Coal mine image defogging method and system

Abstract

The invention discloses a coal mine image defogging method and system, and the method comprises the following steps: collecting an underground coal mine dust fog image signal through a mining camera, and measuring the dust concentration data of an underground environment through a dust sensor; respectively carrying out dark channel prior processing and bright channel prior processing on the dust fog image signal to obtain a dark channel value and a bright channel value; calculating a dust concentration compensation item based on the dust concentration data, and calculating a comparison characteristic parameter and an adaptive proportionality coefficient based on a global characteristic linear difference value of a dark channel value and a bright channel value; using the adaptive proportionality coefficient as a dynamic selection proportion value of the brightest pixel point in ambient light estimation, and performing ambient light estimation in combination with a dust concentration compensation item to obtain an ambient light value; based on the ambient light value, the dark channel value and the bright channel value, transmissivity estimation is carried out in a weighted fusion mode, and a transmissivity map is obtained; and performing image restoration through an atmospheric scattering model by using the optimized transmissivity image and the ambient light value to obtain a defogged clear image.

Description

Technical Field

[0001] This invention belongs to the field of coal mine image dehazing technology, specifically relating to a coal mine image dehazing method and system based on sensor complementarity and adaptive dual-channel fusion. Background Technology

[0002] Intelligent construction in coal mines is a core support for the high-quality development of the coal industry, with computer vision technology playing a crucial role in areas such as underground safety monitoring, equipment status identification, and personnel positioning. However, the underground environment of coal mines is extremely complex: the space is enclosed, lighting relies on artificial light sources and is unevenly distributed, and suspended particulate matter such as dust and water vapor generated during production causes severe scattering and absorption effects on light, resulting in significant "dust and fog" degradation in the acquired images, manifested as image blurring, decreased contrast, color distortion, and loss of detail. This severe image quality deterioration significantly restricts the accuracy of subsequent intelligent analysis algorithms and poses a potential threat to coal mine safety. Therefore, developing image defogging technology that can effectively address the low-light, high-dust environment of underground coal mines is of great significance.

[0003] Currently, image dehazing techniques are mainly divided into three categories: image enhancement methods, deep learning-based methods, and physical model-based methods. Image enhancement methods (such as histogram equalization and homomorphic filtering) improve contrast by directly adjusting pixel distribution, but they ignore the physical causes of image degradation, easily leading to loss of local details or over-enhancement, and have limited effectiveness in underground scenes with uneven dust and fog concentrations. Deep learning-based methods (such as DehazeNet and AOD-Net) rely on training with a large amount of paired data and perform well in dehazing natural images, but it is difficult to obtain real fog-free image data in coal mines, and models trained on synthetic data have insufficient generalization ability. In addition, the models have poor interpretability and high computational complexity, making them difficult to deploy in real time in underground embedded devices. Physical model-based methods have better interpretability, especially atmospheric scattering models, among which Dark Channel Prior (DCP) and Bright Channel Prior (BCP) are classic techniques. DCP is based on statistical observation: in non-sky regions of haze-free images, at least one color channel has extremely low pixel values. However, this method fails when processing coal mine images in bright areas (such as areas directly illuminated by light sources), resulting in an overall darker image after dehazing. Furthermore, inaccurate transmittance estimation can easily produce halo artifacts. BCP focuses on bright areas, which can compensate for the shortcomings of DCP, but it is prone to losing details in dark areas. For example, while the prior art application CN116934621A introduces a strategy for segmenting and fusing bright and dark channels, its fusion weights rely on simple threshold segmentation, which easily produces edge effects. Furthermore, it lacks the ability to adapt to environmental changes because it does not incorporate dust sensor data. Secondly, the prior art application CN116012236A optimizes transmittance through dual-window fitting, but it remains limited to a single prior framework and cannot solve the problem of dark images under low illumination. Additionally, the prior art application CN111402158A uses nonlinear transformation to enhance brightness, but its transmittance estimation method is simple and does not achieve depth-adaptive fusion of bright and dark channels. Existing technologies generally suffer from problems such as strong randomness in parameter estimation, insufficient environmental adaptability, and a lack of multi-source information complementary verification mechanisms, leading to unstable dehazing effects and failing to meet the robustness requirements for high-quality images in intelligent coal mine applications.

