Gangue content detection method and system based on coal gangue depth estimation in coal caving process

Through the coal gangue image defogging treatment and depth value estimation model, the problem of poor coal gangue image quality in harsh underground environments is solved, and the accurate calculation of the gangue content rate during the top coal is realized, reducing resource waste.

CN119006333APending Publication Date: 2024-11-22CHINA UNIV OF MINING & TECH (BEIJING)
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202410792133.0
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-06-19
Publication Date
2024-11-22

AI Technical Summary

Technical Problem

During the process of putting the top coal, the harsh environment and poor lighting conditions underground lead to poor quality of coal gangue images, which affects the accurate calculation of gangue content. Moreover, the depth of coal gangue is large, making it difficult to accurately calculate the gangue content, resulting in waste of resources.

Method used

The coal gangue image defogging treatment and depth value estimation model are used, and the depth value estimation model of local attention enhancement and bar window random field is estimated by image segmentation, combining the fixed threshold method to calculate the coal gangue depth and gangue containing Rate.

Benefits of technology

The accurate calculation of the coal gangue rate during the top coal release process is achieved, the image clarity and depth estimation accuracy are improved, and resource waste is reduced.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119006333A_ABST
    Figure CN119006333A_ABST
Patent Text Reader

Abstract

The invention relates to a gangue content detection method and system based on coal gangue depth estimation in a coal caving process, and the method comprises the steps: carrying out the defogging of a coal gangue image of a coal caving working face, and obtaining a defogged and enhanced image in the coal caving process; the coal gangue image of the coal caving face is an image with water mist acquired at a rear scraper conveyor of an underground fully mechanized caving face by using an industrial explosion-proof camera; training a coal gangue image depth value estimation model, and detecting the defogged and enhanced image in the coal caving process by using the model to obtain a coal gangue depth estimation result; and performing image segmentation on the defogged and enhanced image in the coal caving process by adopting a fixed threshold method to obtain a boundary segmentation result of the coal and gangue blocks, calculating the form of the coal gangue conveyed by the scraper according to the coal gangue depth estimation result and the boundary segmentation result, and estimating the gangue content according to the form. According to the method, firstly, defogging enhancement is performed on a coal gangue image of an underground coal caving working face, depth values of coal and gangue need to be calculated in order to accurately obtain the mass of coal and gangue, and a coal gangue depth value estimation algorithm based on local attention enhancement and a strip-shaped window random field is designed in order to solve the problems that the overall depth of a coal caving scene is small and the individual depth difference of coal and gangue is large. And detecting the defogged and enhanced image in the coal caving process by using a model to obtain a coal and gangue depth estimation result, segmenting the defogged image to obtain a coal and gangue boundary segmentation result, and calculating the gangue content according to the depth estimation result and the boundary segmentation result to realize accurate calculation of the coal gangue content in the top coal caving process.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of coal gangue detection, and in particular to a method and system for detecting gangue content based on estimation of coal gangue depth during coal caving. Background Art

[0002] The multi-round top coal caving process is an important means of full-height mining in the process of mining thick coal seams. After a long period of research and development, the detection and calculation technology of the gangue content of top coal caving has become increasingly mature. However, there are still the following problems in the calculation of the gangue content in the top coal caving process. First, the production environment of the underground top coal caving process is harsh. Coal mining machines and coal cutters will generate a lot of dust during the production process. In order to remove the interference of dust, the production process of coal washing and dust reduction will spray a lot of water mist on the gangue to reduce dust, which in turn leads to uneven water mist concentration in the environment, interfering with the quality of the collected images. The images collected underground will be seriously polluted by water mist due to these factors. At the same time, the lighting conditions underground are poor, the visibility is low, and the color and texture effects of the gangue are not easy to distinguish. Second, in the process of top coal caving, because the top coal and gangue are spontaneously broken and fall due to the gravity of themselves and the upper coal and rock layers, the size and depth of the fallen coal and gangue are different, some are 2-3 meters thick, and some are only tens of centimeters thick. However, the area and perimeter they may occupy in the results of gangue detection and segmentation are relatively small. If the depth value of the gangue cannot be accurately calculated, the gangue content will greatly deviate from the true value, resulting in a waste of mineral resources. All of the above problems have an impact on how to accurately calculate the gangue content. Summary of the invention

[0003] The purpose of the present invention is to provide a method and system for detecting gangue content in the coal caving process based on estimation of gangue depth, which can accurately calculate the gangue content in the top coal caving process.

