Coal mine water exploration and drainage drill rod counting method based on deep learning

Abstract

The invention belongs to the technical field of machine vision and coal mine intellectualization, and discloses a coal mine water exploration and drainage drill rod counting method based on deep learning. According to the specific technical scheme, a coal mine underground video stream is obtained; carrying out preprocessing by utilizing the obtained drill site video stream; analyzing the to-be-detected drilling image by using a pre-constructed target detection model to obtain position information of the key area; calibrating by using the initial position and the end position of the drilling machine; constructing a drill rod counting model by using the calibrated position information so as to determine the rod feeding times of the drill site; according to the drill rod counting method provided by the invention, full-process monitoring and counting of drilling and rod withdrawing can be realized, the workload of workers is reduced, and the situations of false report, missing report and the like are prevented, so that the development efficiency is improved, and the personal safety of coal miners is ensured.

Description

Technical Field

[0001] The present invention belongs to the technical field of machine vision and coal mine intelligence, and specifically relates to a method for counting coal mine water exploration and drainage drill rods based on deep learning. Background Art

[0002] Gas accidents often account for a large proportion of major safety accidents in coal mines. Gas is usually stored in coal seams and released as coal seams are mined. In the narrow working environment underground, when gas reaches a certain concentration, explosion accidents are prone to occur. In order to avoid or reduce the occurrence of gas accidents, coal mines usually use drilling extraction to extract and discharge gas in advance. The depth of the drilling affects the effect of gas extraction.

[0003] So far, traditional video surveillance systems have been widely used in coal mines, but the management of water exploration and drainage is still not perfect. In the coal mine production process, there are three main traditional drill rod counting methods: the drill rig operator manually records the number of drill rods and the drilling depth; the proximity switch is installed on the drill rig. When the drill rod machine moves to a certain position, the proximity switch sends a signal, and then accumulates the count through other devices; the relevant sensor is installed on the drill rig, and the distance parameter sensor between the probe and the drill rig marker is transmitted to the PLC. When the PLC analyzes that the drill bit has reached the specified position, the controller will automatically count, thereby realizing the counting problem of the drill rig. The first method mainly relies on manual recording, which may result in omissions, multiple records, and wrong records. The second method uses proximity switches. The underground environment is harsh, humid, and dusty, and the proximity switches are easily damaged, making the drill rod counting invalid. In addition, the principle is too simple and the error is relatively large. The proximity switches can be manually controlled and the recorded data is unreliable. The third method, PLC, is easy to use, the coding is simple, and the functions are powerful. However, the drilling process of the drilling rig is complicated, and it is difficult to distinguish between drilling and drilling. It mainly requires the installation of a large number of control and detection equipment, and the price of hardware equipment is usually expensive, which requires a large investment and is not conducive to promotion.

[0004] In summary, it can be found that the existing technology is not suitable for the detection method of actual water exploration footage in coal mine drilling sites. Therefore, how to accurately identify the water exploration footage and then apply it to the actual coal mine water exploration and drainage monitoring and management has become a technical problem that needs to be solved urgently. Summary of the invention

[0005] In order to solve the technical problems existing in the prior art, the present invention provides a method for counting water exploration and drainage drill rods in coal mines based on deep learning, which realizes the full process detection and counting of drilling and rod withdrawal, improves development efficiency, and ensures the personal safety of coal miners.

[0006] To achieve the above purpose, the technical solution adopted by the present invention is: a method for counting water drilling rods in coal mines based on deep learning, and the specific steps are as follows:

[0007] Step 1: Obtain downhole drilling video stream;

[0008] Step 2: converting the downhole drilling video stream as the input video stream into a drilling rig image for preprocessing to obtain an enhanced image;

[0009] Step 3: Analyze the drilling rig image using the pre-built target detection model (positioning) to determine the drilling rig detection area for the identification target, thereby obtaining the location information of the drilling rig;

[0010] Step 4: calibrate the drilling rig position information to obtain the starting position and ending position of the drilling rig feeding device;

[0011] Step 5: Use the pre-built counting model to analyze the position information of the drilling rig feeding device, count the rod feeding and rod unloading according to the calibrated position, and obtain the water exploration depth of the current operation.

