An image rapid restoration method and system suitable for complex underground environment

Abstract

This invention discloses a method and system for rapid image restoration in complex underground mining environments, belonging to the field of image processing technology. The invention first converts the image to the HSV color space and adaptively corrects saturation based on the correlation between brightness and saturation. Then, it estimates transmittance through independent pixel-by-pixel calculations, fundamentally avoiding halo artifacts and time-consuming issues caused by block operations. Next, it acquires two sets of atmospheric light using a dark channel prior strategy and a brightness-contrast scoring strategy, respectively, and weights and fuses them using pixel-level transmittance as an indicator to obtain balanced atmospheric light, ensuring that the restoration result simultaneously considers contrast and brightness. Finally, it achieves rapid image restoration based on an atmospheric scattering model. This invention is computationally efficient, processing a single frame of a 1920×1080 resolution image in less than 15ms. It produces artifact-free images with natural brightness, complete detail preservation, and strong generalization capabilities. It can be stably applied in scenarios such as real-time monitoring, target detection, safety inspection, and equipment identification in underground mining without the need for training data.

Description

Technical Field

[0001] This invention relates to the field of image enhancement processing technology, and more specifically, to a method and system for rapid image restoration suitable for complex underground mining environments. Background Technology

[0002] Underground mining operations are often characterized by enclosed and complex environments, with persistent challenges such as high dust concentrations, strong water mist scattering, limited lighting, and uneven illumination. In such conditions, images acquired by image acquisition equipment typically suffer from degradation issues, including low overall contrast, blurred details, loss of edge information, dim brightness, color distortion, and significant noise interference. This severely reduces image usability and directly impacts the accuracy and reliability of subsequent high-level visual tasks such as target detection, personnel tracking, equipment identification, and anomaly monitoring, ultimately hindering the stable operation of intelligent and unmanned monitoring systems in mines.

[0003] Currently, image enhancement and restoration methods under scattering and low-light conditions are mainly divided into three categories: The first category is image restoration methods based on physical imaging models. These methods use atmospheric scattering models as their theoretical foundation and achieve inverse solving of degraded images by estimating key parameters such as transmittance and atmospheric light. A typical method is Dark Channel Prior (DCP), but this method uses local block operations to extract dark channel information, which results in high computational cost and time consumption, making it difficult to meet the real-time processing requirements in mines. At the same time, block operations are prone to halo artifacts and edge distortion, and in non-uniform dust scattering scenarios, transmittance is easily underestimated or overestimated, leading to restoration results that are too dark, discolored, or have residual scattering interference.

[0004] The second category is model-free image enhancement methods, which directly improve image contrast and brightness through pixel value mapping, histogram equalization, and multi-scale fusion. These methods are simple to implement and fast to compute, but lack physical constraints and are prone to problems such as excessive noise amplification, brightness imbalance, color distortion, and local over- or under-enhancement. Under conditions of low illumination, complex textures, and strong interference, it is difficult to balance subjective visual effects with objective quality indicators, and the algorithms lack robustness and stability.

[0005] The third category is image restoration methods based on deep learning. These methods learn the mapping relationship between degraded and clear images through deep neural networks, achieving relatively good performance on standard datasets. However, these methods heavily rely on large-scale labeled datasets. The distribution of real-world mining scene data differs significantly from that of synthetic datasets, resulting in insufficient model generalization ability. Furthermore, deep models have a large number of parameters and slow inference speed, making them difficult to deploy on resource-constrained platforms such as embedded terminals and edge computing devices in mines, thus failing to meet the engineering requirements for real-time monitoring.

[0006] In summary, existing technologies cannot simultaneously meet core engineering requirements such as adaptability to complex underground mining environments, computational efficiency, artifact-free output, balanced brightness and contrast, and real-time deployment capabilities. Summary of the Invention

[0007] To address the aforementioned issues, this invention aims to propose a method and system for rapid image restoration in complex underground mining environments. The entire process employs pixel-by-pixel computation, eliminating block operations and complex optimization processes. While ensuring high-speed computation, it achieves high-quality, artifact-free underground image restoration with balanced brightness and contrast.

