Coal gangue boundary identification method based on depth image identification and adaptive inversion

Abstract

The invention relates to the technical field of coal mining, in particular to a coal gangue boundary recognition method based on depth image recognition and adaptive inversion. The method comprises the following steps: acquiring an RGB texture image and three-dimensional point cloud data of a top coal caving area through an anti-explosion depth camera, and synchronously acquiring lumpiness distribution in real time by using a laser lumpiness analyzer; fusing the texture features, the point cloud curvature and the lumpiness data by adopting an improved model, dynamically improving the recognition weight when a large coal concentrated area is detected, and outputting coal and gangue interface coordinates; constructing a release volume inversion model based on bulk solid mechanics, balancing release volume matching degree and gangue invasion constraint through an objective function, and adaptively adjusting inversion step length and frequency according to lumpiness distribution and interface moving speed; the opening degree is calculated in a segmented mode according to the vertical height from a coal gangue interface to a coal caving opening, an intermittent coal caving strategy is triggered in combination with lumpiness gradation, and meanwhile the dip angle of a linkage support is finely adjusted. The method effectively solves the industrial problems of low caving rate of end top coal and high gangue mixing rate.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of coal mining, in particular to a coal and gangue boundary identification method based on depth image recognition and adaptive inversion. BACKGROUND

[0002] In the top coal caving mining process of the fully mechanized caving face of a coal mine, the low top coal recovery rate and the high gangue mixing rate in the end area are the core problems that have long restricted the resource recovery efficiency, and the root cause lies in the following technical bottlenecks:

[0003] The existing technology mainly relies on visible light cameras or grayscale analysis. In the high-dust and low-illumination environment of the underground mine, the optical characteristics of coal and gangue are confused, which leads to serious degradation of recognition accuracy. The boundary morphology changes in real time during the top coal caving process, and the traditional static recognition model cannot track the dynamic boundary.

[0004] The coal caving strategy mainly relies on artificial experience, and the response lags behind the change of the top coal size distribution. The quantitative correlation between the size distribution, the caving body morphology and the gangue mixing rate has not been established, and the fixed threshold window strategy exacerbates the loss of top coal or the mixing of gangue.

[0005] The offline screening data lags behind for several hours, which cannot be fed back to the control system in real time, and the process optimization window period is lost. The above defects directly cause the end top coal recovery rate to be much lower than the theoretical value, the mixing of gangue exacerbates the equipment wear and tear, and the artificial intervention intensity is high, which seriously restricts the intelligent mining process. SUMMARY

[0006] (I) Technical problems solved

[0007] In view of the deficiencies of the prior art, the present application provides a coal and gangue boundary identification method based on depth image recognition and adaptive inversion, which solves the problem of inaccurate dynamic identification of the coal and gangue boundary.

[0008] (II) Technical solutions

[0009] In order to achieve the above purpose, the present application is implemented by the following technical solutions:

[0010] A coal and gangue boundary identification method based on depth image recognition and adaptive inversion, comprising the following steps:

[0011] S1. An explosion-proof depth camera deployed above the coal caving port is used to obtain the RGB texture image and three-dimensional point cloud data of the top coal caving area, and a laser size rapid analyzer is used to collect the top coal size distribution data in real time;

[0012] S2. An improved YOLOv8 model is used to fuse the RGB texture features, the point cloud edge curvature and the size distribution data to identify the coal and gangue boundary coordinates, wherein the improved YOLOv8 model is embedded with a coordinate attention (CA) module, and an EIoU loss function is used.

[0013] S3. Constructing the drawbody inversion model based on granular mechanics theory, the objective function is: where Ω is the drawbody shape parameter, δ i is the penalty coefficient of gangue invasion, A i is the overlapping area;

[0014] S4. Calculating the opening degree K according to the vertical height H from the coal-gangue interface to the drawgate, the formula is:

[0015]

[0016] and triggering the intermittent caving strategy based on the block size distribution.

[0017] Preferably, in step S1: the explosion-proof depth camera is installed 0.5 m±0.1 m above the drawgate, the pitch angle is 30°±5°, the frame rate is ≥50 fps, the point cloud accuracy is ±1 mm, and the laser block size rapid analyzer measures the block size distribution in the range of 5-500 mm through laser particle size sensing technology, and the sampling frequency is ≥10 Hz.

[0018] Preferably, the block size-visual feature fusion method in step S2 includes: when the laser block size analyzer detects that the block size of >300 mm accounts for >80% in a certain area, mark it as "large block coal area" in the three-dimensional point cloud, and increase the YOLOv8 detection weight coefficient of the marked area by 0.3 to preferentially identify the boundary of large block coal.

