Coal mine underground image defogging method and system

Abstract

The invention provides an underground coal mine image defogging method and system. The method comprises the following steps: extracting a bright channel image and a dark channel image; gamma transformation is carried out on the bright channel map, and quadtree local atmospheric light estimation is carried out on the dark channel map; fusing the dual-channel atmospheric light and the transmissivity, and refining a result by using gradient guided filtering; performing image restoration by using an atmospheric degradation model; and carrying out color correction and CLAHE detail enhancement processing on the image. The technical problems of color distortion, lack of a real data set, high time complexity, poor algorithm instantaneity and adaptability to a complex environment, and image information distortion are solved.

Description

Technical Field

[0001] The present invention relates to the technical field of coal mine underground image processing, and in particular to a coal mine underground image defogging method and system. Background Art

[0002] With my country's increasing emphasis on coal mine safety, the application of intelligent video surveillance systems in underground coal mines is becoming increasingly widespread. Underground image and video technologies encompass a wide range of applications, including video-based personnel recognition, image processing-based fire detection, mine equipment condition monitoring, and production safety monitoring systems. However, the unique environment of underground coal mines often involves large amounts of suspended particles such as dust, soot, and water vapor, creating a severe haze layer that significantly impacts image recognition quality. In these high-dust and low-light environments, traditional image dehazing algorithms often struggle to effectively process dehazed images, resulting in color distortion, lack of detail, and excessive darkness. This severely reduces the reliability and effectiveness of the images and hinders accurate assessment and monitoring of the coal mine environment. The unique underground environment of coal mines, characterized by the presence of large amounts of suspended particles such as dust, soot, and water vapor, creates a distinct haze layer. These factors significantly impact image clarity and contrast, resulting in surveillance images often exhibiting low brightness, low contrast, and blurred details, significantly reducing the recognition accuracy and practicality of the monitoring system.

[0003] The current mainstream image dehazing algorithms mainly include:

[0004] (1) Physical model-based algorithms: A typical example is the Dark Channel Prior (DCP) algorithm, which uses the physical degradation model of the image to inversely estimate the scene transmittance and atmospheric light value to achieve defogging;

[0005] (2) Algorithms based on image enhancement: image contrast is improved through methods such as image histogram equalization and brightness reconstruction;

[0006] (3) Deep learning-based algorithms: Use convolutional neural networks, generative adversarial networks, etc. to learn image restoration patterns from massive samples.

[0007] In recent years, researchers have proposed a variety of image dehazing algorithms. These algorithms can be categorized into three main categories: those based on image enhancement, those based on physical models, and those based on deep learning. Among these, traditional dark channel prior algorithms based on physical models and deep learning-based image dehazing algorithms are the most widely used. However, these dehazing algorithms still suffer from the following drawbacks and deficiencies in practical applications:

[0008] Traditional dehazing algorithms, such as the dark channel prior algorithm, will produce dehazed images with obvious color distortion when there are large white areas in the scene, limiting their application. Traditional dark channel prior algorithms are prone to incorrectly estimating atmospheric light values ​​when processing large bright areas (such as white walls and lights), resulting in dark and color casts in the overall image tone.

[0009] Deep learning algorithms lack real-world datasets for coal mining applications and exhibit high time complexity. Their real-time performance and adaptability to complex environments need improvement. While theoretically promising, deep learning algorithms lack real-world training data in scenarios like coal mining, resulting in insufficient restoration results for edges or textured areas. Deep learning methods require significant computing resources, making them difficult to meet the real-time processing demands of embedded or edge devices.

[0010] Dehazing algorithms must balance clarity and detail preservation in image processing, but traditional dehazing algorithms are prone to over-enhancement, leading to information distortion. Because coal mine images contain both bright and dark areas, a single channel prior (such as a dark channel) struggles to account for these varying brightness levels, often leading to over- or under-processing.

[0011] In summary, the existing technologies have technical problems such as color distortion, lack of real data sets, high time complexity, poor algorithm real-time performance and adaptability to complex environments, and distortion of image information. Summary of the Invention

[0012] The technical problem to be solved by the present invention is: how to solve the technical problems of color distortion, lack of real data sets, high time complexity, poor algorithm real-time performance and adaptability to complex environments, and image information distortion in the existing technology.

