Underground mine blasting fragmentation identification method and system

Abstract

The invention discloses an underground mine blasting lumpiness identification method and system. The method comprises the following steps: acquiring an original image containing a muck pile and a scraper bucket through an image acquisition unit mounted at the top of a scraper; preprocessing the original image to obtain an optimized image; the optimized image is input into a YOLO-Cmodel, and a bounding box of the ore and the bucket is output; wherein the YOLO-Cmodel is obtained by improving a YOLOv13 model, a convolution attention module ConvAttn is introduced into a Backbone part of the YOLOv13 model to replace a depth separable convolution DSConv, an A2C2f module is replaced by a CMLS module, and the CMLS module sequentially comprises a first convolution Con, a second convolution Con, the depth separable convolution DSConv, an MECS module, a first feedforward neural network FNN, an LS convolution and a second feedforward neural network FNN; and calculating the actual lumpiness of each ore based on the boundary frame of the ore and the bucket and the known physical size of the bucket. The method is suitable for an underground mine complex environment and can realize intelligent identification of the blasting lumpiness.

Description

Technical Field

[0001] This invention relates to the field of intelligent mining and computer vision application technology, specifically to a method and system for identifying the size of blasted blocks in underground mines. Background Technology

[0002] Blasting is a core process in mining, and the size of the rock blocks produced after blasting (referred to as "blast block size") is one of the most critical technical indicators for evaluating the blasting effect. An ideal block size distribution can significantly improve the efficiency of subsequent loading, transportation, and crushing processes, and reduce production costs.

[0003] Currently, the measurement of blasted block size in mines mainly relies on manual on-site investigation and experience-based judgment, or on manual screen measurement after taking photographs. These methods have inherent drawbacks such as low efficiency, high subjectivity, poor safety (measurers need to be close to the blast pile), and difficulty in obtaining quantitative data for the entire site.

[0004] Furthermore, most existing vision-based measurement solutions require the additional deployment of calibration plates or objects of known dimensions at the measurement site. This presents significant problems in downhole environments: first, it increases workload and cost; second, calibration objects are easily damaged, contaminated, or displaced in harsh environments such as blasting and loading / unloading, leading to inaccurate measurements; and third, frequent entry and exit from blast areas to deploy calibration objects poses a potential threat to personnel safety. Therefore, a measurement method that does not require additional calibration objects has become an urgent need for downhole applications.

[0005] In recent years, block recognition methods based on computer vision have begun to be applied in open-pit mines. However, the underground mining environment poses more severe challenges to the vision system: (1) Limited space and equipment installation location make it difficult to obtain frontal images and easily cause perspective distortion; (2) Extremely high dust concentration seriously affects image clarity; (3) Vibration is severe when equipment such as loaders and loader are in operation, which easily leads to image blurring; (4) Poor and uneven lighting conditions.

[0006] These factors lead to a significant decrease in identification accuracy and reliability when directly applying existing open-pit mine solutions. Therefore, there is an urgent need for a comprehensive intelligent identification solution for blasted block size that is specifically adapted to the unique environment of underground mines and integrates hardware protection, image preprocessing, and advanced identification algorithms. Summary of the Invention

[0007] The purpose of this invention is to provide a method and system for identifying the blasted block size in underground mines. This method is applicable to the complex environment of underground mines and can effectively solve the special environmental interference of underground mines. In particular, it solves the problems caused by additional calibration materials. At the same time, it improves the YOLOv13 model to achieve automated, high-precision, and high-efficiency identification of blasted block size.

[0008] To achieve the above objectives, the technical solution adopted by this invention is as follows: In a first aspect, the present invention provides a method for identifying the size of blasted blocks in underground mines, the method comprising: The image acquisition unit installed on top of the loader acquires raw images containing the blast pile and the loader bucket; The original image is preprocessed to obtain an optimized image; The optimized image is input into the YOLO-C² model, which outputs the bounding boxes of the ore and the bucket. The YOLO-C² model is an improvement on the YOLOv13 model. In the backbone part of the YOLOv13 model, a convolutional attention module ConvAttn is introduced to replace the depthwise separable convolution DSConv, and the A2C2f module is replaced with a CMLS module. The CMLS module includes, in sequence, a first convolution Con, a second convolution Con, a depthwise separable convolution DSConv, a MECS module, a first feedforward neural network FNN, an LS convolution and a second feedforward neural network FNN. Based on the bounding box of the bucket and its known physical dimensions, the ratio of image pixels to actual physical dimensions is calculated. Combined with the bounding box of the ore, the actual block size of each ore is calculated.