[0012] In step one, the specific acquisition steps are: adjust the camera position to the specified position, adjust the drill rig position and angle to the design requirements, send the drilling start command through the web or mobile app, and when the single hole is completed, send the end command. The AI ​​analysis module encodes and decodes the video stream to obtain the effective drilling video stream.

[0013] In step 2, the specific image preprocessing steps are: decoding the acquired video stream to obtain the original image and enhancing it to obtain an enhanced image, and performing denoising on the enhanced image;

[0014] Specific image enhancement methods include color space conversion, Gamma correction and CLAHE enhancement;

[0015] Specific denoising methods include Gaussian filtering, median filtering and histogram equalization.

[0016] The RGB color space of the original image is converted to the YCbCr color space, the Y brightness component of the YCbCr color space is extracted to perform adaptive equalization with limited contrast, the contrast between the target information and the background is improved, and then it is converted back to the RGB color space to obtain an enhanced image.

[0017] In step three, the specific steps of the target detection model construction method are: receiving manually labeled drilling rig samples, the manually labeled objects at least include the feeding device and the drill rod in the drilling rig image samples; using the labeled drilling rig image samples to train the target detection model, and obtaining the optimal target detection model as the final target detection model.

[0018] The target drilling rig detection area divides the drilling rig into six areas: pneumatic drilling rig feeding device area, pneumatic drilling rig drill rod area, hydraulic drilling rig feeding device area, hydraulic drilling rig drill rod area, crawler drilling rig feeding device area, and crawler drilling rig drill rod area.

[0019] Before manual drilling, the drilling rig feed device is reciprocated once in an unloaded state, and the AI ​​analysis module calibrates the position of the feed device based on the pixel coordinate system to obtain the initial position and end position of the feed device.

[0020] During the drilling process of the drilling rig, the position information of the feed device is tracked in real time, and whether there is a drill rod is detected; when the feed device is in the initial position, it is recorded as 1. During the movement of the drilling rig, it is detected that the drill rod has always existed. When the feed device is in the terminal position, it is recorded as 1, and the number of drill rods is +1. This process is repeated to count the drill rods.

[0021] The Gamma correction function is used to enhance the brightness component of the weak light area and reduce the brightness component of the strong light area. The Gamma function expression is as follows:

[0022] S Gamma (x,y)=S γ(x,y) (x,y)

[0023] Among them, S(x,y) is the original image, γ(x,y) is the correction parameter of the Gamma function, S Gamma (x, y) is the image after correction;

[0024] The correction parameter value of the Gamma function is adaptively adjusted according to the distribution characteristics of the illumination information of each pixel. The function expression of γ(x,y) is as follows:

[0025]

[0026] Among them, γ(x,y) is the correction parameter value of the Gamma function, I(x,y) is the illumination component, and I mean (x, y) is the average illumination component of the area, φ is the margin, I min (x, y) is the minimum illumination component, I max (x,y) is the highest illumination component.

[0027] The edge-preserving filtering algorithm guides filtering to extract the reflection component. The specific expression of the guided filtering function is:

[0028]

[0029] Among them, q is the output image, U is the guide image, a and b are the invariant coefficients of the function when the center of window ω is located at k. According to the unconstrained image restoration method, it is transformed into an optimization problem, and its value function is:

[0030]

[0031] Among them, p i is the image pixel before denoising, a and b are the invariant coefficients of the function when the center of window ω is located at k, ε is the weight value, and the formula is solved using the least squares method:

[0032]

[0033] Among them, μ k represents the mean in the local window ω, U is the guide image, represents the variance in the local window ω, |ω| represents the number of pixels in the local window, and the window sampling operation is taken in the entire image. The final formula for taking the average value is:

[0034]

[0035] in, ω k It refers to all windows containing pixel i, and k is the center position corresponding to different windows.

[0036] Compared with the prior art, the present invention has the following beneficial effects:

[0037] 1. The present invention can meet the needs of water exploration and release operations by selecting the camera position and algorithm scene with the help of mobile phone APP, with low cost. The mobile phone APP can detect the camera position in real time and assist in fixing it in a suitable position, so that the drilling rig and personnel in the water exploration scene are within a reasonable field of vision.

[0038] Second, the present invention converts the image into color space, performs Gamma correction, and CLAHE enhancement to improve the contrast of the image target information. Since the water exploration scene is extremely complex and not fixed, the quality of the video stream is average. By pre-processing the video stream frame by frame, a basis is provided for the subsequent statistics of the pole.

