An underground coal mining image dewatering method based on machine vision

Abstract

The present invention discloses a method for removing water mist from underground coal mining images based on machine vision. First, a picture of the coal wall without a shearer is collected as a prior image. Then, when the shearer passes by, an image with water mist is collected. Through the analysis of the visual image, first, the water mist area is segmented, then the concentration of the water mist in the area is estimated. Finally, the estimated concentration is substituted into the model formula based on atmospheric scattering to estimate the image after removing the water mist in the area and substitute it into the original image. The water mist-free image obtained can accurately extract the information in the image to perform tasks such as coal-rock identification; and the intensity of water mist removal can be adjusted according to the concentration of different water mist areas, which can effectively act on the influence of the concentrated water flow and water mist formed by the shearer-mounted spray system on the image, stably and reliably eliminate the occlusion of the above factors on the coal wall information, and greatly improve the clarity of the image and the accuracy of extracting image information.

Description

Technical Field

[0001] The present invention relates to a method for removing water mist underground, specifically a method for removing water mist from underground coal mining images based on machine vision. Background Art

[0002] With the proposal and development of the concept of fewer or no workers in fully mechanized coal mining faces, obtaining information on the coal mining operation area through visual sensors has become an important part of realizing unmanned operation underground. Since dust is generated when the shearer cuts coal, in order to reduce the floating of dust in the air, continuous dust suppression treatment is carried out on the dust. Dust suppression treatment refers to the process of pressurizing a liquid through a high-pressure unit and then forming natural particles of about 0.01 - 0.15 mm in diameter by an atomizing nozzle, so that dust particles collide with liquid droplets or liquid films and are captured, thereby reducing the dust content in the air. Although the above process can effectively achieve the dust suppression effect, it will cause the captured images obtained by the visual sensors during continuous dust suppression to contain jet-like water mist. And since the water mist is different from the fog with almost uniform concentration in the atmosphere, the concentration of the water mist is uneven, which will block the information of the current position scene obtained, or make the detailed information obtained from the images inaccurate. Therefore, how to provide a new method to remove the water mist in the captured images, so as to obtain the information of the current position scene and ultimately effectively ensure the accuracy of coal mining images is one of the research directions in this industry. Summary of the Invention

[0003] Aiming at the problems existing in the above-mentioned prior art, the present invention provides a method for removing water mist from underground coal mining images based on machine vision, which can remove the water mist in the captured images, thereby obtaining the information of the current position scene and ultimately effectively ensuring the accuracy of coal mining images.

[0004] To achieve the above object, the technical solution adopted by the present invention is: a method for removing water mist from underground coal mining images based on machine vision. The adopted image acquisition system includes multiple cameras and a computer. One camera is installed under the lower part of the top beam of each hydraulic support, and the computer is installed on one of the hydraulic supports. Connect each camera to the computer. The specific steps are as follows:

[0005] Step 1: Before collecting images, manually adjust the angles of each camera so that each camera can collect the top and bottom of the corresponding coal wall, and the sum of all image information collected by all cameras can cover the top and bottom of the entire coal wall. During the image collection process, there are no obstacles and no strong light between the camera and the coal wall;

[0006] Step 2: After setting according to Step 1, before the shearer runs to the coverage area of any camera each time, first use each camera to collect a picture of the coal wall without the shearer and transmit it to the computer for storage as a prior image;

[0007] Step 3: When the shearer passes through the coverage area of any camera, the current camera takes a picture to obtain a water mist image. Each pixel point of the water mist image is processed, and the minimum value among the R, G, and B values of each pixel point is taken as the gray value of this pixel point, so as to obtain the dark channel image of the water mist image;

[0008] Step 4: Gaussian filtering is performed on the obtained dark channel image. After filtering, the image is converted into a binary image, and then an opening operation is performed on the binary image. The obtained image is used as a mask to segment the water mist area in the image, so as to obtain the water mist area image;