[0004] To achieve the above object, the present invention provides the following solutions:

[0005] A method for detecting gangue content in a coal caving process based on estimation of gangue depth comprises:

[0006] The gangue image of the coal caving working face is defogged to obtain a defogged and enhanced image of the coal caving process; the gangue image of the coal caving working face is a water mist image collected at the rear scraper of the underground fully mechanized caving working face using an industrial explosion-proof camera;

[0007] Training a coal gangue image depth value estimation model, using the model to detect the defogging and enhanced image of the coal caving process, and obtaining a coal gangue depth estimation result;

[0008] The fixed threshold method is used to implement image segmentation on the dehazed and enhanced image of the coal discharge process to obtain the boundary segmentation result of the coal and gangue blocks. According to the coal gangue depth estimation result and the boundary segmentation result, the coal gangue morphology conveyed by the scraper conveyor is calculated, and the gangue content is estimated accordingly.

[0009] Optionally, the defogging process is performed on the coal gangue image of the coal caving working face to obtain the defogging and enhanced image, which specifically includes:

[0010] Decomposing the foggy image according to the brightness difference in the grayscale space to obtain a high water fog area image and other low water fog area images;

[0011] Decomposing the high water fog area image by wavelet transform to obtain low-frequency fog component and high-frequency detail component;

[0012] The low-frequency fog component is processed by using a Retinex algorithm of a multi-scale Wiener filter to obtain a processed low-frequency fog component;

[0013] Processing the high-frequency detail components by using guided filtering to obtain processed high-frequency detail components;

[0014] Combining the processed low-frequency fog component and high-frequency detail component by inverse wavelet transform to obtain a combined image;

[0015] The combined image is processed by Gamma transformation with brightness adaptive restoration to obtain an enhanced image of a high water fog area;

[0016] Performing contrast-limited adaptive histogram equalization processing on the low water fog area image to obtain a low water fog area enhanced image;

[0017] The high water fog area enhanced image and the low water fog area enhanced image are merged by using median filtering and convolution operation to obtain a final defogging enhanced image.

[0018] Optionally, the gangue depth value estimation model includes an encoder and a decoder:

[0019] The encoder is a transformer encoder module with enhanced local attention, in which a position encoding scheme with enhanced local position attention is added;

[0020] The decoder is a decoder module based on a strip window fully connected conditional random field, in which a binary potential calculation mechanism is combined with a set of detection heads in vertical and horizontal strip windows.

[0021] Optionally, the image after defogging and enhancement during coal discharge is segmented using a fixed threshold method to obtain a boundary segmentation result of coal and gangue blocks, and the shape of the coal gangue transported by the scraper is calculated according to the coal gangue depth estimation result and the boundary segmentation result, and the gangue content is estimated accordingly, specifically including:

[0022] For the dehazed and enhanced image, a threshold is determined by using a histogram bimodal method;

[0023] According to the threshold value, a fixed threshold value method is used to implement image segmentation, and the boundary segmentation result of the coal and gangue blocks is obtained.

[0024] Calculating the volume of the gangue ore according to the gangue depth estimation result and the boundary segmentation result;

[0025] The gangue content is calculated based on the volume of the gangue ore.