[0013] The present invention solves the above technical problems by adopting the following technical solutions: A method for defogging an image in an underground coal mine comprises:

[0014] S1. Perform dual-channel atmospheric light value fusion, perform maximum filtering and minimum filtering on the input haze image, and obtain the bright channel image and dark channel image corresponding to the haze image respectively;

[0015] S2. Use gamma transform on the bright channel image to adjust the bright channel image and obtain the bright channel atmospheric light value A. g ,Use quadtree to search local atmospheric light in dark channel image and obtain global atmospheric light value of dark channel;

[0016] S3, perform transmittance map estimation and refinement operations, according to the bright channel atmospheric light value A g , dark channel global atmospheric light value, obtain the corresponding bright channel transmittance map, dark channel transmittance map, and fuse the dark channel transmittance t d and the bright channel transmittance t b, get the fused rough transmittance, introduce the gradient guided filter to refine the fused rough transmittance, and get the refined transmittance t;

[0017] S4. Based on the coal mine dust and fog image degradation model, the refined transmittance t and the final atmospheric light value A are used to generate and output a defogging image;

[0018] S5. Perform color correction and CLAHE enhancement on the dehazed image to generate a clear and haze-free image.

[0019] The present invention fully integrates the characteristics of bright channel prior and dark channel prior, and combines key steps such as gamma correction, quadtree atmospheric light estimation, gradient-guided filtering and CLAHE enhancement. On the basis of extracting and fusing image brightness and structure information, it adaptively optimizes transmittance estimation and image enhancement strategy to achieve clear restoration of underground images.

[0020] In a more specific technical solution, in S1, the dark channel map is used to reflect the most opaque area in the image, and the bright channel map is used to compensate for the details of the highlight area.

[0021] In a more specific technical solution, in S2, the following logic is used to perform gamma correction on the bright channel image to reduce the light value deviation:

[0022] L′=L γ ,γ=1.2

[0023] Use the following logic to reduce brightness error:

[0024] A g (x)=(A bright (x)) γ

[0025] Where a bright (x) is the atmospheric light value of the bright channel after filtering, γ is the coefficient of gamma correction transformation, A g (x) is the atmospheric light value of the bright channel obtained after gamma correction.

[0026] In a more specific technical solution, in S2, the atmospheric light value is fused using the following formula:

[0027] A(x)=αA 亮 +(1-α)·A 暗 ,α=0.25.

[0028] In a more specific technical solution, in S2, a quadtree is constructed based on the local blocks of the image, the brightness of each layer of image blocks is sorted, and the atmospheric light value of the high fog area is extracted;

[0029] Gradient-guided filtering is used to refine A(x) to improve image continuity and edge stability.

[0030] This invention significantly improves image brightness issues such as over-brightness and over-enhancement. Specifically, because obtaining the atmospheric light value of the bright channel involves multiple maximum operations, which can result in overly large values, gamma correction is performed on the filtered bright channel image. Because obtaining the atmospheric light value of the dark channel is easily affected by the surrounding environment, large errors can occur in the atmospheric light value, resulting in an overly dark output image. Therefore, a quadtree-based local search for atmospheric light is performed on the filtered dark channel image.

[0031] In a more specific technical solution, in S3, the dark channel transmittance is estimated using the following logic:

[0032]

[0033] The following logic is used to estimate the bright channel transmittance:

[0034]

[0035] Transmittance fusion (adaptive weighting) is performed using the following logic:

[0036] θ=Y / N, Y=number of highlight pixels, N=total number of pixels

[0037] t(x)=θ·t b (x)+(1-θ)·t d (x)

[0038] Gradient-guided filtering is used to refine t(x): an image-guided filtering model is used to minimize the loss function containing an edge-preserving term.

[0039] The present invention improves the performance and accuracy of the algorithm model. Specifically, it estimates model parameters from different perspectives based on dual-channel prior theoretical knowledge and fuses the model parameters based on the characteristics of different channels, improving the process of obtaining atmospheric light and transmittance.

[0040] In a more specific technical solution, gradient-guided filtering is used to refine the transmittance map:

[0041]

[0042] Where, is the output image, G(p) is the guide image, a p′ and b p′ is the linear transformation coefficient.