[0009] Furthermore, the image acquisition unit includes a camera, a sealed transparent dust cover, an automatic cleaning structure, a shock-absorbing bracket, and an angle-adjustable gimbal that allows manual adjustment of the camera's pitch angle based on the tilt angle of the blast pile slope. Before image acquisition, the tilt angle of the blast pile slope is obtained, and the gimbal is manually adjusted so that the camera's optical axis is approximately perpendicular to the surface of the blast pile slope.

[0010] Furthermore, the automatic cleaning structure includes an automatic water spraying device and an electric dust scraping device. When the average gradient value of the image is detected to be less than or equal to a preset threshold, the automatic water spraying device and the electric dust scraping device are activated to perform cleaning.

[0011] Furthermore, the preset threshold is 40.

[0012] Further image preprocessing includes: The multi-scale Retinex algorithm (MSR) is used to suppress illumination inhomogeneity and shadows in the original image; The contrast-limited adaptive histogram equalization (CLAHE) is used to enhance the local contrast of the image after it has been processed by the MSR algorithm. Finally, guided filtering is applied to denoise the enhanced image.

[0013] Furthermore, in the YOLO-C² model, the convolutional attention module ConvAttn simulates the long-range modeling capability of self-attention and instance-dependent weighting through convolution. The MECS module combines channel attention and spatial attention mechanisms to capture and fuse features at different scales. The LS convolution combines large kernel perception with small kernel aggregation to efficiently capture a wide range of perceptual information and achieve accurate feature aggregation for dynamic and complex visual representations.

[0014] Furthermore, the YOLO-C² model is trained based on the adaptive matrix estimation optimizer.

[0015] Furthermore, the image was annotated using the Labelme annotation tool, and the annotation information included the bounding box of the ore, the category and instance segmentation mask, and the bounding box of the bucket.

[0016] Furthermore, based on the bounding box of the bucket and its known physical dimensions, the ratio of image pixels to actual physical dimensions is calculated. Combined with the bounding box of the ore, the actual block size of each ore is calculated, including: Based on the bounding box of the bucket, obtain the pixel size of the bucket in the image; Calculate the ratio K between the image pixels and the actual physical size based on the known physical dimensions of the bucket; For each ore, based on the corresponding instance segmentation mask, the maximum pixel diameter D is calculated using a contour-based caliper diameter algorithm; The actual size of each ore is calculated based on the maximum pixel diameter D and the scaling factor K.

[0017] In a second aspect, the present invention provides an underground mine blasting block size identification system, which applies the underground mine blasting block size identification method described in any one of the first aspects, the system comprising: The manual angle measurement module, including an electronic inclinometer and a data entry terminal, is used to acquire and upload the inclination angle of the blast pile slope; The explosion-proof acquisition and adjustment module includes an image acquisition unit, which is used to adjust the camera pitch angle so that the camera optical axis is approximately perpendicular to the surface of the blast pile slope, and to acquire raw images containing the blast pile and the loader bucket; The preprocessing module includes an illumination correction unit, a contrast enhancement unit, and an image denoising unit. The illumination correction unit is used to suppress illumination unevenness and shadows in the original image using the multi-scale Retinex algorithm (MSR). The contrast enhancement unit is used to enhance the local contrast of the image after MSR algorithm processing using contrast-limited adaptive histogram equalization (CLAHE). The image denoising unit is used to apply guided filtering to denoise the enhanced image. The intelligent recognition module is used to complete block size calculation based on the YOLO-C² model; The data storage and transmission module is used to realize data storage and remote uploading.