[0004] Existing physical model-based dehazing methods have revealed a series of limitations in underground coal mine applications. First, the Dark Channel Prior (DCP) is prone to failure in low-light underground environments: due to uneven illumination, the local minimum pixel value upon which DCP relies may be underestimated in dark areas, leading to overestimation of transmittance and resulting in color shifts or distortion in the dehazed image. Conversely, in bright areas (such as equipment light sources), the DCP assumption does not hold, causing loss of detail. The invention document with publication number CN116012236A attempts to optimize atmospheric light estimation through morphological operations and gamma correction, but it does not overcome the inherent limitations of DCP, and its transmittance fusion still relies on fixed parameters, making it difficult to adapt to dynamic changes in underground dust and fog concentration. Second, while the Bright Channel Prior (BCP) can improve the processing of bright areas, when applied alone in low-light grayscale images, it tends to over-amplify noise, leading to texture blurring. For example, the invention document with publication number CN116934621A discloses a fusion method based on dark and bright channel segmentation, but its weight calculation is only achieved through threshold segmentation, lacking adaptive discrimination of image color features, and is prone to producing discontinuous artifacts in dust and fog transition areas. Furthermore, existing methods largely rely on the image signal itself and do not introduce a complementary mechanism for multi-sensor data in underground mining: dust concentration fluctuates greatly in underground coal mines, and image features alone cannot accurately reflect the real environment; for example, when the lens is damaged or the sensor is malfunctioning, the algorithm cannot self-check. Although the invention document with publication number CN111402158A mentions brightness enhancement, it does not integrate dust sensor data, leading to inaccurate ambient light estimation when dust concentration changes abruptly. These problems result in existing defogging methods producing images with uneven brightness and low color fidelity under complex underground conditions, making it difficult to support high-precision visual analysis.

[0005] Further analysis of the aforementioned prior art reveals three main shortcomings: First, transmittance estimation lacks an adaptive fusion mechanism. While DCP and BCP each have their advantages and disadvantages, simple linear fusion (such as the fixed-weight segmentation disclosed in the invention document with publication number CN116934621A) cannot adapt to the differences in downhole image types (color images versus low-light grayscale images). For example, when an image is classified as a low-light grayscale image, BCP tends to amplify noise, while DCP ignores details. Existing methods do not dynamically adjust the fusion strategy based on features such as the hue histogram. Second, ambient light estimation relies on empirical selection, such as the fixed selection of the brightest pixel ratio described in the invention document with publication number CN116012236A, which does not consider physical compensation for dust concentration, causing the estimated value to deviate from the true value during peak dust concentration periods. Third, the post-processing stage is weak, and dehazed images often suffer from insufficient brightness or saturation distortion. The invention document in application publication number CN111402158A uses histogram equalization to enhance brightness, but it is not optimized in conjunction with transmittance estimation, easily leading to local overexposure. In summary, existing methods fail to deeply integrate dust sensor data with visual features and lack adaptive parameter adjustments for coal mine image types, resulting in insufficient robustness of dehazing effects in low-light, high-dust underground environments. Therefore, overcoming the limitations of existing pure visual algorithms, deeply integrating multi-source heterogeneous sensor information from underground sources, and constructing a robust dehazing model capable of sensing environmental physical states and possessing self-verification and adaptive capabilities is key to solving the problem of image sharpening in underground coal mines. Summary of the Invention

[0006] This invention addresses the aforementioned problems and overcomes the shortcomings of existing technologies by providing a coal mine image dehazing method and system based on sensor complementarity and adaptive dual-channel fusion.

[0007] To achieve the above objectives, the present invention adopts the following technical solution.

[0008] A method for dehazing coal mine images includes the following steps: S1. Real-time acquisition of dust and fog image signals in underground coal mines using mining cameras, and real-time measurement of dust concentration data in the underground environment using dust sensors; S2. Perform dark channel prior processing and bright channel prior processing on the dust and fog image signal respectively to obtain dark channel value and bright channel value; S3. Calculate the dust concentration compensation term based on the dust concentration data, and calculate the comparison feature parameter and adaptive scaling factor based on the global feature linear difference between the dark channel value and the bright channel value; S4. Use the adaptive scaling factor as the dynamic selection scaling factor of the brightest pixel in the ambient light estimation, and combine it with the dust concentration compensation term to perform ambient light estimation and obtain the ambient light value; S5. Based on the ambient light value, dark channel value, and bright channel value, transmittance is estimated by weighted fusion to obtain a transmittance map; S6. Perform weighted guided filtering optimization on the transmittance map to enhance image details; S7. Using the optimized transmittance map and ambient light value, image restoration is performed through an atmospheric scattering model to obtain a clear image after dehazing.

[0009] In a preferred embodiment of the present invention, in step S2, the dark channel prior processing is expressed as follows: The prior processing of the bright channel is expressed as follows: ; in, x Indicates the current pixel position. y Indicates the pixel position within the domain. c Indicates color channels, R , G , B These respectively represent the red channel, green channel, and blue channel; Oh ( x ) indicates with x The center's neighborhood window, J c ( y ) represents an image J The pixel value of a certain color channel.