[0039] 3. The present invention can improve the accuracy of drill rod statistics by calibrating the drill rig position based on the pixel coordinate system. In actual operation scenarios, according to design requirements, the number of holes to be drilled in a single operation is not less than 6, and the position of the drill rig will be adjusted at any time according to the position of the hole. The present invention calibrates the initial position of the single-hole drill rig, thereby reducing the labor intensity of workers.

[0040] Fourth, the present invention performs drill rod statistics based on the actual operation principle of the drilling rig, and can be used for different types of drilling rigs, such as pneumatic drilling rigs, hydraulic drilling rigs, crawler drilling rigs, etc. The present invention performs drill rod statistics based on the initial position of the drilling rig and the size of the rod. Since drilling is a reciprocating motion, the number of reciprocating motions can be adjusted according to the working principle of the drilling rig to achieve drill rod statistics, which is universal.

[0041] 5. The present invention can obtain the actual operation scene of water exploration and release by constructing a target model and a counting composite model, and has the advantages of good versatility, high precision, and good robustness. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 A schematic flow chart of a method for counting water exploration and drainage drill rods in a coal mine provided by an embodiment of the present invention.

[0043] Figure 2 A schematic diagram of the structure of an image acquisition device provided in an embodiment of the present invention.

[0044] Figure 3 A schematic diagram of an image preprocessing process provided by an embodiment of the present invention.

[0045] Figure 4 A schematic diagram of a method for detecting and positioning a water exploration and release target provided in an embodiment of the present invention.

[0046] Figure 5 A schematic diagram of drilling rig target detection model training provided in an embodiment of the present invention.

[0047] Figure 6 A schematic diagram of drill rod counting provided by an embodiment of the present invention.

[0048] In the figure, 1 is the left side, 2 is the front, 3 is the side camera, 4 is the right side, 5 is the first feeding position, 6 is the rear camera, 7 is the drill rod, 8 is the drilling rig, and 9 is the second feeding position. DETAILED DESCRIPTION

[0049] In order to make the technical problems, technical solutions and beneficial effects to be solved by the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0050] like Figure 1 As shown in the figure, a method for counting water drilling rods in coal mines based on deep learning is shown in the figure. The specific steps are as follows:

[0051] Step S101, obtaining a downhole video stream including the drilling rig 8;

[0052] Step S102: pre-processing the acquired image based on OpenCV technology to obtain a clear drilling site image;

[0053] Step S103, using a pre-built target detection and positioning model to analyze the drilling site image to be detected, and determine the detection area;

[0054] Step S104: Calibrate the position information of the feeding device of the drilling rig 8 based on computer vision technology to obtain the initial position and the end position of the drilling rig 8 during its travel;

[0055] Step S105, using the pre-built counting model to track the feeding device in real time, and determining the number of rods fed according to its position information and calibration information.

[0056] In this embodiment, the water exploration and release drilling rod counting system includes: a video stream acquisition device, a key area detection and counting device, and a post-processing device.

[0057] Some embodiments and examples of the present disclosure are described in detail below with reference to the accompanying drawings.

[0058] Step S101: Acquire a video stream image containing the drilling rig 8, such as Figure 2 As shown. The camera is installed on the upper side or side side of the tunnel, the left side 1 is arranged vertically with the front 2, the side camera 3 is installed on the front 2, and the rear camera 6 is installed on the right side 4. The rear camera 6 and the side camera 3 are used to shoot the operation process of the drilling rig. The specific operation process is: the worker connects the drill rod 7 to the drill rig 8, the worker connects the drill rod 7 to the drill rig 8, and the feeding device drives the drill rod 7 into the borehole (here the drill rod 7 is on the drill rig 8) is the drilling process. After the drill rig 8 and the drill rod 7 are completely driven into the borehole, the worker removes the connection between the drill rig 8 and the current drill rod 7 (referred to as unloading the rod, at this time there is no drill rod 7 on the drill rig 8) to connect the next drill rod 7. The process is the unloading process, and so on, a drill rod 7 is driven into the borehole until the total length of the drill rod 7 driven reaches the set drilling depth.