[0009] Step 5: Since the water sprayed by the dust suppression device is in a spray shape, according to the Gaussian diffusion formula, the water mist at different positions will have uneven water mist concentration due to the diffusion of water droplets. The average radius parameter of the artificial water mist droplets generated by the same device does not change greatly. Therefore, the main reason for the difference in the extinction performance of the water mist at different positions is the change in the droplet density. By observing the image, it can be found that the water mist concentration gradually decreases along the spraying direction. Therefore, the concentration of the water mist is related to the distance of this part of the area to the nozzle. Therefore, the relationship between the droplet density and the spatial position of the water mist is established. First, a threshold Q is set, and a 3*3 convolution kernel is used to perform convolution on the water mist area image obtained in Step 4 to find the area where the average gray value exceeds the threshold Q as the center of the sprayed water mist; then, according to the position of the center of the sprayed water mist, the distance Δd between each pixel point in the water mist area image and all the centers of the sprayed water mist is calculated, and the minimum value among them is taken as the basis for judging the concentration;

[0010] Step 6: As the shearer cuts, the coal wall in the same area will have slight changes due to the increase in the number of cuts. Therefore, considering the changes in the coal wall on the time scale, the same area of the waterless mist image after the first three cuts of the shearer is selected as the prior knowledge of the current area during the current cut of the shearer. The square of the difference between the water mist image and the waterless mist image in the same area is weighted and averaged and then square-rooted. The obtained value is used as the gray difference and brightness difference between the water mist image and the waterless mist image at this time; if there are less than three waterless mist images before the current cut of the shearer, the same area of the waterless mist images after all the previous cuts of the shearer is selected as the prior knowledge of the current area; if there is only one before, the difference between the current waterless mist image and the water mist image in the same area is used as the gray difference and brightness difference. If there are two before, that is, there are two waterless mist images, the average value of the differences between the first two waterless mist images and the current water mist image in the same area is used as the gray difference and brightness difference.

[0011] Step 7: Substitute the data obtained in Steps 5 and 6 into the model formula based on atmospheric scattering, so as to obtain the pixel values corresponding to each pixel point in the fog-free image in the water mist area image;

[0012] Step 8: Substitute the pixel values corresponding to each pixel point obtained in Step 7 into the positions corresponding to the water mist area in the water mist image in Step 3, so as to obtain a fog-free image;

[0013] Step 9: After the shearer leaves the camera coverage area this time, take another photo, which is used as new prior knowledge for the next use and also as the basis for adjusting the parameters of the defogging model subsequently.

[0014] Furthermore, the specific content of Step 5 is as follows:

[0015] First, set a threshold Q, perform convolution on the water mist area image obtained in Step 4 using a 3*3 convolution kernel, and find the area where the average gray value exceeds the threshold Q as the center of the sprayed water mist;

[0016] Then set two thresholds D1 and D2, D1 < D2, calculate the distances from the pixel points in the water mist area to all the centers of the sprayed water mist in this area, and take the minimum value d min , if d min is less than D2 and greater than D1, then Δd = d min , if d min is greater than or equal to the threshold D2, then Δd = D2, if d min is less than or equal to the threshold D1, then Δd = D1, substitute Δd into the Gaussian model formula to obtain d, and the specific calculation formula of d is as follows:

[0017]

[0018]

[0019] Among them, Q is the water mist release rate, with the unit of g / s; μ is the average velocity of the fog droplets ejected from the nozzle when releasing the water mist, with the unit of m / s; σ is the standard deviation of atmospheric diffusion, which is related to the atmospheric stability and terrain parameters.

[0020] Furthermore, the specific content of Step 6 is as follows:

[0021] Select the fog-free images after the shearer cuts coal in the previous three times. Since the time intervals from the previous three times to the operation of the shearer this time are different, the weights of the differences are also different. Set the weights of each image as w1, w2, and w3 respectively, and obtain three weight values through linear regression. Then take the fog-free images after the coal wall is cut in the previous three times in the same area and the water mist image when the shearer cuts coal this time, and perform the following operations on the gray values of each pixel point:

[0022]

[0023] where g’(x, y) is the gray value of the image with water mist at (x, y), and g i (x, y) is the gray value of the i adjacent water - mist - free images in the coal wall area at (x, y); a threshold G is set, and when the calculated difference is less than G, the difference is considered to be 0;

[0024] Similarly, for the brightness value, the method of the above - mentioned gray value is also used to judge the magnitude of the brightness difference. The weights of the three nearest HSV images are set as α1, α2, and α3. The brightness difference is obtained through the following formula, and the difference threshold of the brightness value is set as V. When the predicted difference is less than V, the difference is considered to be 0;

[0025]

[0026] where v’(x, y) is the brightness value of the image with water mist at (x, y), and v i (x, y) is the brightness value of the i adjacent water - mist - free images at (x, y).