[0010] As another preferred embodiment of the present invention, in step S3, calculating the comparison feature parameters and the adaptive scaling coefficient includes: First, the dark channel values ​​and bright channel values ​​are normalized to obtain the normalized dark channel mean. m dark and normalized bright channel mean m bright ; Then, the contrast feature parameters are calculated. C = m dark - m bright ; Finally, calculate the adaptive scaling factor. Coef =( C + or ) / R ,in or This is a dust concentration compensation item. R This is a scaling parameter, and its value is dynamically adjusted based on the image's color characteristics. Step S3 also includes a sensor state verification step: verifying the calculated comparison feature parameters. CA consistency comparison is performed with the measured dust concentration data. If the comparison characteristic parameters are consistent... C If the dust concentration level indicated by the processed dust concentration data differs from the measured dust concentration data, the sensor is deemed malfunctioning.

[0011] As another preferred embodiment of the present invention, the proportional adjustment parameter R The value is dynamically adjusted based on the image's color characteristics: when the image is determined to be a color image, R The value range is [1, 30]; when the image is identified as a low-light grayscale image, R The value range is [40, 80]; the dust concentration compensation term or The value range is [0, 1].

[0012] As another preferred embodiment of the present invention, the discrimination of the image color features is achieved through a hue histogram: the image is converted to the HSV color space, the hue channels are extracted and the hue histogram is calculated; the hue histogram is normalized and its maximum value M is calculated; The maximum value M is compared with the preset threshold T. h Comparison: If M <T h If M ≥ T, then it is classified as a color image; h If the image is low-light grayscale, it is determined to be a low-light grayscale image; wherein, the preset threshold T h The value range is [0.95, 1.0].

[0013] In another preferred embodiment of the present invention, in step S4, the ambient light estimation is achieved by a dual-channel weighted fusion mean method: the number of candidate pixels is determined based on the adaptive scaling coefficient, and a joint feature value is calculated by combining the dark channel value and the bright channel value. ,in, α These are the weighting coefficients. B ( x () represents the brightness channel value. D ( x () represents the dark channel value; Select the top with the largest joint eigenvalue N s 1 pixel, of which N s = Coef × N , N The total number of pixels in the image; calculate the... N s The average RGB value of each pixel at its corresponding position in the original dust and fog image is used as the ambient light value.

[0014] As another preferred embodiment of the present invention, in step S5, the transmittance estimation is expressed as: ;in, I c ( y (This refers to a foggy image in the channel) c pixel values, L c ( x Ambient light value, β The weighting coefficients are derived from the comparative feature parameters. C Obtained through constraint transformation.

[0015] As another preferred embodiment of the present invention, in step S6, the weighted guided filtering employs adaptive regularization weights: for the guided image I and input image p Local eigenvalues ​​are calculated using a dual-window strategy, and a weight adjustment parameter is introduced. c Constructing Adaptive Regularized Weights W I This method reduces the smoothing intensity in areas with rich texture and increases the smoothing intensity in areas with poor texture, thereby preserving edge details.

[0016] As another preferred embodiment of the present invention, in step S7, the image restoration is based on an improved atmospheric scattering model: ;in, J ( x The image is the restored image. I ( x (This is a foggy image.) L ( x () represents the ambient light value. t ( x () represents transmittance. t 0 is the minimum transmittance threshold.

[0017] In addition, the present invention provides a coal mine image defogging system for implementing the aforementioned coal mine image defogging method. The coal mine image defogging system includes an image acquisition module, a dust measurement module, a dual-channel processing module, a feature calculation module, an ambient light estimation module, a transmittance estimation module, a filtering optimization module, and an image restoration module. The image acquisition module is used to acquire real-time image signals of dust and fog in underground coal mines using a mining camera; The dust measurement module is used to measure the dust concentration data of the downhole environment in real time through a dust sensor; The dual-channel processing module is used to perform dark channel prior processing and bright channel prior processing on the dust and fog image signal respectively to obtain dark channel value and bright channel value; The feature calculation module is used to calculate a dust concentration compensation term based on the dust concentration data, and to calculate a comparison feature parameter and an adaptive scaling factor based on the global feature linear difference between the dark channel value and the bright channel value. The ambient light estimation module is used to use the adaptive scaling factor as the dynamic selection scaling factor of the brightest pixel in the ambient light estimation, and to combine it with the dust concentration compensation term to perform ambient light estimation and obtain the ambient light value. The transmittance estimation module is used to estimate transmittance based on the ambient light value, dark channel value and bright channel value through a weighted fusion method to obtain a transmittance map. The filtering optimization module is used to perform weighted guided filtering optimization on the transmittance map to enhance image details; The image restoration module is used to restore the image using an optimized transmittance map and ambient light value through an atmospheric scattering model, thereby obtaining a clear image after defogging.