[0019] Preferably, the inversion model adaptive adjustment strategy of step S3 includes: if the >300 mm accounts for >50% in the real-time block size data, the inversion iteration step is adjusted from 0.1 m to 0.05 m; if the coal-gangue interface moving speed is >0.2 m / s, the inversion frequency is increased from 1 Hz to 5 Hz.

[0020] Preferably, the weight coefficient dynamic adjustment rule in step S3 is: when the real-time mixed gangue rate is >5%, increase ω2 to 1.5 times of the original value; when the top coal draw rate is <85%, increase ω2 to 1.2 times of the original value.

[0021] Preferably, the intermittent caving strategy of step S4 is specifically: when the block size analyzer shows that the block size of >200 mm accounts for >60%, start the intermittent caving cycle, i.e. open the drawgate for 20 seconds and close it for 10 seconds.

[0022] Preferably, it further includes hydraulic support posture adjustment: when the inversion shows that the drawbody center offset Δd is >0.3 m, calculate the support inclination adjustment θ = arctan(Δd / h), where h is the thickness of the top coal, and limit θ within the range of ±2°.

[0023] A coal and gangue boundary recognition method control system based on depth image recognition and adaptive inversion is implemented, comprising:

[0024] A data acquisition module, comprising an explosion-proof depth camera, a laser block size rapid analyzer, and a hydraulic support pressure sensor;

[0025] A processing module, comprising an industrial computer loaded with an improved YOLOv8 algorithm, which runs a coal and gangue boundary recognition and release body inversion program in real time;

[0026] An execution module, comprising a coal release opening opening and closing mechanism controlled by an electro-hydraulic proportional valve, and a hydraulic support inclination adjusting device.

[0027] (Three) beneficial effects

[0028] The present application realizes the following fundamental breakthroughs through the depth image multi-source fusion recognition and block size driven adaptive inversion technology chain:

[0029] I. Fusing RGB texture, three-dimensional point cloud edge curvature and real-time block size distribution data, constructing a multi-dimensional feature discrimination system, enhancing the coal and gangue discrimination ability under dust interference, capturing complex boundary morphology through block size feature dynamic weighting mechanism (such as increasing the detection weight of large coal area), establishing the real-time mapping relationship between block size distribution and inversion parameters, dynamically adjusting the step size and frequency according to the working condition, ensuring the accuracy of the release body shape prediction, and the linkage weight coefficient of the mixed gangue rate and the release rate feedback mechanism, dynamically balancing the resource recovery and gangue suppression target.

[0030] II. Based on the segmented opening degree regulation of coal and gangue interface height, effectively blocking the gangue mixing at the critical height, the intermittent coal release strategy triggered by block size matches the top coal flow characteristics, and the natural classification effect is used to reduce the mixing, the explosion-proof equipment and anti-interference algorithm design adapt to the harsh underground environment, the real-time block size analysis-recognition-inversion-control closed loop replaces the lagging artificial decision-making, and the labor intensity is greatly reduced. BRIEF DESCRIPTION OF DRAWINGS

[0031] Fig. 1 The system diagram of the present application;

[0032] Fig. 2 The comparison table of the technical effects of the embodiment of the present application and the conventional method;

[0033] Fig. 3 The adaptive control logic diagram of the present application based on block size feedback. DETAILED DESCRIPTION

[0034] With reference to the drawings of the embodiments of the present application, the technical solutions in the embodiments of the present application will be described clearly and completely. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present application.

[0035] With reference to Figs. 1-3 The present application provides a technical solution: a coal and gangue boundary identification method based on depth image recognition and adaptive inversion, comprising the following steps:

[0036] S1. Obtain the RGB texture image and three-dimensional point cloud data of the top coal caving area through the explosion-proof depth camera deployed above the coal discharge port. The RGB texture image captures the difference between the coal body (gray value < 50) and the gangue (high light spot). The three-dimensional point cloud extracts the boundary surface arc transition zone through an edge detection algorithm. The top coal size distribution data is collected in real time through a laser size rapid analyzer. The explosion-proof depth camera selects Intel RealSense D455 model, is installed at 0.5 m above the coal discharge port, has a pitch angle of 30°, a frame rate of 50 fps, and a point cloud accuracy of ±1 mm;

[0037] S2. Use the improved YOLOv8 model to fuse the RGB texture features, point cloud edge curvature, and size distribution data to identify the coal and gangue boundary coordinates. The improved YOLOv8 model embeds a coordinate attention (CA) module and uses an EIoU loss function. A normalized Beta function is used to enhance the image contrast;

[0038] S3. Construct a release body inversion model based on the theory of granular mechanics. The objective function is: Where Ω is the release body shape parameter, δ i is the gangue intrusion penalty coefficient, A i (Ω) is the overlapping area.