[0043] Compared with the prior art, the transmittance image of the present invention is more delicate and precise: the transmittance image is refined by using gradient-guided filtering, making the obtained transmittance image more delicate and precise, while being able to effectively suppress the block effect and halo effect.

[0044] In a more specific technical solution, an image-guided filtering model is used to minimize a loss function including an edge preservation term;

[0045] Using the following logic, according to the regularization parameter Edge-aware weight F, defines the minimization loss function:

[0046]

[0047] Use the following logic to find the regularization parameter Edge-aware weight F:

[0048]

[0049] Solve the loss function to get the linear parameter a p′ and b p′ , substitute into the formula The output image.

[0050] In a more specific technical solution, in S4, based on the improved atmospheric scattering model, the following logic is used for restoration:

[0051]

[0052] The present invention effectively improves the quality of the defogging image. Specifically, the defogging image is subjected to color correction and adaptive contrast limited histogram equalization (CLANE) processing, so that the obtained image is not only more complete in terms of details, but also clearer and more natural.

[0053] In a more specific technical solution, a coal mine underground image defogging system includes:

[0054] The dual-channel image construction module is used to perform dual-channel atmospheric light value fusion, perform maximum filtering and minimum filtering on the input haze image, and obtain the bright channel image and dark channel image corresponding to the haze image respectively;

[0055] The dual-channel atmospheric light estimation module is used to apply gamma transformation to the bright channel image, adjust the bright channel image, and obtain the bright channel atmospheric light value A g ,The local atmospheric light is searched by quadtree for the dark channel map ,to obtain the global atmospheric light value of the dark channel. The ,dual-channel atmospheric light estimation module is connected with the dual-channel ,map construction module;

[0056] The dual-channel transmittance estimation fusion module is used to estimate and refine the transmittance map according to the bright channel atmospheric light value A g , dark channel global atmospheric light value, obtain the corresponding bright channel transmittance map, dark channel transmittance map, and fuse the dark channel transmittance t d and the bright channel transmittance t b , get the fused rough transmittance, introduce the gradient guided filter to refine the fused rough transmittance, get the refined transmittance t, and connect the dual-channel transmittance estimation fusion module with the dual-channel atmospheric light estimation module;

[0057] The image defogging and restoration module is used to generate and output a defogging image based on the coal mine dust and fog image degradation model using the refined transmittance t and the final atmospheric light value A. The image defogging and restoration module is connected to the dual-channel transmittance estimation and fusion module;

[0058] The image post-processing module is used to perform color correction and CLAHE enhancement on the defogging image to generate a clear and fog-free image. The image post-processing module is connected to the image defogging and restoration module.

[0059] Compared with the prior art, the present invention has the following advantages:

[0060] The present invention fully integrates the characteristics of bright channel prior and dark channel prior, and combines key steps such as gamma correction, quadtree atmospheric light estimation, gradient-guided filtering and CLAHE enhancement. On the basis of extracting and fusing image brightness and structure information, it adaptively optimizes transmittance estimation and image enhancement strategy to achieve clear restoration of underground images.

[0061] This invention significantly improves image brightness issues such as over-brightness and over-enhancement. Specifically, because obtaining the atmospheric light value of the bright channel involves multiple maximum operations, which can result in overly large values, gamma correction is performed on the filtered bright channel image. Because obtaining the atmospheric light value of the dark channel is easily affected by the surrounding environment, large errors can occur in the atmospheric light value, resulting in an overly dark output image. Therefore, a quadtree-based local search for atmospheric light is performed on the filtered dark channel image.

[0062] The present invention improves the performance and accuracy of the algorithm model. Specifically, it estimates model parameters from different perspectives based on dual-channel prior theoretical knowledge and fuses the model parameters based on the characteristics of different channels, improving the process of obtaining atmospheric light and transmittance.

[0063] Compared with the prior art, the transmittance image of the present invention is more delicate and precise: the transmittance image is refined by using gradient-guided filtering, making the obtained transmittance image more delicate and precise, while being able to effectively suppress the block effect and halo effect.

[0064] The present invention effectively improves the quality of the defogging image. Specifically, the defogging image is subjected to color correction and adaptive contrast limited histogram equalization (CLANE) processing, so that the obtained image is not only more complete in terms of details, but also clearer and more natural.