[0018] Compared with the prior art, the beneficial effects of the present invention are as follows: The coal mine image dehazing method and system provided by this invention achieves significant advantages compared to existing technologies: First, this invention introduces a heterogeneous data complementarity strategy between dust sensors and mining cameras, using physically measured dust concentration data as a compensation term for ambient light estimation, and comparing it with image visual features (i.e., contrast feature parameters). C By performing consistency comparisons, the system can automatically identify abnormal sensor states, greatly improving its robustness and environmental adaptability under complex downhole conditions. Secondly, this invention proposes an image color feature discrimination mechanism based on hue histograms, which can intelligently distinguish between color images and low-light grayscale images, and dynamically adjust the scaling parameters accordingly. R The range of values ​​enables differentiated and refined processing of different types of images, improving the color fidelity and naturalness after dehazing from the source. This invention designs a complete adaptive fusion framework for bright and dark channels. By calculating the linear difference between the global features of the dark and bright channels, comparative feature parameters are obtained, and then adaptive scaling coefficients for ambient light estimation are dynamically generated. Coef This replaces the traditional empirical approach of selecting the brightest pixel at a fixed ratio, making ambient light estimation more accurate, while also introducing weighting coefficients. β By incorporating the weighted features of the bright and dark channels into the transmittance estimation formula, the detail recovery of both bright and dark areas of the image is effectively taken into account, overcoming the limitations of a single prior. Finally, this invention employs an improved weighted guided filter to optimize the transmittance map. By introducing adaptive regularization weights, edge details are preserved in textured areas, and smoothing effects are enhanced in flat areas, further suppressing halo artifacts and enhancing image details. Using the coal mine image dehazing method and system of this invention, high-quality underground coal mine images with appropriate brightness, clear details, and natural colors can be recovered, providing reliable technical support for various visual applications in coal mine safety production. Attached Figure Description

[0019] Figure 1 This is a schematic flowchart of a coal mine image dehazing method according to the present invention.

[0020] Figure 2 This is a schematic block diagram of the structure of a coal mine image defogging system according to the present invention.

[0021] Figure 3 This image shows a comparison of the visual effects of using the coal mine image defogging method of the present invention with those of using five existing image processing methods to process dust and fog images.

[0022] The diagram is labeled as follows: 1. Image acquisition module; 2. Dual-channel processing module; 3. Dust measurement module; 4. Feature calculation module; 5. Ambient light estimation module; 6. Transmittance estimation module; 7. Filtering optimization module; 8. Image restoration module. Detailed Implementation

[0023] To make the technical problems solved, the technical solutions, and the beneficial effects of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0024] Please see Figure 1 The present invention provides a method for dehazing coal mine images, comprising the following steps: S1. Real-time acquisition of dust and fog image signals in underground coal mines using mining cameras, and real-time measurement of dust concentration data in the underground environment using dust sensors; S2. Perform dark channel prior processing and bright channel prior processing on the dust and fog image signal respectively to obtain dark channel value and bright channel value; S3. Calculate the dust concentration compensation term based on the dust concentration data, and calculate the comparison feature parameter and adaptive scaling factor based on the global feature linear difference between the dark channel value and the bright channel value; S4. Use the adaptive scaling factor as the dynamic selection scaling factor of the brightest pixel in the ambient light estimation, and combine it with the dust concentration compensation term to perform ambient light estimation and obtain the ambient light value; S5. Based on the ambient light value, dark channel value, and bright channel value, transmittance is estimated by weighted fusion to obtain a transmittance map; S6. Perform weighted guided filtering optimization on the transmittance map to enhance image details; S7. Using the optimized transmittance map and ambient light value, image restoration is performed through an atmospheric scattering model to obtain a clear image after dehazing.

[0025] Specifically, in step S2, the dark channel prior processing is expressed as follows: The prior processing of the bright channel is expressed as follows: ;in, x Indicates the current pixel position. y Indicates the pixel position within the domain. c Indicates color channels, R , G , B These respectively represent the red channel, green channel, and blue channel; Oh ( x ) indicates with x The center's neighborhood window, J c ( y ) represents an image J In the complex environment of a coal mine, the dark channel prior and the bright channel prior exhibit different characteristics when processing images of different colors. Therefore, channel fusion of the dark channel prior and the bright channel prior can improve both the texture of grayscale images and the details of color images, resulting in better processing of dust and fog images.

[0026] Specifically, in step S3, calculating the contrast feature parameters and adaptive scaling coefficient includes: first, normalizing the dark channel values ​​and bright channel values ​​to obtain the normalized dark channel mean. m dark and normalized bright channel mean m bright Then, the contrast feature parameters are calculated. C = m dark - m bright Finally, calculate the adaptive scaling factor. Coef =( C + or ) / R ,in or This is a dust concentration compensation item. R The parameter is a proportional adjustment parameter; step S3 also includes a sensor state verification step: the calculated comparison feature parameter is used as the sensor state verification parameter. CA consistency comparison is performed with the measured dust concentration data. If the comparison characteristic parameters are consistent... C If the dust concentration level indicated by the processed dust concentration data differs from the measured dust concentration data, the sensor is deemed malfunctioning. The proportional adjustment parameter... R The value is dynamically adjusted based on the image's color characteristics: when the image is determined to be a color image, R The value range is [1, 30]; when the image is identified as a low-light grayscale image, R The value range is [40, 80]; the dust concentration compensation term or The value range is [0, 1].