[0027] Further, the specific content of step seven is as follows:

[0028] Substitute Δg, Δv, and d into the formula based on the atmospheric scattering model. The specific formula is:

[0029] I(x) = J(x)t(x)+A(1 - t(x)) (1)

[0030] t(x)=e -Kd(x)c(x) (2)

[0031] Among them, formula (1) is the atmospheric scattering model, x is the position of the pixel point, I(x) is the image with water mist obtained by the camera, J(x) is the water - mist - free image expected to be obtained, t(x) is the transmittance of the atmospheric medium in the light propagation path, and the calculation method is shown in formula (2), and A is the global atmospheric light value at infinity;

[0032] In formula (2), K represents the extinction coefficient, d(x) represents the depth of field of the incident light to the camera, and c(x) is the concentration distribution of the atmospheric medium in the transmission path;

[0033] Since K is a constant and the variation range of d(x) is also very small, taking Kd(x) as a constant β, then t(x) becomes:

[0034] t(x)=e -βc(x) (3)

[0035] The formula for obtaining the water - mist - free image becomes:

[0036]

[0037] Among them, the global atmospheric light value A at infinity is obtained by the following method:

[0038] 1) Take the top 0.1% of the pixel positions in the dark channel map according to the gray level;

[0039] 2) Among these positions, find the value of the pixel point with the highest gray level in the original water mist image as the A value;

[0040] The water mist concentration is calculated according to the characteristic quantity obtained above, and the expression of the water mist concentration is as follows:

[0041] c(x) = θ1Δg + θ2Δv + θ3d + θ0 (5)

[0042] Then the formula for obtaining the water mist-free image becomes

[0043]

[0044] Let βθ1 = C1, βθ2 = C2, βθ3 = C3, βθ0 = C0

[0045] Then the final formula becomes

[0046]

[0047] Among them, C0 to C3 are obtained by the following method:

[0048] Decompose C1Δg + Cθ2Δv + C3d + C0 by formula (8) to get

[0049]

[0050] Obtain a data set of water mist images and water mist-free images through experiments, obtain the values of C0 to C3 through linear regression, substitute them into formula (8) to finally obtain the formula for solving the water mist-free image; obtain the corresponding pixel values of each pixel point in the water mist area image in the fog-free image through the above formula.

[0051] Furthermore, when forming the water mist-free image in step eight, in order to prevent obvious boundaries from being generated, first use median filtering to process the image to remove the boundaries, and then use the original image as a guidance map to perform guided filtering on the image, so as to ensure the detail information of the generated water mist-free image.

[0052] Compared with the prior art, the present invention first collects pictures of the coal wall without the shearer as prior images, and then collects images with water mist when the shearer passes by. Through the analysis of visual images, the water mist area is first segmented, then the concentration of water mist in the area is estimated, and finally the estimated concentration is brought into the model formula based on atmospheric scattering to estimate the image after removing the water mist in the area and bring it into the original image. The obtained water-mist-free image can accurately extract the information in the image for coal-rock identification and other work. Therefore, the present invention can adjust the intensity of removing water mist according to the concentration of different water mist areas, can effectively act on the influence of the clustered water flow and water mist formed by the shearer-mounted spray system on the image, stably and reliably eliminate the occlusion of the above factors on the coal wall information, and greatly improve the clarity of the image and the accuracy of extracting image information. Brief Description of the Drawings

[0053] Figure 1 is a schematic diagram of the installation position of the camera in the present invention;

[0054] Figure 2 is the work flow chart of the present invention. Detailed Embodiment

[0055] The present invention will be further described below.

