an initial foggy state to a clear image.

2. The image dehazing enhancement and intelligent perception collaborative optimization method based on flow matching according to claim 1 is characterized in that: The dehazing process is governed by the ordinary differential equation: ； in represents the time step, It's time images, is Parameterized fog-aware vector field, is the initial time step, For the input image, by solving the ordinary differential equation, the input image will gradually change from a foggy state to a clear state along a probability path.

3. The image dehazing enhancement and intelligent perception collaborative optimization method based on flow matching according to claim 2 is characterized in that: The processing process of the fog perception vector field on the input image is expressed as: ; in, represents the output of the fog perception vector field, is the current input image, is the output image obtained by the atmospheric scattering purifier, It is the result of color adjustment through the dehazing-aware color lookup table. is a scaling factor that controls the intensity of the color transform.

4. The image dehazing enhancement and intelligent perception collaborative optimization method based on flow matching according to claim 3 is characterized in that: An atmospheric scattering model is designed in the atmospheric scattering purifier, which is expressed as: ； in, It is a clear image. is the input foggy image, is a learnable scaling parameter, is a learnable bias term with an initial value of 1.

5. The image dehazing enhancement and intelligent perception collaborative optimization method based on flow matching according to claim 4 is characterized in that: Convolutional neural networks (CNNs) are used to estimate directly from the input , the dehazed image can be obtained by: ； in, is the current input image, is the output of the enhanced convolutional neural network, which consists of three downsampling encoder modules, a spatial attention module, and three upsampling decoder modules: ； in, is the current input image, represents the downsampling encoder module, Refers to the attention mechanism, Table upsampling decoder module.

6. The image dehazing enhancement and intelligent perception collaborative optimization method based on flow matching according to claim 5 is characterized in that: A 3D lookup table represents a 3D lattice, consisting of elements, denoted as ,in is the number of quantization levels for each color channel; Each element Corresponds to RGB color index and its transformed output RGB color , the accuracy of this transformation is determined by The transformation of the dehazing-aware color lookup table is achieved through two key steps: lookup and trilinear interpolation.

7. The image dehazing enhancement and intelligent perception collaborative optimization method based on flow matching according to claim 6 is characterized in that: For the input RGB color , the coordinates in the lookup table is calculated as follows: ; in, Is the maximum color value; after determining the position, the output RGB color It is obtained by trilinear interpolation from 8 adjacent lattice vertices; finally, the dehazing-aware color lookup table process is expressed as: ; in, is the input image, represents the trilinear interpolation function, is a dehaze-aware color lookup table containing the sampled output RGB values.

8. An ultra-high-definition image defogging system based on flow matching, characterized in that: include: Data acquisition module, used to obtain foggy input images; An equation building module, constructing an ordinary differential equation defining an image transformation, wherein the ordinary differential equation is defined by a fog perception vector field; A fog-aware vector field, comprising: an atmospheric scattering purifier and a defogging-aware color lookup table, wherein the atmospheric scattering purifier is used to extract multi-scale features from a foggy input image to generate a clearer output; and the defogging-aware color lookup table adaptively adjusts the image color through a nonlinear color transformation mechanism; The image defogging module solves the ordinary differential equation using the RK4 solver and iteratively updates the foggy input image at each time step to ensure that the input image can be smoothly and stably transformed from the initial foggy state to a clear image.

9. A computer-readable storage medium, characterized in that The computer-readable storage medium stores computer-executable instructions, and the computer-executable instructions are used to enable a computer to execute the image dehazing enhancement and intelligent perception collaborative optimization method based on flow matching according to any one of claims 1 to 7.

Images