[0027] Dust concentration compensation item or The value is linearly proportional to the dust concentration value, using comparative characteristic parameters. C The verification process for the measured dust concentration data: When the lens of the mining camera is relatively clean, the acquired images can describe the actual underground environment. C The value reflects the concentration of dust in the environment. C When the value is low, the dust concentration is high, and the measurement value of the dust sensor should also be high. C When the value is large, the dust concentration is low, and the measurement of the dust sensor should also be low. All of the above are normal situations, and the converted dust concentration measurement value will be used as the dust concentration compensation term. or However, when the visual image and the dust sensor measurement values ​​are inconsistent, it is considered an anomaly. This is mainly manifested in the following ways: an unclear image with a low dust concentration measurement may be due to poor lens cleanliness of the mining camera; conversely, a clear image with a high dust concentration measurement may be caused by an improper installation location of the dust sensor. By fusing heterogeneous data from the mining camera and dust sensor, anomalies can be effectively eliminated, ensuring the accuracy of ambient light estimation.

[0028] The image color feature discrimination is achieved through hue histogram: the image is converted to the HSV color space, the hue channels are extracted and the hue histogram is calculated; the hue histogram is normalized and its maximum value M is calculated; Specifically, this invention uses hue histogram to identify image color features. First, the input image is converted to the HSV color space, the hue channels are segmented, and then converted into a one-dimensional vector. h According to the formula: The hue histogram was calculated. history [ k ]; in the formula, k This represents the number of groups in the histogram, with a value range of [0, 29]. N The hue histogram is normalized to represent the total number of pixels in the image, which is the product of the image height and width. history norm And its maximum value M is obtained statistically. The formula for calculating the maximum value M is as follows: , The maximum value M is compared with the preset threshold T. h Comparison: If M <T h If M ≥ T, then it is classified as a color image; h If the image is low-light grayscale, it is determined to be a low-light grayscale image; wherein, the preset threshold T h The value range is [0.95, 1.0]. Within this range, the preset threshold T is... h The preferred setting is 0.98.

[0029] Specifically, in step S4, the ambient light estimation is achieved through a dual-channel weighted fusion mean method: the number of candidate pixels is determined based on the adaptive scaling coefficient, and a joint feature value is calculated by combining the dark channel value and the bright channel value. ,in, α This is the weighting coefficient, and its value ranges from 0 to... α <1, at this point, both the brighter areas of the dark channel and the bright channel receive attention, thereby reducing the dependence on features of a single channel. In this invention... α The preferred value is 0.8; B ( x () represents the brightness channel value. D ( x () represents the dark channel value; select the value with the largest joint eigenvalue. N s 1 pixel, of which N s = Coef × N , N The total number of pixels in the image; calculate the... N s The average RGB value of each pixel at its corresponding position in the original dust and fog image is used as the ambient light value.

[0030] Specifically, in step S5, the transmittance estimate is expressed as: ;in, I c ( y (This refers to a foggy image in the channel) c pixel values, L c ( x Ambient light level; β The weighting coefficients are derived from the comparative feature parameters. C Obtained by constraint transformation, and compared with characteristic parameters C Weighting coefficients β Multiples of.

[0031] Specifically, in step S6, the weighted guided filtering employs adaptive regularized weights: for the guided image I and input image p Local eigenvalues ​​are calculated using a dual-window strategy, and a weight adjustment parameter is introduced. c Constructing Adaptive Regularized Weights W I This method reduces the smoothing intensity in areas with rich texture and increases the smoothing intensity in areas with poor texture, thereby preserving edge details.

[0032] The specific implementation process of the weighted guided filtering is as follows: For the guided image... I and input image p A dual-window strategy is adopted. First, mean filtering is used to perform small-window convolution, and then the guiding image... I The total number of pixels is N When calculating its local mean. I Guide image I The result of the squared mean filter is corr1 I and guide image I Local variance var1 I Adaptive regularized weights are constructed using the normalized statistic S and the weight adjustment parameter γ. W I Then the execution radius is r Large window mean filtering operation to calculate the guiding image I and input image p Local feature values, including the guide image I Local mean I Input image p Local mean p Guide image I The result of the squared mean filter is corr2 I Guide image I and input image p mean-filtered result of the product, corr Ip Guide image I Local variance var2 I Guide image I and input image p covariance Ip In regularization parameters e Linear coefficients a and b In the calculation process, adaptive regularization weights are introduced. W I When adjusting the weight W I In some cases, the regularization intensity can be reduced in areas with rich texture. e / W I To preserve edge details, while in areas with less texture, the regularization intensity is increased to promote smoothness. Further adjustments are made to the linear coefficients. a and b Perform mean filtering separately to obtain the mean a and mean b Finally, a linear transformation is performed to obtain the output image. q The algorithm flow of weighted guided filtering is shown in Table 1 below.