[0065] The present invention solves the technical problems existing in the prior art, such as color distortion, lack of real data sets, high time complexity, poor algorithm real-time performance and adaptability to complex environments, and image information distortion. BRIEF DESCRIPTION OF THE DRAWINGS

[0066] Figure 1 This is a schematic diagram of the basic steps of a method for defogging an image in an underground coal mine according to embodiment 1 of the present invention;

[0067] Figure 2 This is a schematic diagram of data flow processing of a coal mine underground image defogging method according to embodiment 1 of the present invention;

[0068] Figure 3 Grayscale images under different Gamma values ​​in Example 1 of the present invention;

[0069] Figure 4 This is the atmospheric light effect diagram estimated by the quadtree in Example 1 of the present invention;

[0070] Figure 5 The atmospheric light map of Example 1 of the present invention;

[0071] Figure 6 is a transmittance diagram of Example 1 of the present invention;

[0072] Figure 7 The gradient-guided filtering refined transmittance map of Example 1 of the present invention;

[0073] Figure 8 Comparison between the CLAHE enhanced grayscale image and histogram according to Example 1 of the present invention. DETAILED DESCRIPTION

[0074] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0075] Example 1

[0076] like Figure 1 and Figure 2 As shown, the present invention provides a coal mine underground image defogging method, which includes the following basic steps:

[0077] S1. Perform dual-channel atmospheric light value fusion and perform maximum and minimum filtering operations on the input haze image to obtain the corresponding bright channel image and dark channel image respectively;

[0078] In this embodiment, a 15×15 filter window is used to perform maximum and minimum filtering on the haze image to generate light and dark channel images. The dark channel image reflects the most opaque area in the image, and the light channel image is used to compensate for the details of the highlight area. Step S1 provides a basis for subsequent atmospheric light and transmittance calculations.

[0079] S2. Apply gamma transform to the filtered bright channel image and adjust the bright channel image to obtain the bright channel atmospheric light value A. g The filtered dark channel image is searched for local atmospheric light using a quadtree to obtain the global atmospheric light value of the dark channel.

[0080] like Figure 3 As shown, in this embodiment, the bright channel image is gamma corrected (L′=L γ ,γ=1.2), reduce the brightness error, the formula is: A g (x)=(A bright (x)) γ Where: A bright (x) is the atmospheric light value of the bright channel after filtering; γ is the coefficient of gamma correction transformation; A g (x) is the atmospheric light value in the bright channel after gamma correction. Introducing gamma correction can effectively reduce the error caused by excessive brightness in the bright channel image.

[0081] like Figure 4 As shown, in this embodiment, since the atmospheric light value obtained in the dark channel is easily affected by the surrounding environment, when there are objects with strong reflective characteristics such as light sources and white blank areas in the scene, there will be large errors in the atmospheric light value obtained. Therefore, a quadtree local search method for atmospheric light is used on the filtered dark channel image to calculate the local atmospheric light value of the dark channel and obtain the global atmospheric light value;

[0082] S3, dual-channel atmospheric light value fusion;

[0083] like Figure 5 As shown, in this embodiment, the bright channel atmospheric light value A obtained above is g Linear fusion is performed with the dark channel atmospheric light value A0. The fusion formula is: A(x) = αA g (x)+(1-α)A0. α is a scaling factor (0≤α≤1). Experiments show that setting α to 0.25 can balance global and local information and reduce errors.

[0084] S4, perform transmittance map estimation and refinement;

[0085] like Figure 6 As shown, in this embodiment, in this embodiment, A g Linearly fuse A(x) with A0 and use gradient-guided filtering to perform refinement to obtain the final atmospheric light value A. Use gradient-guided filtering to refine the fused atmospheric light value A(x) to obtain the final atmospheric light value A. Use gradient-guided filtering to refine A(x) to improve image continuity and edge stability.

[0086] In this embodiment, the atmospheric light value is fused using the following formula:

[0087] A(x)=αA 亮 +(1-α)·A 暗 ,α=0.25

[0088] Linear fusion is used to balance the estimation errors of bright and dark channels in different regions.