[0008] To achieve the above objectives, the technical solution of the present invention is implemented as follows: A fast image restoration method suitable for complex underground mining environments includes the following steps: S1. Input a degraded image caused by dust or water mist in the mine. An image restoration framework is constructed based on an atmospheric scattering model; S2. Convert the image from RGB color space to HSV color space and separate the saturation channels. With brightness channel ; S3 based on saturation channel With brightness channel The correlation coefficient p is used to adaptively correct the saturation, resulting in the corrected saturation. ; S4, Based on modified saturation With brightness channel The mapping relationship is used to calculate pixel-level transmittance using a pixel-by-pixel independent calculation method. No local window filtering or block operations are used throughout the process; S5. Two sets of initial atmospheric light estimates were obtained by using the dark channel prior strategy and the brightness-contrast scoring strategy, respectively. and ; S6, with pixel-level transmittance As an adaptive weight, the two sets of atmospheric light are weighted and fused to obtain balanced atmospheric light that takes into account both contrast and brightness. ; S7. Based on the atmospheric scattering model, using pixel-level transmittance... With balance atmospheric light For degraded images Perform inverse transform to solve the problem and output a clear restored image. .

[0009] Furthermore, in step S2, the saturation channel With brightness channel The calculation formulas are as follows:

[0010]

[0011] in, This represents the pixel values ​​of the image across the three color channels (RGB).

[0012] Furthermore, in step S3, the formula for calculating the correlation coefficient p is:

[0013] Where H and W are the height and width of the input image, respectively, and std is the standard deviation; This is the squared term of the difference between the pixel value of the luminance channel and its mean. This is the squared term of the difference between the pixel value of the saturation channel and its mean. and These are the average values ​​for the luminance and saturation channels, respectively.

[0014] Furthermore, in step S3, the saturation correction formula is:

[0015] in, This is the corrected saturation. For the original saturation channel at the pixel The value at that location, For the luminance channel at the pixel The value at; For saturation channel With brightness channel The correlation coefficient; This is a saturation correction factor, with a value ranging from 0.1 to 0.5, used to prevent the transmittance from being underestimated under low illumination.

[0016] Furthermore, in step S4, the formula for calculating the pixel-level transmittance is: .

[0017] Furthermore, in step S5: Atmospheric light is estimated using two independent strategies: the dark channel transmittance optimization algorithm is used to calculate the first atmospheric light. And the second atmospheric light is calculated using an image block brightness-contrast comprehensive evaluation algorithm. First atmospheric light The specific algorithm is as follows: ; in, for The component in the c-th channel of the RGB spectrum; This is the set of pixels with the lowest transmittance, representing the top 0.1% of pixels. This represents the total number of selected pixels. For the degraded image Channel pixel values. Therefore, the first atmospheric light can be... Represented as a vector form consisting of RGB three-channel components: ; in , , They are respectively The components in the red, green, and blue color channels; Second atmospheric light The image block brightness-contrast comprehensive evaluation algorithm is used to calculate the brightness evaluation index of the k-th sub-block. First, the degraded image is uniformly divided into several local sub-blocks of the same size. The calculation formula is as follows: ; in, , These represent the height and width of a single sub-block, respectively. In the k-th sub-block, the coordinates are The pixel value of the c-th RGB channel, and the contrast evaluation index of the k-th sub-block. The calculation formula is as follows:

[0018] in, Let c be the pixel mean of the c-th RGB channel of the k-th sub-block; to achieve a balance between high brightness and low contrast, a comprehensive evaluation function for the sub-block is constructed. The details are as follows: ; Select the sub-region k with the highest score The region is then further divided into four smaller sub-regions, and the scoring process described above is repeated iteratively until the size of each sub-region is less than a preset threshold, resulting in the final candidate region Ω. Within the candidate region Ω, all pixels are traversed, and the distance between their RGB values ​​and pure white (255, 255, 255) is calculated. The pixel with the smallest distance is selected. This is a weighting coefficient that balances brightness and contrast, used to control the algorithm's preference for bright areas, as shown below:

[0019] in, These are the pixel values ​​of pixel x in the red, green, and blue channels, respectively; the RGB values ​​of this pixel are used as the second atmospheric light. ,Right now: .