[0033] Table 1. Algorithm flow of weighted guided filtering

[0034] As can be seen from the above, the present invention uses weighted guided filtering with adaptive regularization weights to process images, which can optimize the transmittance map, not only ensuring a good dehazing effect, but also enhancing image edges.

[0035] Specifically, in step S7, the image restoration is based on an improved atmospheric scattering model: ;in, J ( x The image shown is the restored image, i.e., the image after dehazing and enhancement. I ( x (This is a foggy image.) L ( x () represents the ambient light value. t ( x () represents transmittance. t 0 represents the minimum transmittance threshold. In this invention, the minimum transmittance threshold... t The preferred value for 0 is 0.1.

[0036] The following specific experiments verify the effectiveness of this invention: Considering the complexity of the underground coal mine environment, this invention selects typical dust and fog scenes for experimental analysis. The coal mine image defogging method proposed in this invention has been tested and compared experimentally. For example, in the coal mining face scene, under the influence of factors such as uneven light distribution, limited illumination, dark environment, and dust and fog, it is necessary to restore the outline and details of targets such as coal and rock structures, coal mining machines, scraper conveyors, and hydraulic supports; Figure 3 The image shown is a comparison of the visual effects of using the coal mine image dehazing method of the present invention and using five existing image processing methods to process dust and fog images. The five existing image processing methods are: DCP method (dark channel prior method), BCP method (bright channel prior method), ABiCP method (adaptive dual-channel prior method), TCSF method (dark and bright channel segmentation and fusion method), and TSDM method (improved coal mine dust image dehazing method). Figure 3In the images, (a) and (b) are color images of light and dense fog under sufficient lighting, while (c) and (d) are grayscale images of light and dense fog under insufficient lighting. Using the DCP method, some color distortion occurs in local areas, particularly with over-enhancement in dark areas, resulting in an overall dark image. The BCP method exhibits significant color distortion in color images, and also over-enhancement in dark areas when processing grayscale images. The ABiCP method also shows significant color distortion, with noticeable dust and fog residue in the top area and loss of detail in dark areas. The TCSF method has good strong light suppression, but over-enhancement in dark areas causes the image to lose its natural lighting effect. The TSDM method also exhibits over-enhancement in some dark areas.

[0037] This invention introduces Peak Signal-to-Noise Ratio (PSNR) and Structural Similarity Index Measurement (SSIM) image quality assessment metrics for comparative analysis, providing objective evaluation results. Both metrics reflect changes in illumination, color, and structural features between compared images. PSNR reflects the enhancement effect and color distortion of the image, while SSIM assesses the brightness characteristics, reflecting the rationality of the brightness distribution and the level of brightness enhancement. Higher values ​​for both metrics indicate better image enhancement. The coal mine image defogging method proposed in this invention exhibits relatively stable performance, effectively removing dust and fog while maintaining a natural image effect without significant artifacts or distortion. Calculations based on the quantitative metrics in Table 2 show that the PSNR and SSIM values ​​(higher values ​​indicate better image restoration quality) of this invention's coal mine image defogging method are on average 21.25% and 8.84% higher than the other five methods, respectively. Therefore, in coal mining face scenarios, the image restoration effect of this invention's coal mine image defogging method is quite ideal.

[0038] Table 2 Comparison of objective quantitative indicators of different methods

[0039] In the experimental description of this invention, the coal mine image dehazing method proposed in this invention is compared with the above five existing methods: DCP (dark channel prior method), BCP (bright channel prior method), ABiCP (adaptive dual-channel prior method), TCSF (dark and bright channel segmentation and fusion method), and TSDM (improved coal mine dust image dehazing method). PSNR (peak signal-to-noise ratio) and SSIM (structural similarity index) are used as objective evaluation criteria to demonstrate the comprehensive advantages of the dehazing method of this invention in restoring image details, maintaining color naturalness, and improving overall visual quality.

[0040] On the other hand, please see Figure 1 This invention provides a coal mine image defogging system for implementing the aforementioned coal mine image defogging method. The coal mine image defogging system includes an image acquisition module 1, a dust measurement module 3, a dual-channel processing module 2, a feature calculation module 4, an ambient light estimation module 5, a transmittance estimation module 6, a filtering optimization module 7, and an image restoration module 8. The image acquisition module 1 is connected to the feature calculation module 4 after processing by the dual-channel processing module 2. The dust measurement module 3 is directly connected to the feature calculation module 4. The feature calculation module 4 is connected to the ambient light estimation module 5, the ambient light estimation module 5 is connected to the transmittance estimation module 6, the transmittance estimation module 6 is connected to the filtering optimization module 7, and the filtering optimization module 7 is connected to the image restoration module 8. The image acquisition module 1 is used to acquire real-time dust and fog image signals from underground coal mines using a mining camera. The dust measurement module 3 is used to measure real-time dust concentration data from the underground environment using a dust sensor. The dual-channel processing module... 2. The module performs dark channel prior processing and bright channel prior processing on the dust and fog image signal to obtain dark channel values ​​and bright channel values ​​respectively; the feature calculation module 4 calculates a dust concentration compensation term based on the dust concentration data, and calculates a contrast feature parameter and an adaptive scaling factor based on the global feature linear difference between the dark channel values ​​and bright channel values; the ambient light estimation module 5 uses the adaptive scaling factor as the dynamic selection ratio value of the brightest pixel in the ambient light estimation, and combines it with the dust concentration compensation term to perform ambient light estimation to obtain an ambient light value; the transmittance estimation module 6 performs transmittance estimation based on the ambient light value, dark channel value, and bright channel value through a weighted fusion method to obtain a transmittance map; the filtering optimization module 7 performs weighted guided filtering optimization on the transmittance map to enhance image details; the image restoration module 8 uses the optimized transmittance map and ambient light value to perform image restoration through an atmospheric scattering model to obtain a clear image after defogging.