[0059] In this embodiment, when the drill rig 8 is in the second feeding position 9, a start drilling instruction is sent through the mobile phone APP, and the algorithm starts the recognition work. When the drill rig 8 is in the first feeding device 5, an end drilling instruction is sent through the mobile phone APP, and the algorithm ends the recognition work.

[0060] Step S201: pre-processing the acquired image containing the decoded video stream of the drilling rig 8 to obtain an enhanced image.

[0061] In order to enhance the contrast of low-light areas, the Gamma correction function is used to enhance the brightness component of weak-light areas and reduce the brightness component of strong-light areas, thereby solving the problem of uneven illumination of the initial image. The Gamma function expression is as follows:

[0062] S Gamma (x,y)=S γ(x,y) (x,y)

[0063] Among them, S(x,y) is the original image, γ(x,y) is the correction parameter of the Gamma function, S Gamma (x,y) is the image after correction.

[0064] The Gamma value is usually 2.2, but there are multiple light source pixels in the coal mine environment, not only in areas with low illumination, but also in areas with high illumination intensity. If a unified Gamma value is used for correction, the image will be distorted. The correction parameter value of the Gamma function is adaptively adjusted according to the distribution characteristics of the illumination information of each pixel. The function expression of γ(x,y) is as follows:

[0065]

[0066] Among them, γ(x,y) is the correction parameter value of the Gamma function, I(x,y) is the illumination component, and I mean (x, y) is the average illumination component of the area, φ is the margin, I min (x, y) is the minimum illumination component, I max (x, y) is the highest illumination component. Using the γ(x, y) parameter function can prevent the gamma correction from enhancing the illumination component in high-brightness areas, and can better enhance areas with too low illumination components in the initial image.

[0067] The Gaussian filter function used by the Retinex algorithm takes the same operation on each pixel of the image, and cannot maintain the edge features of the image content. The present invention uses a filter algorithm that can maintain the edge to guide the filter to extract the reflection component. Through the guide image of the guide filter, the guide filter becomes an edge-maintaining filter operation, which can be used for noise reduction and detail smoothing. The specific expression of the guide filter function is:

[0068]

[0069] Among them, q is the output image, U is the guide image, a and b are the invariant coefficients of the function when the center of window ω is located at k. This method assumes that p and q are the results before filtering, and satisfies the minimum difference between p and q. According to the unconstrained image restoration method, it can be transformed into an optimization problem, and its value function is the formula:

[0070]

[0071] Among them, p i is the image pixel before denoising, ε is the residual value, a and b are the invariant coefficients of the function when the center of window ω is located at k, and the formula is solved by the least squares method:

[0072]

[0073] Among them, μ k represents the mean in the local window ω, U is the guide image, represents the variance in the local window ω, |ω| represents the number of pixels in the local window, and the window sampling operation is performed in the entire image. Finally, the final formula of the formula is obtained by taking the mean:

[0074]

[0075] in, ω k It refers to all windows containing pixel i, and k is the center position corresponding to different windows.

[0076] Step 301: Figure 3 , 4 As shown, the input image is subjected to detection processing to determine the drilling rig 8 area including the identification target.

[0077] The pre-built positioning model adopts the mainstream Yolo algorithm model, is end-to-end, and has the advantages of fast detection speed.

[0078] Before building the network, you need to collect and organize the images of the input model and convert them into JPG format. Then, use the labelimg tool to manually label the image data, and then output the label format in XML format. When training the network, you need to divide the data into training set and test set. Here, 80% of the data set is used as the training set of the network, and the remaining 20% ​​of the data set is used as the test set.

[0079] After completing the preprocessing of the data set, it is necessary to build the network model of the yolo algorithm and configure the running environment. Use pycharm as the IDE of the project, and use conda to import the downloaded pytorch environment into the project. After the environment configuration is completed, it is necessary to set the parameters of yolo and build the model. The yolo model mainly includes the Input input terminal, Backbone backbone network, Neck network layer, and Head output terminal. The input terminal processes the input image, and uses Mixup data enhancement, adaptive anchor box calculation, and adaptive image scaling to improve the speed of model training and the accuracy of the network; the Backbone network uses CBS structure and ELAP structure to build a neural network of image features; the Neck network uses FPN+PAN structure to enhance the network feature fusion capability; the Head uses EIOU as the loss function to predict image features and generate bounding boxes and category confidence.