[0020] Furthermore, in step S6, the balanced atmospheric light fusion formula is: .

[0021] Furthermore, in step S7, the image restoration formula is: ; in, To clearly restore the image; Input a degraded image; To balance atmospheric light; Pixel-level transmittance; To prevent extremely small positive numbers with a denominator of 0.

[0022] To achieve the above objectives, the present invention also provides a rapid image restoration system suitable for complex underground mining environments, comprising the following components connected in sequence: Color space conversion module: used to convert RGB images to HSV color space, separating the saturation channel and the brightness channel; Saturation Adaptive Correction Module: Used to calculate the correlation coefficient between brightness and saturation, and to adaptively correct the saturation. Pixel-level transmittance estimation module: used to calculate transmittance pixel by pixel based on the corrected saturation and luminance channels; Dual-strategy atmospheric light solution module: used to calculate two sets of initial atmospheric light using the dark channel prior strategy and the brightness-contrast scoring strategy respectively; Balanced Atmospheric Light Fusion Module: Used to fuse two sets of atmospheric light with pixel-level transmittance as a weight to obtain balanced atmospheric light; Image inverse model restoration module: used to solve for clear restored images based on the inverse operation of the atmospheric scattering model; Quantization output and storage module: used to quantize the restored image into standard 8-bit RGB format and output and store it; The system is used to perform the above-described method for rapid restoration of underground mining images.

[0023] Beneficial effects: No artifacts, high speed: The entire process adopts independent pixel-by-pixel operation, abandoning the block operation of the traditional dark channel method, thus eliminating halo artifacts and edge distortion from the mechanism; the algorithm has low complexity, and the processing time of a single frame of a 1920×1080 resolution image is less than 15ms, which can meet the real-time processing requirements of 30fps video streams in mines.

[0024] Dynamic balance between brightness and contrast: An innovative dual-strategy atmospheric light estimation and transmittance weighted fusion mechanism is proposed, which solves the common problem of traditional methods either restoring images that are too dark or too exposed, so that the restored images have a natural brightness distribution and moderate contrast.

[0025] Strong robustness and good generalization: Based on the physical model design, it does not rely on any training data and can adapt to complex scenarios with different dust concentrations, different lighting conditions and different coal mine working conditions, thus avoiding the problem of insufficient generalization of deep learning methods.

[0026] Easy to deploy: The algorithm has simple logic and low computational load, and can be directly deployed on embedded edge devices such as FPGA and ARM. It has strong engineering practicality and can significantly improve the performance of systems such as target detection and safety inspection in mines. Attached Figure Description

[0027] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an undue limitation of the invention. In the drawings: Figure 1 This is a flowchart of the main process of the image fast restoration method applicable to complex underground mining environments as described in this embodiment of the invention; Figure 2 This image comparison chart shows the image restoration effects of the fast image restoration method applicable to complex underground mining environments of the present invention, the GDCP method, and the MR method on several common degraded images in underground mining environments. Figure 3 This is a schematic diagram of the structure of the image rapid restoration system suitable for complex underground mining environments as described in an embodiment of the present invention. Detailed Implementation

[0028] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other.