[0041] The working principle of the coal mine image defogging system of this invention is as follows: This system uses a mining camera to collect coal mine image information, and a dust sensor to measure the dust concentration data of the underground environment. For dust and fog image signals, the image first undergoes prior processing of the bright channel and dark channel using a dual-channel processing module 2. The linear difference of the global features of the two channels is used to obtain comparative feature parameters, and an adaptive scaling factor is calculated. The adaptive scaling factor is used to dynamically select the scaling value of the brightest pixel in the ambient light estimation, and the dust concentration value is used as a compensation term for the ambient light estimation. By comparing and analyzing the comparative feature values ​​with the dust concentration values, sensor anomalies can be determined. The hue histogram is used as the basis for judging the image color features. Different processing methods are used for color images and low-light grayscale images. A weighting strategy is introduced to effectively fuse the bright and dark channels, performing weighted ambient light estimation and weighted transmittance estimation respectively. Weighted guided filtering is used to enhance image details, and finally, the image restoration operation is completed to obtain a clear image after defogging.

[0042] The working process of the coal mine image defogging system of this invention is as follows: Under the premise of ensuring that the required equipment of the system is installed correctly, the system uses a mine camera and a dust sensor to complement each other to identify sensor abnormalities. When the dust concentration is low and has no significant impact on image quality, no defogging is required. When the dust concentration increases and affects the image quality, defogging operation is required. It can effectively process color images under good illumination and grayscale images under low illumination, and finally restore a clear image.

[0043] In summary, the coal mine image defogging method and system provided by this invention introduces a heterogeneous data complementarity strategy between dust sensors and mining cameras, using physically measured dust concentration data as a compensation term for ambient light estimation. It also intelligently identifies the image type (color or low-light grayscale image) using the image hue histogram, dynamically adjusting the brightness and darkness channel fusion parameters to achieve adaptive optimization of transmittance estimation and ambient light estimation. Furthermore, it utilizes weighted guided filtering to enhance transmittance details and incorporates brightness equalization and saturation correction in the image restoration process. Ultimately, it effectively overcomes problems such as image darkening, color distortion, and detail loss in existing technologies, significantly improving the brightness, contrast, and color fidelity of defogging images. This demonstrates stronger robustness and practicality in the complex low-light, high-dust environment of underground coal mines.

[0044] It is understood that, although embodiments of the invention have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for dehazing coal mine images, characterized in that, Includes the following steps: S1. Real-time acquisition of dust and fog image signals in underground coal mines using mining cameras, and real-time measurement of dust concentration data in the underground environment using dust sensors; S2. Perform dark channel prior processing and bright channel prior processing on the dust and fog image signal respectively to obtain dark channel value and bright channel value; S3. Calculate the dust concentration compensation term based on the dust concentration data, and calculate the comparison feature parameter and adaptive scaling factor based on the global feature linear difference between the dark channel value and the bright channel value; S4. Use the adaptive scaling factor as the dynamic selection scaling factor of the brightest pixel in the ambient light estimation, and combine it with the dust concentration compensation term to perform ambient light estimation and obtain the ambient light value; S5. Based on the ambient light value, dark channel value, and bright channel value, transmittance is estimated by weighted fusion to obtain a transmittance map; S6. Perform weighted guided filtering optimization on the transmittance map to enhance image details; S7. Using the optimized transmittance map and ambient light value, image restoration is performed through an atmospheric scattering model to obtain a clear image after dehazing.

2. The method for dehazing coal mine images according to claim 1, characterized in that, In step S2, the dark channel prior processing is expressed as follows: The prior processing of the bright channel is expressed as follows: ; in, x Indicates the current pixel position. y Indicates the pixel position within the domain. c Indicates color channels, R , G , B These respectively represent the red channel, green channel, and blue channel; Ω ( x ) indicates with x The center's neighborhood window, J c ( y ) represents an image J The pixel value of a certain color channel.