[0080] The precision and recall are used as indicators to measure the model, and the formulas are:

[0081]

[0082] Among them, TP is the positive sample that is correctly identified, FP is the negative sample that is identified as a positive sample, and FN is the negative sample that is correctly identified. In order to solve the divergence problem of IOU during training, CIOU is used as the loss function of Boundingbox. EIOU takes into account the distance, overlap rate, scale and penalty factor between the target and the anchor, strengthens the stability of the target frame regression, and speeds up the convergence of the predicted frame and the real frame regression. The formula is as follows:

[0083]

[0084] Step 401: Obtain the position information of the detection area and calibrate it based on the pixel coordinate system, which mainly includes the following steps:

[0085] When the angles of the side camera 3 and the rear camera 6 are adjusted appropriately, and the workers have prepared the preliminary work such as adjusting the position and angle of the drill rig 8 appropriately, the drill rig 8 is in an unloaded state, and when the feed device of the drill rig 8 is at the initial position, the mobile phone APP is used to record the coordinate information of the initial position; when the feed device of the drill rig 8 is at the final position, the mobile phone APP is used to record the coordinate information of the final position.

[0086] The coordinate information of the initial position and the final position mentioned above is stored and sent to the counting model through the background, so as to prepare for the next step of drilling rod 7 statistics.

[0087] Repeat this process and follow the above steps when drilling other holes.

[0088] Step 501: using the counting model to count the number and depth information of the drill rods 7, mainly including the following steps:

[0089] According to the target detection model described above, the position information of the feed device of the drilling rig 8 is obtained, and its position information is tracked and recorded. Its coordinate information is judged with the calibrated coordinate value. If the initial position and the end position appear in pairs, the number of drill rods 7 is +1.

[0090] In the process of determining the coordinate position, it is necessary to determine whether the target of the drill rod 7 appears or changes in size to determine whether it should be counted in the number of rods; if the rod is advanced, the size of the drill rod 7 gradually decreases; if the rod is withdrawn, the size of the drill rod 7 gradually increases, such as Figure 6 The peak represents the completion of rod advancement, the trough represents the completion of rod withdrawal, and the time between the peak and the trough represents the actual time of the drill rod travel.

[0091] In the embodiment, different types of coal mine water exploration drills have different reciprocating times for completing one feed rod, and the algorithm model can be dynamically adjusted according to the actual operation of the drill.

[0092] In the embodiment, the preprocessed image is detected, wherein the detection area is divided into six blocks, specifically: pneumatic drilling rig feeding device, pneumatic drilling rig drill rod, hydraulic drilling rig feeding device, hydraulic drilling rig drill rod, crawler drilling rig feeding device, crawler drilling rig drill rod. The feeding device is tracked and counted to obtain relevant information of the drill rod 7 based on computer vision recognition technology.

[0093] In an embodiment, Figure 5 As shown in the figure, the intelligent processing platform receives the water exploration video stream through the streaming media server, performs image preprocessing on it, obtains clear and high-quality images, and loads the pre-trained model to extract its features to obtain the detected target and location information. The optimal model is obtained by repeatedly verifying the training results and adjusting the hyperparameters.

[0094] In an embodiment, by analyzing the video stream images of the drilling site, the scale of the water exploration and release drilling can be recorded and counted.

[0095] 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 and improvements made within the spirit and principle of the present invention should be included in the scope of the present invention.

Claims

1. A method for counting water drilling rods in coal mines based on deep learning, characterized in that: The specific steps are as follows: Step 1: Obtain downhole drilling video stream; Step 2: converting the downhole drilling video stream as the input video stream into a drilling rig image for preprocessing to obtain an enhanced image; Step 3: Analyze the drilling rig image using the pre-built target detection model to determine the drilling rig detection area for the identification target, thereby obtaining the location information of the drilling rig; Step 4: calibrate the drilling rig position information to obtain the starting position and ending position of the drilling rig feeding device; Step 5: Use the pre-built counting model to analyze the position information of the drilling rig feeding device, count the rod feeding and rod unloading according to the calibrated position, and obtain the water exploration depth of the current operation.