[0029] The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0030] Example 1 See Figure 1 A method for rapid image restoration suitable for complex underground mining environments includes the following steps: S1. Input a degraded image caused by dust or water mist in the mine. An image restoration framework is constructed based on an atmospheric scattering model; S2. Convert the image from RGB color space to HSV color space and separate the saturation channels. With brightness channel ; S3 based on saturation channel With brightness channel The correlation coefficient p is used to adaptively correct the saturation, resulting in the corrected saturation. ; S4, Based on modified saturation With brightness channel The mapping relationship is used to calculate pixel-level transmittance using a pixel-by-pixel independent calculation method. No local window filtering or block operations are used throughout the process; S5. Two sets of initial atmospheric light estimates were obtained by using the dark channel prior strategy and the brightness-contrast scoring strategy, respectively. and ; S6, with pixel-level transmittance As an adaptive weight, the two sets of atmospheric light are weighted and fused to obtain balanced atmospheric light that takes into account both contrast and brightness. ; S7. Based on the atmospheric scattering model, using pixel-level transmittance... With balance atmospheric light For degraded images Perform inverse transform to solve the problem and output a clear restored image. .

[0031] This embodiment is based on an atmospheric scattering physical model. It first decouples brightness and color information through HSV color space conversion, then corrects saturation based on their correlation, and finally calculates transmittance pixel-by-pixel. Innovatively, it employs a dual strategy to estimate two sets of atmospheric light, one emphasizing contrast and the other brightness, and finally fuses them using transmittance as a weight to obtain balanced atmospheric light, which is then substituted into the inverse model to solve for a clear image. The entire process abandons the local block operations of traditional methods; all core steps are performed independently pixel-by-pixel.

[0032] This embodiment fundamentally solves the problems of halo artifacts and edge distortion caused by block filtering in traditional dark channel prior methods; it overcomes the inherent defects of single atmospheric light estimation and achieves a dynamic balance between brightness and contrast in the restored image; the algorithm has low overall complexity and high computational efficiency, meeting the real-time processing requirements of underground video streams; it does not rely on any training data, has strong generalization ability, and can adapt to the complex working conditions of different coal mines.

[0033] In a specific example, in step S2, the saturation channel With brightness channel The calculation formulas are as follows:

[0034]

[0035] in, This represents the pixel values ​​of the image across the three color channels (RGB).

[0036] This embodiment clarifies the specific calculation methods for the saturation channel S(x) and the brightness channel V(x) in the HSV color space: saturation is defined as 1 minus the ratio of the minimum to the maximum value of the three RGB channels, and brightness is defined as the maximum value of the three RGB channels. This definition completely separates the coupled color information and illumination information in the RGB space.

[0037] This embodiment achieves complete decoupling of color saturation and brightness, avoiding parameter estimation deviations caused by mutual interference between the three channels in the RGB space; the calculation method is simple and efficient, requiring only pixel-level maximum and minimum value calculations, without complex calculations; it provides independent and accurate feature inputs for subsequent correlation calculations and saturation correction.

[0038] In a specific example, in step S3, the formula for calculating the correlation coefficient p is:

[0039] Where H and W are the height and width of the input image, respectively, and std is the standard deviation; This is the squared term of the difference between the pixel value of the luminance channel and its mean. This is the squared term of the difference between the pixel value of the saturation channel and its mean. and These are the average values ​​for the luminance and saturation channels, respectively.

[0040] This embodiment defines a formula for calculating the correlation coefficient p between the luminance channel and the saturation channel. The linear correlation between the two channels is quantified by calculating the sum of the products of the squared differences between the two channels and the mean, and then dividing by the square root of the product of the total number of pixels in the image and the standard deviation of the two channels.

[0041] This embodiment accurately quantifies the negative correlation between brightness and saturation in low-light environments in mines (the lower the brightness, the lower the saturation); this embodiment provides an objective quantitative basis for subsequent adaptive saturation correction, avoiding the subjectivity of manually setting parameters; this embodiment can automatically adjust the correction intensity according to the actual lighting conditions of different images, improving the robustness of the algorithm.