[0026] A gangue content detection system based on gangue depth estimation in coal caving process, specifically comprising:

[0027] The camera login module is used to obtain the image information collected by the underground explosion-proof camera. When using the camera opening function in the function bar, you need to enter the camera's IP and device port number first. After clicking login, different field values ​​will be passed to the background. Then, when the camera is opened, these field information will be used to obtain the camera's image information through IP access;

[0028] A gangue ore depth value estimation module is used to estimate the depth value of the gangue ore through a gangue depth value estimation model to obtain the depth information of the gangue ore;

[0029] The gangue identification and segmentation module is used to display the gangue segmentation result image and obtain the area of ​​each closed area after the gangue segmentation;

[0030] An underground hydraulic support control module is used to control the opening and closing of the top coal caving opening of the underground hydraulic support;

[0031] The gangue content calculation module is used to calculate the gangue content in the screen in real time.

[0032] Optionally, the gangue content calculation module specifically includes:

[0033] The first calculation unit converts the depth value estimation map into a gangue ore thickness map, and then superimposes the gangue segmentation result with the gangue thickness result to obtain the gangue ore cross-sectional area and the gangue ore thickness information;

[0034] The second calculation unit calculates the gangue content in the picture in real time according to the cross-sectional area and thickness information of the gangue ore.

[0035] According to the specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0036] The present invention discloses a method and system for detecting gangue content in a coal caving process based on estimation of gangue depth. The method comprises: defogging a gangue image of a coal caving working face to obtain a defogging and enhanced image; the gangue image of the coal caving working face is an image with water mist collected at a scraper behind a fully mechanized caving working face in an underground well by using an industrial explosion-proof camera; training a model for estimating the depth value of a gangue image, detecting the defogging and enhanced image of the coal caving process with the model to obtain a gangue depth estimation result; segmenting the defogging and enhanced image of the coal caving process using a fixed threshold method to obtain a boundary segmentation result of coal and gangue blocks, calculating the shape of the gangue transported by the scraper according to the gangue depth estimation result and the boundary segmentation result, and estimating the gangue content accordingly. The present invention first performs dehazing and enhancement on the gangue image of the underground coal caving working face. To accurately obtain the quality of coal and gangue, it is necessary to calculate their depth values. In order to solve the problem that the overall depth of the coal caving scene is small and the individual depth differences of gangue are large, a gangue depth value estimation algorithm based on local attention enhancement and strip window random field is designed. The dehazing and enhanced image of the coal caving process is detected by the model to obtain the gangue depth estimation result, and the dehazing image is segmented to obtain the gangue boundary segmentation result. Then, the gangue content is calculated according to the depth estimation result and the boundary segmentation result, so as to realize the accurate calculation of the gangue content of coal in the process of top coal caving. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0038] Figure 1 A flow chart of a method for estimating the gangue depth value and detecting the gangue content in the top coal caving process provided in Example 1 of the present invention;

[0039] Figure 2 A flow chart of coal gangue image defogging provided in Example 1 of the present invention;

[0040] Figure 3 A schematic diagram of coal gangue image defogging provided in Example 1 of the present invention;

[0041] Figure 4 A structural diagram of a depth value estimation model provided in Example 1 of the present invention;

[0042] FIG5( a) is a structural diagram of a transformer encoder module based on local attention enhancement provided in Example 1 of the present invention;

[0043] FIG5( b ) is a structural diagram of a decoder module based on a strip-window fully connected conditional random field according to Embodiment 1 of the present invention;

[0044] Figure 6 A schematic diagram of depth value estimation results provided in Example 1 of the present invention;

[0045] Figure 7 This is a schematic diagram of the coal gangue image segmentation results provided in Example 1 of the present invention. DETAILED DESCRIPTION

[0046] In order to clearly and completely describe the technical solutions in the embodiments of the present invention in combination with the drawings in the embodiments of the present invention, it is obvious that the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0047] The purpose of the present invention is to provide a gangue content detection method and system based on gangue depth estimation in the coal caving process, which can accurately calculate the mass of gangue in the process of top coal caving.