[0056] As Figure 1 shown, the image acquisition system adopted by the present invention includes multiple cameras and a computer. A camera is installed under the lower part of the top beam of each hydraulic support, and the computer is installed on one of the hydraulic supports. Each camera is connected to the computer. As Figure 2 shown, the specific steps are as follows:

[0057] Step 1: Before collecting images, manually adjust the angles of each camera so that each camera can collect the top and bottom of the corresponding coal wall, and the sum of all image information collected by all cameras can cover the top and bottom of the entire coal wall. There are no occlusions and no strong light between the camera and the coal wall during the image acquisition process;

[0058] Step 2: After setting according to Step 1, before the shearer runs to the coverage area of any camera each time, first use each camera to collect a picture of the coal wall without the shearer and transmit it to the computer for storage as a prior image;

[0059] Step 3: When the shearer passes through the coverage area of any camera, the current camera takes a picture to obtain an image with water mist. Each pixel point of the image with water mist is processed, and the minimum value among the R, G, and B values of each pixel point is taken as the gray value of this pixel point, so as to obtain the dark channel map of the image with water mist;

[0060] Step 4: Perform Gaussian filtering on the obtained dark channel image. After filtering, convert the image into a binary image, and then perform opening operation on the binary image. Use the obtained image as a mask to segment the water mist area in the image, so as to obtain the water mist area image;

[0061] Step 5: Since the water sprayed by the dust suppression device is in a spray shape, according to the Gaussian diffusion formula, the water mist at different positions will have the phenomenon of uneven water mist concentration due to the diffusion of fog droplets. The average radius parameter of the artificial water mist droplets generated by the same device does not change greatly. Therefore, the main reason for the difference in the light extinction performance of the water mist at different positions is the change in the fog droplet density. By observing the image, it can be found that the water mist concentration gradually decreases along the spraying direction. So the concentration of the water mist is related to the distance of this part of the area to the nozzle. Therefore, establish the relationship between the fog droplet density and the spatial position of the water mist, specifically as follows:

[0062] First, set a threshold Q, and use a 3*3 convolution kernel to perform convolution on the water mist area image obtained in Step 4 to find the area where the average gray value exceeds the threshold Q as the center of the sprayed water mist;

[0063] Then set two thresholds D1 and D2, D1 < D2, calculate the distance from the pixel points in the water mist area to all the centers of the sprayed water mist in this area, and take the minimum value d min , if d min is less than D2 and greater than D1, then Δd = d min , if d min is greater than or equal to the threshold D2, then Δd = D2, if d min is less than or equal to the threshold D1, then Δd = D1. Substitute Δd into the Gaussian model formula to get d, and the specific calculation formula of d is as follows:

[0064]

[0065]

[0066] Among them, Q is the water mist release rate, with the unit of g / s; μ is the average velocity of the fog droplets ejected from the nozzle when releasing the water mist, with the unit of m / s; σ is the standard deviation of atmospheric diffusion, which is related to the atmospheric stability and terrain parameters. Since there is no wind in the underground environment and it is relatively dim, the diffusion coefficient in a relatively stable state is taken as the standard deviation. After checking the table, the standard deviation is 0.065.

[0067] Step 6. As the shearer cuts, the coal wall in the same area will have slight changes due to the increase in the number of cuts. Therefore, considering the changes in the coal wall on the time scale, the same area of the anhydrous mist images after the first three cuts of the shearer is selected as the prior knowledge of the current area during the current shearer cut. Take the square of the difference between the water mist image and the anhydrous mist image in the same area, then calculate its weighted average and finally take the square root. The value obtained is used as the gray difference and brightness difference between the water mist image and the anhydrous mist image at this time, specifically as follows:

[0068] Select the anhydrous mist images after the first three cuts of the shearer. Since the time intervals between the first three cuts and the current shearer operation are different, the weights of the differences are also different. Weights w1, w2, and w3 are set for each image respectively, and three weight values are obtained through linear regression. Then, select the anhydrous mist images after the first three cuts of the coal wall in the same area and the water mist image during the current shearer cut, and perform the following operations on the gray values of each pixel:

[0069]

[0070] where g’(x,y) is the gray value of the water mist image at (x,y), and g i (x,y) is the gray value of the anhydrous mist images of the coal wall area near i at (x,y); set a threshold G. When the calculated difference is less than G, the difference is considered 0;

[0071] Similarly, for the brightness value, the method of the above gray value is used to judge the size of the brightness difference. The weights of the nearest three HSV images are set as α1, α2, and α3. The brightness difference is obtained through the following formula, and the difference threshold of the brightness value is set as V. When the predicted difference is less than V, the difference is considered 0;

[0072]

[0073] where v’(x,y) is the brightness value of the water mist image at (x,y), and v i (x,y) is the brightness value of the anhydrous mist images near i at (x,y).