[0042] In a specific example, in step S3, the saturation correction formula is:

[0043] in, This is the corrected saturation. For the original saturation channel at the pixel The value at that location, For the luminance channel at the pixel The value at; For saturation channel With brightness channel The correlation coefficient; This is a saturation correction factor, with a value ranging from 0.1 to 0.5, used to prevent the transmittance from being underestimated under low illumination.

[0044] This embodiment adaptively corrects the original saturation based on the calculated correlation coefficient p: The difference between the brightness value and (1-p) times the brightness value is multiplied by a correction coefficient α, and the result is finally constrained within the [0,1] interval. The correction strength is jointly controlled by the correlation coefficient p and the correction coefficient α.

[0045] This embodiment specifically addresses the problem of severely underestimated saturation in low-light environments in mines; it avoids deviations in subsequent transmittance calculations caused by underestimation of saturation, thus improving the accuracy of transmittance estimation; the correction intensity can be flexibly adjusted through the α coefficient to prevent color distortion caused by over-correction; the correction results are constrained within a reasonable range, ensuring the numerical stability of subsequent calculations.

[0046] In a specific example, in step S4, the formula for calculating the pixel-level transmittance is: .

[0047] This embodiment directly utilizes the mapping relationship between the corrected saturation and brightness channels to calculate the transmittance pixel by pixel through simple arithmetic operations: t(x) = 1 - (1 - corrected saturation) × brightness.

[0048] This step does not use any local window filtering or block operations; the calculation of each pixel is completely independent. This is the core implementation of the artifact-free characteristics of this invention: it fundamentally eliminates the edge blurring and halo problems caused by block operations; the calculation logic is extremely simple, requiring only one multiplication and one subtraction, and the operation speed is extremely fast; all pixels can be calculated in parallel, making it particularly suitable for acceleration on parallel computing platforms such as GPUs and FPGAs; the transmittance estimation accuracy is high, and it can accurately reflect the spatial distribution of non-uniform dust scattering under the mine.

[0049] In a specific instance, in step S5: Atmospheric light is estimated using two independent strategies: the dark channel transmittance optimization algorithm is used to calculate the first atmospheric light. And the second atmospheric light is calculated using an image block brightness-contrast comprehensive evaluation algorithm. First atmospheric light The specific algorithm is as follows: ; in, for The component in the c-th channel of the RGB spectrum; This is the set of pixels with the lowest transmittance, representing the top 0.1% of pixels. This represents the total number of selected pixels. For the degraded image Channel pixel values. Therefore, the first atmospheric light can be... Represented as a vector form consisting of RGB three-channel components: ; in , , They are respectively The components in the red, green, and blue color channels; Second atmospheric light The image block brightness-contrast comprehensive evaluation algorithm is used to calculate the brightness evaluation index of the k-th sub-block. First, the degraded image is uniformly divided into several local sub-blocks of the same size. The calculation formula is as follows: ; in, , These represent the height and width of a single sub-block, respectively. In the k-th sub-block, the coordinates are The pixel value of the c-th RGB channel, and the contrast evaluation index of the k-th sub-block. The calculation formula is as follows:

[0050] in, Let c be the pixel mean of the c-th RGB channel of the k-th sub-block; to achieve a balance between high brightness and low contrast, a comprehensive evaluation function for the sub-block is constructed. The details are as follows: ; Select the sub-region k with the highest score The region is then further divided into four smaller sub-regions, and the scoring process described above is repeated iteratively until the size of each sub-region is less than a preset threshold, resulting in the final candidate region Ω. Within the candidate region Ω, all pixels are traversed, and the distance between their RGB values ​​and pure white (255, 255, 255) is calculated. The pixel with the smallest distance is selected. This is a weighting coefficient that balances brightness and contrast, used to control the algorithm's preference for bright areas, as shown below:

[0051] in, These are the pixel values ​​of pixel x in the red, green, and blue channels, respectively; the RGB values ​​of this pixel are used as the second atmospheric light. ,Right now: .