[0048] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0049] Example 1

[0050] Figure 1 The flowchart of the gangue content detection method based on the estimation of the gangue depth value in the top coal caving process provided in Example 1 of the present invention is as follows: Figure 1 As shown, the method includes:

[0051] Step 101: Defogging the coal gangue image of the coal caving working face to obtain a defogging enhanced image; the coal gangue image of the coal caving working face is an image with water mist collected at the rear scraper of the underground fully mechanized caving working face using an industrial explosion-proof camera. Step 101 specifically includes:

[0052] Step 1011: Decompose the foggy image according to the brightness difference in the grayscale space to obtain a high water fog area image and other low water fog area images.

[0053] Step 1012: Decompose the high water fog area image using wavelet transform to obtain low-frequency fog components and high-frequency detail components.

[0054] Step 1013: Process the low-frequency fog component using a Retinex algorithm of a multi-scale Wiener filter to obtain a processed low-frequency fog component.

[0055] Step 1014: Process the high-frequency detail components using guided filtering to obtain processed high-frequency detail components.

[0056] Step 1015: The processed low-frequency fog component and high-frequency detail component are combined by inverse wavelet transform to obtain a combined image.

[0057] Step 1016: Process the merged image using Gamma transformation with adaptive brightness restoration to obtain an enhanced image of a high water fog region.

[0058] Step 1017: performing contrast-limited adaptive histogram equalization processing on the low water mist region image to obtain a low water mist region enhanced image.

[0059] Step 1018: For the high water fog area enhanced image and the low water fog area enhanced image, the high and low water fog areas are merged by using median filtering and convolution operation to obtain a final defogging enhanced image.

[0060] Step 102: Using the trained gangue image depth value estimation model, the model is used to detect the defogging and enhanced image during the coal caving process to obtain a gangue depth estimation result. In this embodiment, the gangue depth value estimation model includes an encoder and a decoder.

[0061] The encoder is a transformer encoder module with enhanced local attention, in which a position encoding scheme with enhanced local position attention is added;

[0062] The decoder is a decoder module based on a strip window fully connected conditional random field, in which a binary potential calculation mechanism is combined with a set of detection heads in vertical and horizontal strip windows.

[0063] Step 103: Use a fixed threshold method to segment the image after defogging and enhancement during the coal discharge process to obtain the boundary segmentation result of the coal and gangue blocks, calculate the shape of the coal gangue transported by the scraper according to the coal gangue depth estimation result and the boundary segmentation result, and estimate the gangue content accordingly. Step 103 specifically includes:

[0064] Step 1031: for the dehazed and enhanced image, a threshold is determined using a histogram bimodal method.

[0065] Step 1032: Based on the threshold, a fixed threshold method is used to implement image segmentation to obtain boundary segmentation results of coal and gangue blocks.

[0066] Step 1033: Calculate the volume of the gangue ore according to the gangue depth estimation result and the boundary segmentation result;

[0067] Step 1034: Calculate the gangue content according to the volume of the gangue ore.

[0068] The principle of the present invention is specifically described below:

[0069] Step 1: Dehazing and enhancing coal gangue images based on water mist region decomposition

[0070] The explosion-proof camera and the tunnel light are installed side by side at the rear tail beam of the underground hydraulic support to form a supporting equipment to collect the coal gangue image of the underground rear scraper conveyor. Due to the extremely harsh underground coal mining environment, the collected images often have low illumination, low clarity, high fog area, and low contrast. Many existing defogging and enhancement algorithms generally have the problem of weak processing of high water fog areas and over-processing of low water fog areas because they perform the same processing transformation on the entire image. Therefore, the present invention adopts a method based on water fog area decomposition to defog and enhance coal gangue images. Figure 2 This is a flow chart of coal gangue image defogging provided in Example 1 of the present invention. The process is as follows:

[0071] 1. Image mist region decomposition

[0072] First, the color image is converted into grayscale space. Then, according to the grayscale value of each pixel, a threshold β is set to divide the converted grayscale image and its corresponding color image into two areas, one is the high fog area Iwhite which is seriously polluted by fog, and the other is the low fog area Igray which needs image enhancement. The mathematical expression of the division scheme of the two areas is as follows:

[0073]

[0074] Wherein: Vgray(x, y) is the corresponding value image matrix of the color RGB image converted to the grayscale space; the value range of β is between [0, 1], and β is set to 0.4 as an empirical value in the present invention;

[0075] 2. Defogging enhancement in high fog areas based on frequency domain decomposition

[0076] The image is converted from a color image RGB three-channel grayscale image, so that the image becomes a one-dimensional image distributed between 0-255, and then the image is converted into frequency domain space by Fourier transform.