2. A method for counting coal mine water exploration and drainage drill rods based on deep learning according to claim 1, characterized in that: In step one, the specific acquisition steps are: adjust the camera position to the specified position, adjust the drill rig position and angle to the design requirements, send a drilling start command, and when a single hole is completed, send an end command. The AI ​​analysis module encodes and decodes the video stream to obtain a valid drilling video stream.

3. A method for counting coal mine water exploration and drainage drill rods based on deep learning according to claim 2, characterized in that: In step 2, the specific image preprocessing steps are: decoding the acquired video stream to obtain the original image and enhancing it to obtain an enhanced image, and performing denoising on the enhanced image; Specific image enhancement methods include color space conversion, Gamma correction and CLAHE enhancement; Specific denoising methods include Gaussian filtering, median filtering and histogram equalization.

4. A method for counting coal mine water exploration and drainage drill rods based on deep learning according to claim 3, characterized in that: The RGB color space of the original image is converted into the YCbCr color space, the Y brightness component of the YCbCr color space is extracted for adaptive equalization with limited contrast, and then converted back to the RGB color space to obtain an enhanced image.

5. A method for counting coal mine water exploration and drainage drill rods based on deep learning according to claim 4, characterized in that: In step three, the specific steps of the target detection model construction method are: receiving manually labeled drilling rig samples, the manually labeled objects at least include the feeding device and the drill rod in the drilling rig image samples; using the labeled drilling rig image samples to train the target detection model, and obtaining the optimal target detection model as the final target detection model.

6. A method for counting coal mine water exploration and drainage drill rods based on deep learning according to claim 5, characterized in that: The target drilling rig detection area divides the drilling rig into six areas: pneumatic drilling rig feeding device area, pneumatic drilling rig drill rod area, hydraulic drilling rig feeding device area, hydraulic drilling rig drill rod area, crawler drilling rig feeding device area, and crawler drilling rig drill rod area.

7. A method for counting coal mine water exploration and drainage drill rods based on deep learning according to claim 6, characterized in that: Before manual drilling, the drilling rig feed device is reciprocated once in an unloaded state, and the AI ​​analysis module calibrates the position of the feed device based on the pixel coordinate system to obtain the initial position and end position of the feed device.

8. A method for counting coal mine water exploration and drainage drill rods based on deep learning according to claim 7, characterized in that: During the drilling process of the drilling rig, the position information of the feed device is tracked in real time, and whether there is a drill rod is detected; when the feed device is in the initial position, it is recorded as 1. During the movement of the drilling rig, it is detected that the drill rod has always existed. When the feed device is in the terminal position, it is recorded as 1, and the number of drill rods is +1. This process is repeated to count the drill rods.

9. A method for counting coal mine water exploration and drainage drill rods based on deep learning according to claim 3, characterized in that: The Gamma correction function is used to enhance the brightness component of the weak light area and reduce the brightness component of the strong light area. The Gamma function expression is as follows: S Gamma (x,y)=S γ(x,y) (x,y) Among them, S(x,y) is the original image, γ(x,y) is the correction parameter of the Gamma function, S Gamma (x, y) is the image after correction; The correction parameter value of the Gamma function is adaptively adjusted according to the distribution characteristics of the illumination information of each pixel. The function expression of γ(x,y) is as follows: Among them, γ(x,y) is the correction parameter value of the Gamma function, I(x,y) is the illumination component, and I mean (x, y) is the average illumination component of the area, φ is the margin, I min (x, y) is the minimum illumination component, I max (x,y) is the highest illumination component.

10. A method for counting coal mine water exploration and drainage drill rods based on deep learning according to claim 9, characterized in that: The edge-preserving filtering algorithm guides filtering to extract the reflection component. The specific expression of the guided filtering function is: Among them, q is the output image, U is the guide image, a and b are the invariant coefficients of the function when the center of window ω is located at k. According to the unconstrained image restoration method, it is transformed into an optimization problem, and its value function is: Among them, p i is the image pixel before denoising, a and b are the invariant coefficients of the function when the center of window ω is located at k, ε is the weight value, and the formula is solved using the least squares method: Among them, μ k represents the mean in the local window ω, U is the guide image, represents the variance in the local window ω, |ω| represents the number of pixels in the local window, and the window sampling operation is taken in the entire image. The final formula for taking the average value is: in, ω k It refers to all windows containing pixel i, and k is the center position corresponding to different windows.

Images