[0052] It should be noted that this embodiment, through dual-strategy atmospheric light estimation, fundamentally solves the inherent defects of traditional single atmospheric light estimation methods, laying a core foundation for subsequent balanced atmospheric light fusion. The specific technical effects are as follows: This embodiment achieves a complementary approach of prioritizing contrast and prioritizing brightness. The first atmospheric light A1 is based on the dark channel prior principle, which focuses on preserving the detail contrast of the image and can accurately restore the texture information of the weakly scattering area, avoiding the problem of the restored image being grayish. The second atmospheric light A2 is based on a comprehensive evaluation of brightness and contrast, focusing on improving the overall brightness of the image. It can effectively improve the dimness in low-light and high-dust areas in mines and avoid the overall darkness of the restored image.

[0053] Significantly improved the robustness of atmospheric light estimation Traditional dark channel prior methods directly select the brightest 0.1% pixel in the dark channel as atmospheric light, which easily misclassifies dark objects (such as black equipment or coal) as atmospheric light, resulting in an underestimation. This embodiment filters candidate pixels by sorting by transmittance, ensuring that the candidate region is indeed the region with the strongest scattering, which greatly reduces the probability of misclassification.

[0054] Traditional brightness-first methods directly select the brightest pixel globally, which is susceptible to interference from local high-brightness noise such as mine lamp reflection and metal surface reflection. This embodiment effectively eliminates the influence of complex texture areas and noise points through iterative block division and comprehensive evaluation of brightness and contrast, making the atmospheric light estimate closer to the real ambient scattered light.

[0055] Perfectly adapted for non-uniform scattering scenarios in mining environments The underground environment exhibits significant non-uniform dust distribution (e.g., high dust concentration near the coal mining machine and low dust concentration in distant roadways), making it impossible for a single atmospheric light source to simultaneously meet the restoration needs of different areas. This embodiment provides two sets of atmospheric light sources corresponding to optimal estimates of different scattering intensities, offering high-quality foundational data for subsequent adaptive fusion weighted by transmittance, ultimately achieving a dynamic balance between global brightness and contrast.

[0056] In a specific example, in step S6, the balanced atmospheric light fusion formula is: In step S7, the image restoration formula is: ; in, To clearly restore the image; Input a degraded image; To balance atmospheric light; Pixel-level transmittance; To prevent extremely small positive numbers with a denominator of 0.

[0057] In this embodiment, pixel-level transmittance is used as an adaptive weight to perform weighted fusion of the two sets of atmospheric light: the lower the transmittance (the denser the dust), the greater the weight of A2; the higher the transmittance (the lighter the dust), the greater the weight of A1. Finally, the inverse formula of the atmospheric scattering model is substituted to solve for the clear image, and the ε parameter is used to prevent the denominator from being zero.

[0058] This embodiment achieves a dynamic adaptive balance between global brightness and contrast: high dust areas automatically increase brightness, and low dust areas automatically enhance contrast; it solves the common problem of traditional methods restoring images that are either too dark overall or overexposed in certain areas; the ε parameter ensures the stability of numerical calculations and avoids the error of dividing by zero; the inverse operation process is simple and efficient, requiring only pixel-by-pixel addition, subtraction, multiplication, and division operations.

[0059] To verify the effectiveness of the rapid image restoration method of this invention applicable to complex underground mining environments, several common degraded underground images were selected for testing and verification of its enhancement performance. Figure 2 Figures (a), (b), and (c) show the results. As can be seen from the figures, the GDCP method only improves the overall brightness of the image and has little effect on removing the haze effect. The MR method can improve image contrast to some extent, but there is a phenomenon of low illumination in distant background areas. In contrast, the restoration effect of this invention can significantly reduce the hazy background of the image and improve the overall contrast and brightness of the restored image.

[0060] Table 1 shows... Figure 2 Results of different enhancement methods Score

[0061] Table 2 is... Figure 2 Results of different enhancement methods Entropy Score

[0062] Tables 1 and 2 respectively give the normalized gradient mean of the visible edges ( ) and information entropy ( Entropy As can be seen from Table 1, this invention has achieved the highest [percentage missing]. The score indicates that the restoration result has high contrast. Table 2 shows that this invention achieved the highest or second-highest score. Entropy The score indicates that the restoration result contains a wealth of scene information.