[0077] The image is decomposed into high-frequency components and low-frequency components in the frequency domain space using two-dimensional discrete wavelet transform, obtaining a low-frequency fog component of the image and three high-frequency detail components, namely horizontal detail component, diagonal detail component and vertical detail component.

[0078] 3. Retinex algorithm based on multi-scale Wiener filtering to process low-frequency fog components

[0079] Based on the principle of multi-scale Retinex algorithm, Wiener filters of different scales are taken for weighted fusion to obtain the processing result image. This method overcomes the problems of excessive brightness difference, large color brightness change, poor continuity between image pixels, etc., and has a good defogging effect on low-frequency foggy components.

[0080] 4. Denoising high-frequency detail components based on guided filtering

[0081] The window generated with each pixel as the center is used as the guiding image, so that the edge information of the image after guided filtering is roughly consistent with the guiding map. The high-frequency component not only preserves certain detail information, but also removes a large number of noise points in the image, achieving the purpose of denoising the edge details of the original image.

[0082] 5. Gamma transformation for adaptive brightness restoration

[0083] Through the inverse wavelet function, the processed low-frequency components and high-frequency components are reorganized in the frequency domain to obtain a synthetic image, which is then processed by adaptive Gamma transformation to further suppress the foggy area and enhance the contrast of the overall image, making the coal gangue ore originally covered by water mist appear clearer. The mathematical expression of this method is as follows:

[0084]

[0085] Where: Iwhite represents the number of pixels in the high fog area; Igray represents the number of pixels in the low fog area; D is the scale parameter.

[0086] 6. Merging high and low water mist areas of coal gangue images

[0087] The low fog area is enhanced by using the contrast-limited adaptive histogram averaging method, and then the high and low fog areas are merged. In the merging process, median filtering and convolution operations are added to make the fusion of the boundary area more harmonious, and the defogging and enhanced image is obtained. Figure 3 Schematic diagram of coal gangue image defogging provided in Example 1 of the present invention.

[0088] Step 2: Coal gangue depth estimation model based on local attention enhancement and strip window random field

[0089] The realization of intelligent top coal caving technology in coal mine fully mechanized caving working face relies on accurately calculating the gangue content of coal on the rear scraper conveyor, that is, accurately calculating the quality of gangue and coal blocks. In order to obtain the quality of gangue ore, it is necessary to first obtain the depth value of gangue ore through various means of computer vision. The present invention designs a gangue depth value estimation model based on local attention enhancement and strip window random field. Figure 4This is a structural diagram of a depth value estimation model provided in Example 1 of the present invention; Figure 4 As shown in the figure, the input image of the network is firstly subjected to feature extraction by the improved shift window transformer. The feature extraction is carried out in four layers, corresponding to the input image The corresponding improved strip window fully connected conditional random field decoder is also designed as a four-layer structure based on these four levels and resolutions, with 4, 8, 16, and 32 detection heads in each layer. The top pyramid pooling layer can aggregate feature information from the entire image and estimate the initial label X based on the features extracted from the fourth layer. Then, the improved strip window fully connected conditional random field decoder in the fourth layer is fed into the unary potential network training to obtain the unary potential The Q and K in the figure are the same as the self-attention calculation module, which are obtained by multiplying the input feature matrix by the corresponding coefficients. P is the relative position encoding matrix. The corresponding parameters of the binary potential are obtained by softmax operation and then multiplied by the estimated label X to obtain the binary potential. Then, the energy function E estimated by multiple detection heads is passed through the fusion network to obtain the estimation result X' of this layer. At the same time, X' is passed to the next layer as the estimation result of the upper layer and the same operation is performed again as X. After completing four layers, the final depth value map is obtained.