[0074] If there are less than three anhydrous mist images before the current shearer cut, select the same area of the anhydrous mist images after all previous cuts of the shearer as the prior knowledge of the current area, and repeat this step to complete the calculation of the gray difference and brightness difference; if there is only one before, select the difference between the current anhydrous mist image and the water mist image in the same area as the gray difference and brightness difference. If there are two before, that is, there are two anhydrous mist images, then take the average of the differences between the first two anhydrous mist images and the current water mist image in the same area as the gray difference and brightness difference.

[0075] Step 7: Substitute the data obtained in Steps 5 and 6 into the model formula based on atmospheric scattering, so as to obtain the pixel values corresponding to each pixel point in the foggy area image in the fog-free image. Specifically:

[0076] Substitute Δg, Δv, and d into the formula based on the atmospheric scattering model. The specific formula is:

[0077] I(x) = J(x)t(x) + A(1 - t(x)) (1)

[0078] t(x) = e -Kd(x)c(x) (2)

[0079] Among them, formula (1) is the atmospheric scattering model, x is the position of the pixel point, I(x) is the foggy image obtained by the camera, J(x) is the expected fog-free image, t(x) is the transmittance of the atmospheric medium in the light propagation path, and the calculation method is shown in formula (2), and A is the global atmospheric light value at infinity;

[0080] In formula (2), K represents the extinction coefficient, d(x) represents the depth of field of the incident light to the camera, and c(x) is the concentration distribution of the atmospheric medium in the transmission path;

[0081] Since K is a constant and the variation range of d(x) is also very small, so take Kd(x) as a constant β, then t(x) becomes:

[0082] t(x) = e -βc(x) (3)

[0083] The formula for obtaining the fog-free image becomes:

[0084]

[0085] Among them, the global atmospheric light value A at infinity is obtained by the following method:

[0086] 1) Take the pixel positions of the top 0.1% according to the gray scale from the dark channel map;

[0087] 2) Among these positions, find the value of the pixel point with the highest gray scale in the original foggy image as the A value;

[0088] The fog concentration is calculated according to the characteristic quantities obtained above. The expression of the fog concentration is as follows:

[0089] c(x) = θ1Δg + θ2Δv + θ3d + θ0 (5)

[0090] Then the formula for obtaining the fog-free image becomes

[0091]

[0092] Let βθ1 = C1, βθ2 = C2, βθ3 = C3, βθ0 = C0

[0093] Then the final formula becomes

[0094]

[0095] where C0 to C3 are obtained by the following method:

[0096] Decompose C1Δg + Cθ2Δv + C3d + C0 through formula (8) to get

[0097]

[0098] Obtain a dataset with water mist images and water - mist - free images through experimental methods, obtain the values of C0 to C3 through linear regression, substitute them into formula (8) to finally obtain the formula for solving water - mist - free images; obtain the corresponding pixel values of each pixel point in the water mist area image in the fog - free image through the above formula.

[0099] Step eight: Substitute the pixel values corresponding to each pixel point obtained in step seven into the corresponding positions of the water mist area in the water - mist - filled image in step three. When forming a water - mist - free image, in order to prevent obvious boundaries from being generated, first perform median filtering on the image to remove the boundaries, and then use the original image as a guidance image to perform guided filtering on the image, so as to ensure the detail information of the generated water - mist - free image, and finally obtain a water - mist - free image;