[0039] S4. Calculate the opening degree K according to the vertical height H of the coal and gangue boundary surface to the coal discharge port, and trigger the intermittent coal release strategy based on the size distribution.

[0040] In step S1, the laser size rapid analyzer uses a LMS-511 laser scanner through laser particle size sensing technology, is installed on the inlet side of the scraper conveyor, has a measurement range of 5-500 mm, a sampling frequency of 10 Hz, and the laser analyzer outputs the size gradation every 0.1 second.

[0041] The block-visual feature fusion method in step S2 includes: when the laser block analysis instrument detects that the proportion of a region >300 mm block is >80%, marking it as a "large block coal region" in the three-dimensional point cloud, and increasing the YOLOv8 detection weight coefficient of the marked region by 0.3 to preferentially identify the boundary of the large block coal.

[0042] The inversion model adaptive adjustment strategy of step S3 includes: if the proportion of >300 mm in the real-time block data is >50%, the inversion iteration step is adjusted from 0.1 m to 0.05 m; if the coal and gangue interface moving speed is >0.2 m / s, the inversion frequency is increased from 1 Hz to 5 Hz.

[0043] The weight coefficient dynamic adjustment rule in step S3 is: when the real-time mixed gangue rate is >5%, ω2 is increased to 1.5 times of the original value; when the top coal release rate is <85%, ω2 is increased to 1.2 times of the original value.

[0044] The intermittent caving strategy of step S4 is specifically: when the block analysis instrument shows that the proportion of >200 mm block is >60%, the intermittent caving cycle is started, that is, the caving opening is opened for 20 seconds and then closed for 10 seconds.

[0045] It also includes hydraulic support posture adjustment: 4 KYB-SP300 type sensors are arranged on the top beam of the hydraulic support to monitor the top coal caving pressure in real time, when the inversion shows that the center offset Δd of the release body is >0.3 m, the support inclination adjustment amount θ is calculated as arctan(Δd / h), where h is the thickness of the top coal, and θ is limited within ±2°.

[0046] A coal and gangue interface identification method control system based on depth image recognition and adaptive inversion includes:

[0047] The data acquisition module includes an explosion-proof depth camera, a laser block rapid analyzer, and a hydraulic support pressure sensor.

[0048] The processing module includes an industrial computer loaded with an improved YOLOv8 algorithm, which runs a coal and gangue interface identification and release body inversion program in real time.

[0049] The execution module includes a caving opening opening and closing mechanism controlled by an electro-hydraulic proportional valve, and a hydraulic support inclination adjusting device.

[0050] The "large block coal region" marked by the block analysis instrument is mapped to the point cloud data, and the detection weight of YOLOv8 in the region is increased: input RGB texture (coal and gangue color features) + point cloud curvature (morphological features) + block distribution (physical properties); output coal and gangue interface center coordinates (X, Y, Z).

[0051] V 实测 Calculation:

[0052]

[0053] Where W is the conveyor weight (ton), P ＞200mm The proportion of >200mm block size.

[0054] When the interface recognizes gangue, the penalty term triggers, δ i =1, otherwise 0.

[0055] Input the coal-gangue interface height H (unit: meters), and calculate the opening degree K according to the piecewise function:

[0056]

[0057] Embodiment

[0058] Embodiment one:

[0059] The block size analyzer scans the right side of the coal discharge port (X: 1.5-2.0m), and outputs: >300mm accounts for 83% (threshold value is 80%). The processing module marks this area as "large block coal area" (red highlight) in the point cloud, and the YOLOv8 detection weight is increased to 1.3.

[0060] The large block coal boundary recognition error is reduced from 0.18m to 0.07m, and the coal-gangue interface coordinates are updated to (X: 3.1m, Y: 0.9m, Z: 0.7m). The system detects that >300mm accounts for 83%>50%, and the inversion step is adjusted to 0.05m. The discharge body half-cone angle is corrected from 30° to 34.5°, and the mixed gangue rate prediction value is reduced from 7.1% to 4.3%.

[0061] Embodiment two:

[0062] On the basis of embodiment one, the coal-gangue interface height H=0.4m (less than the critical value 0.5m). The opening degree K=30% (the calculation formula triggers the third segment); 70% of the coal discharge port cross-sectional area is closed.