[0090] FIG5(a) is a structural diagram of a transformer encoder module based on local attention enhancement provided in Example 1 of the present invention, and FIG5(b) is a structural diagram of a decoder module based on a strip window fully connected conditional random field provided in Example 1 of the present invention; as shown in FIG5(a), the encoder mainly includes four stages, each of which is composed of a linear embedding layer or a patch merging layer and multiple shift window Transformer modules. Before entering these four stages, the input image needs to pass through a patch splitting module. The encoder module has a certain enhancement in the feature calculation effect of different positions of a single target, which is closer to the coloring effect of the real depth value collected. As shown in FIG5(b), the binary potential calculation mechanism combined with the improved vertical and horizontal strip window detection head increases the long-range field of view of the energy function of the fully connected conditional random field, making the global depth perception more accurate, the global depth performance richer and closer to reality, the depth value comparison between the gangue ore farther away from the depth camera and the gangue ore closer to the depth camera is more obvious, and the long-range information of the estimated map is richer. Figure 6 This is a schematic diagram of the depth value estimation result provided in Example 1 of the present invention.

[0091] The depth value estimation model in the present invention has the following advantages:

[0092] At the encoder level, the original method has a weak ability to extract the relative position information of samples, that is, different arrangement orders of sample points will obtain the same extracted features in the same perception window. The present invention proposes an improved local attention enhanced position encoding strategy that is decoupled from the self-attention mechanism, so that the depth value changes of different local areas of individual gangue ore are more real and accurate.

[0093] At the decoder level, the energy potential function of each pixel in the original neural window fully connected conditional random field is calculated using a window interaction method, which has a small receptive field and cannot fully calculate the binary potential information. The present invention proposes a binary potential calculation strategy that combines improved vertical and horizontal strip window detection heads, which makes the global depth perception more accurate and increases the long-range field of view of the neural fully connected conditional random field.

[0094] Step 3: Coal gangue image segmentation based on fixed threshold

[0095] The gray value difference of the coal gangue color image converted into gray space is used as the detection and identification index of coal block and gangue, the fixed threshold method is used to realize the coal gangue segmentation, and the histogram bimodal method is used to determine the threshold to realize the detection of coal gangue and obtain the coal gangue segmentation result. Figure 7 This is a schematic diagram of the coal gangue segmentation results provided in Example 1 of the present invention.

[0096] Step 4: 3D modeling of coal gangue ore and calculation of gangue content

[0097] The depth value image obtained by superimposing the gangue segmentation image with the depth value image of the gangue ore image is the depth value image of the gangue. The average depth value of the gangue depth value image is taken to get the average depth value of the gangue in the current picture. The sum of the areas of the closed areas composed of pixels with a grayscale value of 255 in the gangue segmentation image is the total cross-sectional area of ​​the gangue. After obtaining the data required for modeling, the system performs three-dimensional modeling of the gangue ore. A piece of gangue ore is approximated as a right prism with a polygonal cross-section. The volume calculation formula of the right prism is the bottom area multiplied by the length of the prism, that is, the cross-sectional area obtained by gangue segmentation multiplied by the average thickness of the gangue ore. The calculation formula for the gangue content is:

[0098]

[0099] Where: 煤 , ρ 矸 They represent the density of coal mine and gangue respectively; l and p represent the thickness and cross-sectional area of ​​gangue; q and d represent the thickness and cross-sectional area of ​​coal block.