[0089] S5. Obtain the corresponding transmittance graphs of the bright channel and dark channel according to the atmospheric light values ​​obtained for the dark channel and bright channel, and then calculate the transmittance of the dark channel t d and the bright channel transmittance t b The fusion is performed to obtain the fused rough transmittance, and the gradient guided filter is introduced to further refine the fused rough transmittance to obtain the refined transmittance t;

[0090] In traditional dark channel image processing, the dark channel value of the highlight area is larger, and the transmittance is calculated It is necessary to prevent it from approaching zero, otherwise the brightness of the dark area will be excessively enhanced and noise will be introduced. Therefore, in this embodiment, the transmittance t for dark light and bright light areas is calculated using the dark channel map and the bright channel map respectively. d and t b .

[0091] Specifically, the dark channel transmittance is estimated using the following formula:

[0092]

[0093] The bright channel transmittance is estimated using the following formula:

[0094]

[0095] Because the ratio of pixels in the highlight area is b The influence is small, and the adaptive weight coefficient β is introduced: β = Y / N (Y is the number of highlight pixels with grayscale value greater than 220, and N is the total number of pixels). The fusion transmittance formula is: t(x) = βt b +(1-β)t d This method is adaptive and can better balance the transmittance estimation of highlight and dark areas.

[0096] like Figure 7 As shown, in this embodiment, in order to optimize the transmittance map and reduce halo and block artifacts, the gradient guided filtering method is used to refine the transmittance map. This method has excellent performance in edge preservation and tone mapping. The formula is in is the output image, G(p) is the guide image, a p′ and b p′ is the linear transformation coefficient, which is obtained by minimizing the loss function. The minimization loss function is defined as:

[0097]

[0098] The loss function definition involves regularization parameters and edge-aware weights F, where F is calculated by local variance. F and γ p′ The specific formula is:

[0099]

[0100] The image-guided filtering model is used to minimize the loss function including the edge preservation term, improve the boundary processing effect of the transmittance map, and eliminate halo and block effects.

[0101] Solve the loss function to get the linear parameter a p′ and b p′ , substitute into the formula The output image is obtained. After gradient-guided filtering, the transmittance map is more refined, effectively suppressing blocking and halo effects, and the image is more natural and clear. This method effectively balances processing efficiency and dehazing effect, especially performing well in image edge processing.

[0102] The dark channel prior theory was proposed by He Kaiming et al., which means that in each pixel of a clear image, there is at least one channel with a grayscale value close to zero. Combined with the dust and fog degradation model, the transmittance expression can be derived: The dark channel prior theory provides an effective mathematical tool for removing dust and fog by assuming that natural images have the characteristics of a low-value dark channel when there is no haze interference. However, it does not work well in bright scenes and color distortion is prone to occur.

[0103] The bright channel prior theory is applicable to high-brightness scenes, pointing out that the grayscale value of a certain channel of most pixels in haze images is close to the atmospheric light value, and the atmospheric light value of fog-free images is close to 1. Bright channel value: The transmittance calculation formula is: The effect is significant when processing highlight areas, and it can complement the dark channel prior to optimize image restoration quality. However, when processing images with strong contrast between highlight and low-light areas, the bright channel prior may not be able to accurately distinguish between real details and the influence of scattered light, resulting in the loss of some detail information, especially details at the edges of highlight areas.

[0104] S6. Generate a defogging image using the refined transmittance t and the final atmospheric light value A obtained above according to the coal mine dust and fog image degradation model, and output the defogging image.

[0105] In this embodiment, image dehazing is achieved by estimating the atmospheric light value A(x) and transmittance t(x) based on an atmospheric scattering model, combined with a coal mine dust and fog image degradation model. To avoid whitening and loss of detail due to low transmittance, a minimum transmittance threshold t0 is set to 0.1. The restoration formula is:

[0106]

[0107] S7, performing color correction on the obtained defogging image and processing edge details of the defogging image using the CLAHE (contrast limited adaptive histogram equalization) method, finally generating a clear and fog-free image and outputting the image;

[0108] In this embodiment, the image is converted to LAB space, the L channel is normalized and contrast-limited; CLAHE (contrast-limited adaptive histogram equalization) is applied to enhance image edges and details; and the image is restored to RGB space to obtain a final clear, natural, and haze-free output image.

[0109] In the optional implementation of this embodiment, the filtering module can be switched to guided filtering or bilateral filtering; the fusion method can be replaced by weighted average or maximum entropy criterion; CLAHE can be replaced by histogram matching or Retinex method;

[0110] Each module can be encapsulated as a software component or hardware chip and used in embedded terminals or edge computing systems.