[0063] It should be noted that the relationship terms such as first and second in this text are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between the entities or operations. Moreover, the term "includes", "contains" or any other variant thereof is intended to cover non-exclusive inclusion, so that the process, method, article or equipment including a series of elements not only includes those elements, but also includes other elements not explicitly listed or inherent to such process, method, article or equipment. Without more limitations, the element defined by the statement "includes a" does not exclude the presence of another identical element in the process, method, article or equipment including the element.

[0064] While embodiments of the application have been shown and described, it is to be understood that the application is not limited to the details of the embodiments described, since numerous changes, modifications, substitutions and variations can be made thereto without departing from the spirit and scope of the application as defined by the appended claims and their equivalents.

Claims

1. A method for identifying coal and gangue boundaries based on depth image recognition and adaptive inversion, characterized in that, Includes the following steps: S1. Obtain RGB texture images and 3D point cloud data of the top coal collapse area through an explosion-proof depth camera deployed above the coal outlet, and simultaneously collect top coal block size distribution data in real time through a laser block size rapid analyzer; S2. An improved YOLOv8 model is used to fuse RGB texture features, point cloud edge curvature and block size distribution data to identify the coordinates of the coal gangue interface. The improved YOLOv8 model embeds a coordinate attention (CA) module and uses the EIoU loss function. S3. Construct an inversion model for the released volume based on the theory of granular mechanics, with the objective function as follows: Where Ω represents the morphological parameters of the emitted organism, and δ i Let A be the penalty coefficient for gangue intrusion. i (Ω) represents the overlapping area; S4. Calculate the opening degree K based on the vertical height H from the coal and gangue interface to the coal discharge port, and trigger the intermittent coal discharge strategy based on the block size distribution.

2. The method for identifying coal and gangue boundaries based on depth image recognition and adaptive inversion according to claim 1, characterized in that, In step S1: the explosion-proof depth camera is installed 0.5m±0.1m above the coal discharge port, with a pitch angle of 30°±5°, a frame rate of ≥50fps, and a point cloud accuracy of ±1mm. The laser particle size rapid analyzer measures the particle size distribution within a range of 5-500mm using laser particle size sensing technology, with a sampling frequency of ≥10Hz.

3. The method for identifying coal and gangue boundaries based on depth image recognition and adaptive inversion according to claim 1, characterized in that, The block size-visual feature fusion method in step S2 includes: when the laser block size analyzer detects that a certain area has a block size ratio of >300mm >80%, it is marked as "large coal block area" in the three-dimensional point cloud, and the YOLOv8 detection weight coefficient of 0.3 is added to the marked area to prioritize the identification of large coal block boundaries.

4. The method for coal gangue boundary identification based on depth image recognition and adaptive inversion according to claim 1, characterized in that, The adaptive adjustment strategy of the inversion model in step S3 includes: if the proportion of >300mm in the real-time block size data is >50%, the inversion iteration step size is adjusted from 0.1m to 0.05m; if the coal gangue interface moving speed is >0.2m / s, the inversion frequency is increased from 1Hz to 5Hz.

5. The method for identifying coal gangue boundaries based on depth image recognition and adaptive inversion according to claim 1, characterized in that, The dynamic adjustment rule for the weight coefficient in step S3 is as follows: when the real-time mixed gangue rate is >5%, increase ω2 to 1.5 times the original value; when the top coal release rate is <85%, increase ω2 to 1.2 times the original value.

6. The method for identifying coal and gangue boundaries based on depth image recognition and adaptive inversion according to claim 1, characterized in that, The intermittent coal feeding strategy in step S4 is as follows: when the block size analyzer shows that the proportion of blocks with a size greater than 200mm is greater than 60%, the intermittent coal feeding cycle is started, that is, the coal feeding port is opened for 20 seconds and then closed for 10 seconds.

7. The method for identifying coal and gangue boundaries based on depth image recognition and adaptive inversion according to claim 1, characterized in that, It also includes hydraulic support attitude adjustment: when the inversion shows that the offset of the center of the released body Δd > 0.3m, the support tilt adjustment θ = arctan(Δd / h) is calculated, where h is the thickness of the top coal, and θ is limited to ±2°.

8. A control system implementing any one of claims 1-7, characterized in that, include: The data acquisition module includes an explosion-proof depth camera, a laser block size rapid analyzer, and a hydraulic support pressure sensor. The processing module includes an industrial computer equipped with an improved YOLOv8 algorithm, which runs a coal gangue boundary identification and emission body inversion program in real time. The execution module includes a coal discharge port opening and closing mechanism controlled by an electro-hydraulic proportional valve, and a hydraulic support tilt adjustment device.

Images