[0100] Step nine: When the shearer leaves the area covered by the camera, take another photo, which is used as new prior knowledge for the next use, convert it into a grayscale image and an HSV image, which are used as two new prior knowledge for the next use and also as the basis for adjusting the parameters of the defogging model in the follow - up. The specific adjustment process is as follows: Compare the water - mist - free image taken after the shearer leaves this time with the water - mist - free image calculated by the above method during the cutting of the shearer this time. The specific comparison method is: Convert the two images into grayscale images and HSV images, and take the difference images of the grayscale subtraction and the brightness value subtraction. Add the values of each pixel point in the difference image. The obtained results are that the difference of the grayscale image is F1 and the difference of the HSV brightness is F2. If the result exceeds the set value, it is determined that the processing result of this image is not ideal and needs to be adjusted appropriately. If the difference in the grayscale image is too large, adjust C1, and if the difference in the brightness value of the HSV image is too large, adjust C2. By adjusting in this way, the accuracy of the subsequent water - mist - removing image can be higher.

[0101] The above are only the preferred embodiments of the present invention. It should be noted that for those of ordinary skill in the art, without departing from the principle of the present invention, several improvements and refinements can be made, and these improvements and refinements should also be regarded as the protection scope of the present invention.

Claims

1. A method for removing water mist from underground coal mining images based on machine vision, characterized in that, The adopted image acquisition system includes multiple cameras and a computer. A camera is installed at the lower part of the top beam of each hydraulic support, and the computer is installed on one of the hydraulic supports. Each camera is connected to the computer. The specific steps are as follows: Step 1: Before collecting images, manually adjust the angles of each camera so that each camera can collect the top and bottom of the corresponding coal wall, and the sum of all the image information collected by all cameras can cover the top and bottom of the entire coal wall. Step 2: After setting according to Step 1, before the shearer runs to the coverage area of any camera each time, first use each camera to collect a picture of the coal wall without the shearer and transmit it to the computer for storage as a prior image. Step 3: When the shearer passes through the coverage area of any camera, the current camera takes a picture to obtain a water mist image. Process each pixel point of the water mist image, and take the minimum value among the R, G, and B values of each pixel point as the gray value of this pixel point, so as to obtain the dark channel image of the water mist image. Step 4: Perform Gaussian filtering on the obtained dark channel image. After filtering, convert the image into a binary image, and then perform opening operation on the binary image. Use the obtained image as a mask to segment the water mist area in the image, so as to obtain the water mist area image. Step 5: First set a threshold Q. Use a 3*3 convolution kernel to perform convolution on the water mist area image obtained in Step 4, and find the area where the average gray value exceeds the threshold Q as the water spray center. Then calculate the distances Δd between each pixel point in the water mist area image and all the water spray centers respectively according to the positions of the water spray centers, and take the minimum value among them as the basis for judging the concentration. Step 6: As the shearer cuts, the coal wall in the same area will have slight changes due to the increase in the number of cuts. Therefore, considering the changes in the coal wall on the time scale, the same area of the waterless images after the previous three cuts of the shearer is selected as the prior knowledge of the current area during the current cut of the shearer. Take the square of the difference between the water mist image and the waterless image in the same area, then calculate its weighted average and finally take the square root. The obtained value is used as the gray difference and brightness difference between the water mist image and the waterless image at this time. If there are less than three waterless images before the current cut of the shearer, select the same area of the waterless images after all previous cuts of the shearer as the prior knowledge of the current area. Step 7: Substitute the data obtained in Steps 5 and 6 into the model formula based on atmospheric scattering, so as to obtain the pixel values corresponding to each pixel point in the water mist area image in the fogless image. Step 8: Substitute the pixel values corresponding to each pixel point obtained in Step 7 into the positions corresponding to the water mist area in the water mist image in Step 3, so as to obtain a fogless image. Step 9: After the current shearer leaves the coverage area of the camera, take another picture, which is used as new prior knowledge for the next use and also as the basis for adjusting the parameters of the defogging model later.