[0111] like Figure 8 As shown in this example, after dehazing, the image may suffer from color distortion due to factors such as lighting conditions and sensor performance, requiring color correction in the LAB color space. The specific steps are: converting the RGB image to LAB space, normalizing the L channel (L / 100), setting a contrast limit threshold to prevent excessive noise enhancement, restoring the enhanced luminance channel to its original range, and then converting the image back to RGB space to ensure the valid pixel value range, ultimately resulting in a high-quality, haze-free image.

[0112] In real life, light is easily affected by other objects and media such as suspended particles in the atmosphere during its propagation, causing it to change direction and form scattering. When dehazing haze images, the atmospheric scattering model proposed by SGNarasimha et al. is often used. The atmospheric scattering model is the basis for image dehazing and is expressed as: I(x) = J(x)t(x) + A(1-t(x)).

[0113] The formula for the coal mine dust and fog image degradation model derived from the atmospheric scattering model and the special environment of underground coal mines is: I(x) = J(x)t(x) + L(x)(1-t(x)), where I(x) is the foggy image, J(x) is the clear image, L(x) is the ambient light value, and t(x) = exp(-β(λ)d(x)). From this formula, we can obtain J(x) = (I(x)-L(x)) / (t(x))+L(x). To avoid noise introduced by excessively low transmittance, a minimum transmittance t0 is introduced, generally set to t0 = 0.1. This gives J(x) = (I(x)-L(x)) / (max(t0,t(x)))+L(x), which is suitable for image restoration in underground coal mines.

[0114] In summary, the present invention fully integrates the characteristics of bright channel prior and dark channel prior, and combines key steps such as gamma correction, quadtree atmospheric light estimation, gradient-guided filtering, and CLAHE enhancement. On the basis of extracting and fusing image brightness and structure information, it adaptively optimizes the transmittance estimation and image enhancement strategy to achieve clear restoration of downhole images.

[0115] This invention significantly improves image brightness issues such as over-brightness and over-enhancement. Specifically, because obtaining the atmospheric light value of the bright channel involves multiple maximum operations, which can result in overly large values, gamma correction is performed on the filtered bright channel image. Because obtaining the atmospheric light value of the dark channel is easily affected by the surrounding environment, large errors can occur in the atmospheric light value, resulting in an overly dark output image. Therefore, a quadtree-based local search for atmospheric light is performed on the filtered dark channel image.

[0116] The present invention improves the performance and accuracy of the algorithm model. Specifically, it estimates model parameters from different perspectives based on dual-channel prior theoretical knowledge and fuses the model parameters based on the characteristics of different channels, improving the process of obtaining atmospheric light and transmittance.

[0117] Compared with the prior art, the transmittance image of the present invention is more delicate and precise: the transmittance image is refined by using gradient-guided filtering, making the obtained transmittance image more delicate and precise, while being able to effectively suppress the block effect and halo effect.

[0118] The present invention effectively improves the quality of the defogging image. Specifically, the defogging image is subjected to color correction and adaptive contrast limited histogram equalization (CLANE) processing, so that the obtained image is not only more complete in terms of details, but also clearer and more natural.

[0119] The present invention solves the technical problems existing in the prior art, such as color distortion, lack of real data sets, high time complexity, poor algorithm real-time performance and adaptability to complex environments, and image information distortion.

[0120] 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 aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A coal mine underground image defogging method, characterized in that: The method comprises: S1. Perform dual-channel atmospheric light value fusion, perform maximum filtering and minimum filtering on the input haze image, and obtain a bright channel image and a dark channel image corresponding to the haze image respectively; S2. Apply gamma transformation to the bright channel image to adjust the bright channel image and obtain the bright channel atmospheric light value A. g , searching the local atmospheric light for the dark channel map using a quadtree to obtain the global atmospheric light value of the dark channel; S3, perform transmittance map estimation and refinement operations, according to the bright channel atmospheric light value A g , the global atmospheric light value of the dark channel, obtain the corresponding bright channel transmittance map, dark channel transmittance map, and fuse the dark channel transmittance t d and bright channel transmittance t b , obtain the fused rough transmittance, introduce the gradient guided filter to refine the fused rough transmittance, and obtain the refined transmittance t; S4. Generate and output a defogging image using the refined transmittance t and the final atmospheric light value A according to a coal mine dust and fog image degradation model; S5. Perform color correction and CLAHE enhancement on the defogging image to generate a clear and fog-free image.