[0063] In summary, the processed underground images in this embodiment completely eliminate the whitening effect of dust fogging, clearly restore details such as tunnel edges, equipment outlines, and cable textures, and have uniform and natural brightness distribution, moderate contrast, no color cast, and no artifacts. The algorithm has low processing time per frame and can be adapted to real-time deployment of embedded edge devices underground, possessing strong engineering practical value and scenario generalization ability.

[0064] Example 2 To achieve the above objectives, see Figure 3 This embodiment also provides a rapid image restoration system for mines based on pixel-level transmittance and balanced atmospheric light estimation, comprising the following components connected in sequence: Color space conversion module: used to convert RGB images to HSV color space, separating the saturation channel and the brightness channel; Saturation Adaptive Correction Module: Used to calculate the correlation coefficient between brightness and saturation, and to adaptively correct the saturation. Pixel-level transmittance estimation module: used to calculate transmittance pixel by pixel based on the corrected saturation and luminance channels; Dual-strategy atmospheric light solution module: used to calculate two sets of initial atmospheric light using the dark channel prior strategy and the brightness-contrast scoring strategy respectively; Balanced Atmospheric Light Fusion Module: Used to fuse two sets of atmospheric light with pixel-level transmittance as a weight to obtain balanced atmospheric light; Image inverse model restoration module: used to solve for clear restored images based on the inverse operation of the atmospheric scattering model; Quantization output and storage module: used to quantize the restored image into standard 8-bit RGB format and output and store it; The system is used to perform the above-described method for rapid restoration of underground mining images.

[0065] The pixel-level transmittance and balanced atmospheric light estimation fast image restoration system for mines in this embodiment has the same advantages over the prior art as the pixel-level transmittance and balanced atmospheric light estimation fast image restoration method for mines described above, and will not be repeated here.

[0066] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for rapid image restoration suitable for complex underground mining environments, characterized in that, Includes the following steps: S1. Input a degraded image caused by dust or water mist in the mine. An image restoration framework is constructed based on an atmospheric scattering model; S2. Convert the image from RGB color space to HSV color space and separate the saturation channels. With brightness channel ; S3, based on the saturation channel With brightness channel The correlation coefficient p is used to adaptively correct the saturation, resulting in the corrected saturation. ; S4, Based on modified saturation With brightness channel The mapping relationship is used to calculate pixel-level transmittance using a pixel-by-pixel independent calculation method. No local window filtering or block operations are used throughout the process; S5. Two sets of initial atmospheric light estimates were obtained by using the dark channel prior strategy and the brightness-contrast scoring strategy, respectively. and ; In step S5: Atmospheric light is estimated using two independent strategies: the dark channel transmittance optimization algorithm is used to calculate the first atmospheric light. And the second atmospheric light is calculated using an image block brightness-contrast comprehensive evaluation algorithm. First atmospheric light The specific algorithm is as follows: ； in, for The component in the c-th channel of the RGB spectrum; This is the set of pixels with the lowest transmittance, representing the top 0.1% of pixels. This represents the total number of selected pixels. For the degraded image Channel pixel values; thus, the first atmospheric light can be obtained. Represented as a vector form consisting of RGB three-channel components: ； in , , They are respectively The components in the red, green, and blue color channels; Second atmospheric light The image block brightness-contrast comprehensive evaluation algorithm is used to calculate the brightness evaluation index of the k-th sub-block. First, the degraded image is uniformly divided into several local sub-blocks of the same size. The calculation formula is as follows: ； in, , These represent the height and width of a single sub-block, respectively. In the k-th sub-block, the coordinates are The pixel value of the c-th RGB channel, and the contrast evaluation index of the k-th sub-block. The calculation formula is as follows: in, Let c be the pixel mean of the c-th RGB channel of the k-th sub-block; to achieve a balance between high brightness and low contrast, a comprehensive evaluation function for the sub-block is constructed. The details are as follows: ； The highest-scoring sub-region is selected and further divided into four smaller sub-regions. This scoring process is repeated iteratively until the sub-region size is less than a preset threshold, resulting in the final candidate region Ω. Within candidate region Ω, all pixels are traversed, and the distance between their RGB values ​​and pure white (255, 255, 255) is calculated. The pixel with the smallest distance is selected. This is a weighting coefficient that balances brightness and contrast, used to control the algorithm's preference for bright areas, as shown below: in, These are the pixel values ​​of pixel x in the red, green, and blue channels, respectively; the RGB values ​​of this pixel are used as the second atmospheric light. ,Right now: ； S6, with pixel-level transmittance As an adaptive weight, the two sets of atmospheric light are weighted and fused to obtain balanced atmospheric light that takes into account both contrast and brightness. ; S7. Based on the atmospheric scattering model, using pixel-level transmittance... With balance atmospheric light For degraded images Perform inverse transform to solve the problem and output a clear restored image. .