[0100] Example 2

[0101] This embodiment provides a gangue content detection system based on gangue depth estimation during coal caving process, including:

[0102] The camera login module is used to obtain the image information collected by the underground explosion-proof camera. When using the camera opening function in the function bar, you need to enter the camera's IP and device port number first. After clicking login, different field values ​​will be passed to the background. Then, when the camera is opened, these field information will be used to obtain the camera's image information through IP access;

[0103] A gangue ore depth value estimation module is used to estimate the depth value of the gangue ore through a gangue depth value estimation model to obtain the depth information of the gangue ore;

[0104] The gangue identification and segmentation module is used to display the gangue segmentation result image and obtain the area of ​​each closed area after the gangue segmentation;

[0105] An underground hydraulic support control module is used to control the opening and closing of the top coal caving opening of the underground hydraulic support;

[0106] The gangue content calculation module is used to calculate the gangue content in the screen in real time.

[0107] In this embodiment, the gangue content calculation module specifically includes:

[0108] The first calculation unit converts the depth value estimation map into a gangue ore thickness map, and then superimposes the gangue segmentation result with the gangue thickness result to obtain the gangue ore cross-sectional area and the gangue ore thickness information;

[0109] The second calculation unit calculates the gangue content in the picture in real time according to the cross-sectional area and thickness information of the gangue ore.

[0110] According to the specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0111] (1) The present invention proposes a coal gangue image defogging and enhancement algorithm for coal caving working face, and a new ore image defogging process idea is proposed in this algorithm. The coal gangue ore image is partitioned by regional decomposition. The high water fog area focuses on removing water fog, and the low water fog area focuses on image enhancement. It also integrates multiple methods such as frequency domain decomposition, wavelet change regional decomposition and merging; on the basis of the first point of regional division, the division result is used to innovate the Gamma change based on brightness adaptive recovery, which can adjust the suppression intensity according to the degree of water fog pollution in the image, and enhance the overall robustness and universality of the algorithm; a Retinex algorithm based on multi-scale Wiener filtering is also proposed. The Wiener filter can effectively extract the low-frequency fog component in the image, which is beneficial for our removal. At the same time, the Wiener filter does not need to be linearized, and the application scenario is more extensive. The processed picture changes are smooth.

[0112] (2) The present invention proposes a coal gangue depth value estimation algorithm based on local attention enhancement and strip window random field. In this algorithm, at the encoder level, the transformer encoder module based on local attention enhancement applies local attention enhancement position encoding to the position information between local adjacent patches and directly adds it to the calculation result of the self-attention score, thereby adding the position information of the adjacent patches, thereby enhancing the feature calculation effect of the local position of a single target to a certain extent. At the decoder level, the decoder module based on the strip window fully connected conditional random field adopts a binary potential calculation strategy combining vertical and horizontal strip window detection heads. The detection heads in the vertical and horizontal directions are calculated in parallel, which increases the long-range field of view of the energy function of the fully connected conditional random field, making the global depth perception more accurate.

[0013] (3) The present invention proposes a rapid estimation system for the depth value of coal gangue and detection of gangue content. In the system software, the coal gangue image defogging enhancement algorithm and the coal gangue depth value prediction algorithm are comprehensively used, the fixed threshold method is used for coal gangue segmentation, a three-dimensional ore model of coal gangue is established to realize rapid calculation of gangue content, and the modbus-tcp protocol is used to realize interactive control with the underground hydraulic support. In terms of system hardware, a complete hardware architecture for the underground and aboveground is constructed.

[0114] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the system disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.

[0115] The principles and implementation methods of the present invention are described in this article using specific examples. The description of the above embodiments is only used to help understand the method and core idea of ​​the present invention. At the same time, for those skilled in the art, according to the idea of ​​the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting the present invention.

Claims

1. A method for detecting gangue content in coal caving process based on estimation of gangue depth. It is characterized in that include: Dehazing the gangue image of the coal caving working face to obtain the dehazing and enhanced image of the coal caving process; The coal gangue image of the coal caving working face is a water mist image collected by an industrial explosion-proof camera at the rear scraper of the underground fully mechanized caving working face; Training a coal gangue image depth value estimation model, using the model to detect the defogging and enhanced image of the coal caving process, and obtaining a coal gangue depth estimation result; The fixed threshold method is used to implement image segmentation on the dehazed and enhanced image of the coal discharge process to obtain the boundary segmentation result of the coal and gangue blocks. According to the coal gangue depth estimation result and the boundary segmentation result, the coal gangue morphology conveyed by the scraper conveyor is calculated, and the gangue content is estimated accordingly.