2. The method for removing water mist from underground coal mining images based on machine vision according to claim 1, characterized in that, The specific content of Step 5 is as follows: First, set a threshold Q. Use a 3*3 convolution kernel to convolve the water mist area image obtained in Step 4, and find the area where the average gray value exceeds the threshold Q as the center of the sprayed water mist; Then, set two thresholds D1 and D2, where D1 < D2, calculate the distance from the pixel points in the water mist area to all the centers of the sprayed water mist in this area, and take the minimum value d min , if d min is less than D2 and greater than D1, then Δd = d min , if d min is greater than or equal to the threshold D2, then Δd = D2, if d min is less than or equal to the threshold D1, then Δd = D1. Substitute Δd into the Gaussian model formula to obtain d. The specific calculation formula of d is as follows: Among them, Q is the water mist release rate, with the unit of g / s; μ is the average droplet ejection velocity at the nozzle when releasing the water mist, with the unit of m / s; σ is the atmospheric diffusion standard deviation, which is related to the atmospheric stability and terrain parameters.

3. The method for removing water mist from underground coal mining images based on machine vision according to claim 2, characterized in that, The specific content of Step 6 is as follows: Select the water-mist-free images after the first three coal shearer cuts. Set weights w1, w2, and w3 for each image respectively. Obtain three weight values through linear regression. Then, take the water-mist-free images after the first three coal wall cuts in the same area and the water-mist image during the current coal shearer cut, and perform the following operations on the gray values of each pixel: where g’(x,y) is the gray value of the image with water mist at (x,y), and g i (x,y) is the gray value of the i adjacent water-mist-free images in the coal wall area at (x,y); a threshold G is set, and when the calculated difference is less than G, the difference is considered to be 0; Similarly, for the brightness value, use the above method for the gray value to judge the size of the brightness difference. Set the weights of the last three HSV images as α1, α2, and α3. Obtain the difference in brightness through the following formula, and set the difference threshold of the brightness value as V. When the predicted difference is less than V, the difference is considered 0; where v’(x, y) is the brightness value of the water mist image at (x, y), and v i (x, y) is the brightness value of the i nearby waterless mist images at (x, y).

4. The method for removing water mist from underground coal mining images based on machine vision according to claim 3, characterized in that, The specific content of Step 7 is as follows: Substitute Δg, Δv, and d into the formula based on the atmospheric scattering model. The specific formula is: I(x) = J(x)t(x) + A(1 - t(x)) (1) t(x) = e -Kd(x)c(x) (2) Among them, formula (1) is the atmospheric scattering model, x is the position of the pixel, I(x) is the water-mist image obtained by the camera, J(x) is the water-mist-free image expected to be obtained, t(x) is the atmospheric medium transmittance in the light propagation path, and the calculation method is shown in formula (2), and A is the global atmospheric light value at infinity; In formula (2), K represents the extinction coefficient, d(x) represents the depth of field of the incident light to the camera, and c(x) is the concentration distribution of the atmospheric medium in the transmission path; Since K is a constant and the variation range of d(x) is also very small, take Kd(x) as a constant β, then t(x) becomes: t(x) = e -βc(x) (3) The formula for obtaining the water-mist-free image becomes: Among them, the global atmospheric light value A at infinity is obtained through the following method: 1) Take the top 0.1% of the pixel positions in the dark channel map according to the gray level; 2) Among these positions, find the value of the pixel with the highest gray level corresponding to it in the original water-mist image as the value of A; The water mist concentration is calculated according to the above-derived characteristic quantities. The expression of the water mist concentration is as follows: c(x) = θ1Δg + θ2Δv + θ3d + θ0 (5) Then the formula for obtaining the water-mist-free image becomes Let βθ1 = C1, βθ2 = C2, βθ3 = C3, βθ0 = C0 Then the final formula becomes Among them, C0 to C3 are obtained through the following method: Decompose C1Δg + Cθ2Δv + C3d + C0 through formula (8) to get Obtain a dataset of water-mist images and water-mist-free images through experiments. Obtain the values of C0 to C3 through linear regression and substitute them into formula (8) to finally obtain the formula for solving the water-mist-free image; obtain the pixel values corresponding to each pixel in the water mist area image in the fog-free image through the above formula.

5. The method for removing water mist from underground coal mining images based on machine vision according to claim 1, characterized in that, When performing the eighth step to form a fog-free image, in order to prevent obvious boundaries from being generated, first use median filtering to process the image to remove the boundaries, and then use the original image as a guidance map to perform guided filtering on the image, so as to ensure the detailed information of the generated fog-free image.

Images