2. The rapid image restoration method applicable to complex underground mining environments according to claim 1, characterized in that, In step S2, the saturation channel With brightness channel The calculation formulas are as follows: in, This represents the pixel values ​​of the image across the three color channels (RGB).

3. The rapid image restoration method applicable to complex underground mining environments according to claim 1, characterized in that, In step S3, the formula for calculating the correlation coefficient p is: Where H and W are the height and width of the input image, respectively, and std is the standard deviation; This is the squared term of the difference between the pixel value of the luminance channel and its mean. This is the squared term of the difference between the pixel value of the saturation channel and its mean. and These are the average values ​​for the luminance and saturation channels, respectively.

4. The rapid image restoration method applicable to complex underground mining environments according to claim 1, characterized in that, In step S3, the saturation correction formula is: in, This is the corrected saturation. For the original saturation channel at the pixel The value at that location, For the luminance channel at the pixel The value at; For saturation channel With brightness channel The correlation coefficient; This is a saturation correction factor, with a value ranging from 0.1 to 0.5, used to prevent the transmittance from being underestimated under low illumination.

5. The rapid image restoration method applicable to complex underground mining environments according to claim 1, characterized in that, In step S4, the formula for calculating the pixel-level transmittance is: 。 6. The rapid image restoration method applicable to complex underground mining environments according to claim 1, characterized in that, In step S6, the balanced atmospheric light fusion formula is: .

7. The rapid image restoration method applicable to complex underground mining environments according to claim 1, characterized in that, In step S7, the image restoration formula is: ； in, To clearly restore the image; Input a degraded image; To balance atmospheric light; Pixel-level transmittance; To prevent extremely small positive numbers with a denominator of 0.

8. A rapid image restoration system suitable for complex underground mining environments, characterized in that, Including those connected sequentially: Color space conversion module: used to convert RGB images to HSV color space, separating the saturation channel and the brightness channel; Saturation Adaptive Correction Module: Used to calculate the correlation coefficient between brightness and saturation, and to adaptively correct the saturation. Pixel-level transmittance estimation module: used to calculate transmittance pixel by pixel based on the corrected saturation and luminance channels; Dual-strategy atmospheric light solution module: used to calculate two sets of initial atmospheric light using the dark channel prior strategy and the brightness-contrast scoring strategy respectively; Balanced Atmospheric Light Fusion Module: Used to fuse two sets of atmospheric light with pixel-level transmittance as a weight to obtain balanced atmospheric light; Image inverse model restoration module: used to solve for clear restored images based on the inverse operation of the atmospheric scattering model; Quantization output and storage module: used to quantize the restored image into standard 8-bit RGB format and output and store it; The system is used to perform the rapid restoration method for underground images as described in any one of claims 1 to 7.

Images