2. The coal mine underground image defogging method according to claim 1, characterized in that: In S1, the dark channel image is used to reflect the most opaque area in the image, and the bright channel image is used to compensate for the details of the highlight area.

3. The coal mine underground image defogging method according to claim 1, characterized in that: In S2, the following logic is used to perform gamma correction on the bright channel image to reduce the light value deviation: L′=L γ ,γ=1.2 Use the following logic to reduce brightness error: A g (x)=(A briht (x)) γ Where A bright (x) is the atmospheric light value of the bright channel after filtering, γ is the coefficient of gamma correction transformation, A g (x) is the atmospheric light value of the bright channel obtained after gamma correction.

4. The coal mine underground image defogging method according to claim 1, characterized in that: In S2, the atmospheric light value is fused using the following formula: A(x)=αA 亮 +(1-a)·A 暗 ,α=0.

25.

5. The coal mine underground image defogging method according to claim 1, characterized in that: In S2, a quadtree is constructed based on the local blocks of the image, the brightness of each layer of image blocks is sorted, and the atmospheric light value of the high fog area is extracted; Gradient-guided filtering is used to refine A(x) to improve image continuity and edge stability.

6. The coal mine underground image defogging method according to claim 1, characterized in that: In S3, the dark channel transmittance is estimated using the following logic: The following logic is used to estimate the bright channel transmittance: Transmittance fusion (adaptive weighting) is performed using the following logic: θ=Y / N, Y=number of highlight pixels, N=total number of pixels t(x)=θ·t b (x)+(1-θ)t d (x) Gradient-guided filtering is used to refine t(x): an image-guided filtering model is used to minimize the loss function containing an edge-preserving term.

7. The coal mine underground image defogging method according to claim 6, characterized in that: The gradient-guided filtering is used to refine the transmittance map: Where, is the output image, G(p) is the guide image, a p′ and b p′ is the linear transformation coefficient.

8. The method for defogging an underground coal mine image according to claim 6, characterized in that: Using an image-guided filtering model, minimize the loss function including edge preservation terms; Using the following logic, according to the regularization parameter Edge-aware weight F, defines the minimization loss function: Using the following logic, the regularization parameter is obtained The edge-aware weight F: Solve the loss function to get the linear parameter a p′ and b p′ , substitute into the formula The output image.

9. The method for defogging an image in an underground coal mine according to claim 1, characterized in that: In S4, based on the improved atmospheric scattering model, the following logic is used for restoration:

10. A coal mine underground image defogging system, characterized in that: The system comprises: A dual-channel image construction module is used to perform dual-channel atmospheric light value fusion, perform maximum filtering and minimum filtering on the input haze image, and obtain the bright channel image and dark channel image corresponding to the haze image respectively; A dual-channel atmospheric light estimation module is used to apply gamma transformation to the bright channel image, adjust the bright channel image, and obtain the bright channel atmospheric light value A g , searching the local atmospheric light for the dark channel map using a quadtree to obtain the global atmospheric light value of the dark channel, and connecting the dual-channel atmospheric light estimation module with the dual-channel map construction module; Dual-channel transmittance estimation fusion module, used to perform transmittance map estimation and refinement operations, according to the bright channel atmospheric light value A g , the global atmospheric light value of the dark channel, obtain the corresponding bright channel transmittance map, dark channel transmittance map, and fuse the dark channel transmittance t d and the bright channel transmittance t b , obtaining a fused rough transmittance, introducing a gradient-guided filter to refine the fused rough transmittance to obtain a refined transmittance t, and connecting the dual-channel transmittance estimation fusion module to the dual-channel atmospheric light estimation module; an image defogging and restoration module, configured to generate and output a defogging image based on a coal mine dust and fog image degradation model using the refined transmittance t and the final atmospheric light value A, the image defogging and restoration module being connected to the dual-channel transmittance estimation and fusion module; The image post-processing module is used to perform color correction and CLAHE enhancement on the defogging image to generate a clear and fog-free image. The image post-processing module is connected to the image defogging and restoration module.

Images