2. A method for detecting gangue content in a coal caving process based on estimation of gangue depth according to claim 1, It is characterized in that The defogging process is performed on the coal gangue image of the coal caving working face to obtain the defogging enhanced image, specifically including: Decomposing the foggy image according to the brightness difference in the grayscale space to obtain a high water fog area image and other low water fog area images; Decomposing the high water fog area image by wavelet transform to obtain low-frequency fog component and high-frequency detail component; The low-frequency fog component is processed by using a Retinex algorithm of a multi-scale Wiener filter to obtain a processed low-frequency fog component; Processing the high-frequency detail components by using guided filtering to obtain processed high-frequency detail components; Combining the processed low-frequency fog component and high-frequency detail component by inverse wavelet transform to obtain a combined image; The combined image is processed by Gamma transformation with brightness adaptive restoration to obtain an enhanced image of a high water fog area; Performing contrast-limited adaptive histogram equalization processing on the low water fog area image to obtain a low water fog area enhanced image; The high water fog area enhanced image and the low water fog area enhanced image are merged by using median filtering and convolution operation to obtain a final defogging enhanced image.

3. The method for detecting gangue content in a coal caving process based on estimation of gangue depth according to claim 1, It is characterized in that The gangue depth value estimation model includes an encoder and a decoder: The encoder is a transformer encoder module with enhanced local attention, in which a position encoding scheme with enhanced local position attention is added; The decoder is a decoder module based on a strip window fully connected conditional random field, in which a binary potential calculation mechanism is combined with a set of detection heads in vertical and horizontal strip windows.

4. The method for detecting gangue content in a coal caving process based on estimation of gangue depth according to claim 1, It is characterized in that The fixed threshold method is used to segment the image after defogging and enhancement during the coal discharge process to obtain the boundary segmentation result of the coal and gangue blocks. According to the coal gangue depth estimation result and the boundary segmentation result, the coal gangue shape conveyed by the scraper is calculated, and the gangue content is estimated accordingly, which specifically includes: For the dehazed and enhanced image, a threshold is determined by using a histogram bimodal method; According to the threshold value, a fixed threshold value method is used to implement image segmentation, and the boundary segmentation result of the coal and gangue blocks is obtained. Calculating the volume of the gangue ore according to the gangue depth estimation result and the boundary segmentation result; The gangue content is calculated based on the volume of the gangue ore.

5. A gangue content detection system based on gangue depth estimation during coal caving process, It is characterized in that include: The camera login module is used to obtain the image information collected by the underground explosion-proof camera. When using the camera opening function in the function bar, you need to enter the camera's IP and device port number first. After clicking login, different field values ​​will be passed to the background. Then, when the camera is opened, these field information will be used to obtain the camera's image information through IP access; A gangue ore depth value estimation module is used to estimate the depth value of the gangue ore through a gangue depth value estimation model to obtain the depth information of the gangue ore; The gangue identification and segmentation module is used to display the gangue segmentation result image and obtain the area of ​​each closed area after the gangue segmentation; An underground hydraulic support control module is used to control the opening and closing of the top coal caving opening of the underground hydraulic support; The gangue content calculation module is used to calculate the gangue content in the screen in real time.

6. A gangue content detection system based on gangue depth estimation in coal caving process according to claim 6, It is characterized in that The gangue content calculation module specifically includes: The first calculation unit converts the depth value estimation map into a gangue ore thickness map, and then superimposes the gangue segmentation result with the gangue thickness result to obtain the gangue ore cross-sectional area and the gangue ore thickness information; The second calculation unit calculates the gangue content in the picture in real time according to the cross-sectional area and thickness information of the gangue ore.

Citation Information

Patent Citations

  • Coal gangue mass calculation method and system in top coal caving process

    CN112446914A

  • IN2835MU2014A