3. The method for dehazing coal mine images according to claim 1, characterized in that, In step S3, calculating the contrast feature parameters and adaptive scaling coefficients includes: First, the dark channel values ​​and bright channel values ​​are normalized to obtain the normalized dark channel mean. μ dark and normalized bright channel mean μ bright ; Then, the contrast feature parameters are calculated. C = μ dark - μ bright ; Finally, calculate the adaptive scaling factor. Coef =( C + η ) / R ,in η This is a dust concentration compensation item. R This is a scaling parameter, and its value is dynamically adjusted based on the image's color characteristics. Step S3 also includes a sensor state verification step: verifying the calculated comparison feature parameters. C A consistency comparison is performed with the measured dust concentration data. If the comparison characteristic parameters are consistent... C If the dust concentration level indicated by the processed dust concentration data differs from the measured dust concentration data, the sensor is deemed malfunctioning.

4. The method for dehazing coal mine images according to claim 3, characterized in that, The proportional adjustment parameter R The value is dynamically adjusted based on the image's color characteristics: when the image is determined to be a color image, R The value range is [1, 30]; when the image is identified as a low-light grayscale image, R The value range is [40, 80]; the dust concentration compensation term η The value range is [0, 1].

5. The method for dehazing coal mine images according to claim 4, characterized in that, The image color features are determined by using a hue histogram: the image is converted to the HSV color space, the hue channels are extracted and the hue histogram is calculated; the hue histogram is normalized and its maximum value M is calculated. The maximum value M is compared with the preset threshold T. h Comparison: If M < T h If M ≥ T, then it is classified as a color image; h If the image is low-light grayscale, it is determined to be a low-light grayscale image; wherein, the preset threshold T h The value range is [0.95, 1.0].

6. The method for dehazing coal mine images according to claim 1, characterized in that, In step S4, the ambient light estimation is achieved through a dual-channel weighted fusion mean method: the number of candidate pixels is determined based on the adaptive scaling coefficient, and a joint feature value is calculated by combining the dark channel value and the bright channel value. ,in, α These are the weighting coefficients. B ( x () represents the brightness channel value. D ( x () represents the dark channel value; Select the first one with the largest joint eigenvalue N s 1 pixel, of which N s = Coef × N , N The total number of pixels in the image; calculate the... N s The average RGB value of each pixel at its corresponding position in the original dust and fog image is used as the ambient light value.

7. The method for dehazing coal mine images according to claim 1, characterized in that, In step S5, the transmittance estimate is expressed as: ; in, I c ( y (This refers to a foggy image in the channel) c pixel values, L c ( x Ambient light value, β The weighting coefficients are derived from the comparative feature parameters. C Obtained through constraint transformation.

8. The method for dehazing coal mine images according to claim 1, characterized in that, In step S6, the weighted guided filtering employs adaptive regularization weights: for the guided image I and input image p Local eigenvalues ​​are calculated using a dual-window strategy, and a weight adjustment parameter is introduced. γ Constructing Adaptive Regularized Weights W I This method reduces the smoothing intensity in areas with rich texture and increases the smoothing intensity in areas with poor texture, thereby preserving edge details.

9. A method for dehazing coal mine images according to claim 1, characterized in that, In step S7, the image restoration is based on an improved atmospheric scattering model: ;in, J ( x The image is the restored image. I ( x (This is a foggy image.) L ( x () represents the ambient light value. t ( x () represents transmittance. t 0 is the minimum transmittance threshold.

10. A coal mine image dehazing system, characterized in that, The coal mine image defogging system, used to implement any one of claims 1 to 9, comprises an image acquisition module, a dust measurement module, a dual-channel processing module, a feature calculation module, an ambient light estimation module, a transmittance estimation module, a filtering optimization module, and an image restoration module. The image acquisition module is used to acquire real-time image signals of dust and fog in underground coal mines using a mining camera; The dust measurement module is used to measure the dust concentration data of the downhole environment in real time through a dust sensor; The dual-channel processing module is used to perform dark channel prior processing and bright channel prior processing on the dust and fog image signal respectively to obtain dark channel value and bright channel value; The feature calculation module is used to calculate a dust concentration compensation term based on the dust concentration data, and to calculate a comparison feature parameter and an adaptive scaling factor based on the global feature linear difference between the dark channel value and the bright channel value. The ambient light estimation module is used to use the adaptive scaling factor as the dynamic selection scaling factor of the brightest pixel in the ambient light estimation, and to combine it with the dust concentration compensation term to perform ambient light estimation and obtain the ambient light value. The transmittance estimation module is used to estimate transmittance based on the ambient light value, dark channel value and bright channel value through a weighted fusion method to obtain a transmittance map. The filtering optimization module is used to perform weighted guided filtering optimization on the transmittance map to enhance image details; The image restoration module is used to restore the image using an optimized transmittance map and ambient light value through an atmospheric scattering model, thereby obtaining a clear image after defogging.

Images