[0018] In summary, compared with the prior art, the above-described technical solutions conceived by this invention can achieve the following beneficial effects: By mounting the camera on top of the loader and installing dustproof and shock-absorbing devices, and by measuring the inclination angle of the blast pile in advance and adjusting the camera angle, the influence of the distortion of the ore blocks and bucket in the image on the actual block size measurement is greatly avoided.

[0019] By manually adjusting the pan-tilt head to reduce distortion at the source of data acquisition, and combining this with the innovative method of using the bucket in the same frame as a natural reference, residual perspective errors are effectively offset. This eliminates the need for complex software distortion correction algorithms, reducing system computational complexity while ensuring the accuracy of block size measurements. The system as a whole is more robust and better suited to the actual engineering environment of underground mines.

[0020] By using a cascaded preprocessing workflow of MSR + CLAHE + guided filtering, the problems of extreme uneven illumination, shadows, and low contrast caused by downhole auxiliary light sources are innovatively solved, providing high-quality input images for the YOLO-C² model, which is a key prerequisite for achieving high-precision recognition.

[0021] The YOLOv13 model's backbone introduces a convolutional attention module, ConvAttn, to replace the original depthwise separable convolution DSConv. ConvAttn simulates the long-range modeling capability of self-attention and instance dependency weighting through convolution, significantly reducing computational overhead and memory usage in block-level image super-resolution tasks. Furthermore, a lightweight module, CMLS, is proposed to replace the original A2C2f module. CMLS incorporates depthwise separable convolution DSConv, MECS, and LS convolutions. The MECS module combines channel attention and spatial attention mechanisms, enabling the capture and fusion of features at different scales. LS (Large-Small) convolutions combine large kernel perception with small kernel aggregation, efficiently capturing broad perceptual information and achieving precise feature aggregation for dynamic and complex visual representations, thus enabling skillful processing of visual information.

[0022] Finally, the innovative use of the bucket itself as a dimensional reference completely eliminates the need for manually pre-placed, easily damaged, and contaminated special calibration objects (such as calibration plates). This not only simplifies the operation process and eliminates systematic errors caused by calibration object displacement or damage, but also avoids frequent entry of measurement personnel into the hazardous area of ​​the explosive pile, greatly improving operational safety and making fully automated online measurement possible. Attached Figure Description

[0023] Figure 1This is a flowchart illustrating an embodiment of a method for identifying the blasting block size in underground mines provided by the present invention. Figure 2 This is a schematic diagram of the structure of an embodiment of an image acquisition unit provided by the present invention; Figure 3 This invention provides an overall architecture diagram of a YOLOv13 model; Figure 4 An architecture diagram of a lightweight CMLS module provided for this invention; Figure 5 This is a schematic diagram of the structure of an underground mine blasting block size identification system provided by the present invention.

[0024] In the picture: 1. Camera; 2. Sealed transparent dust cover; 3. Automatic cleaning structure; 4. Shock-absorbing bracket; 5. Angle-adjustable gimbal. Detailed Implementation

[0025] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. All other embodiments obtained by those skilled in the art based on the embodiments provided by this invention without inventive effort are within the scope of protection of this invention.

[0026] Obviously, the accompanying drawings described below are merely some examples or embodiments of the present invention. Those skilled in the art can apply the present invention to other similar scenarios based on these drawings without any inventive effort. Furthermore, it is understood that although the efforts made in this development process may be complex and lengthy, for those skilled in the art related to the content disclosed in this invention, modifications to design, manufacturing, or production based on the technical content disclosed in this invention are merely conventional technical means and should not be construed as insufficient disclosure of the present invention.

[0027] In this invention, the reference to "embodiment" means that a specific feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of the invention. The appearance of this phrase in various places in the specification does not necessarily refer to the same embodiment, nor is it a mutually exclusive, independent, or alternative embodiment. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described in this invention may be combined with other embodiments without conflict.

[0028] Unless otherwise defined, the technical or scientific terms used in this invention shall have the ordinary meaning understood by one of ordinary skill in the art to which this invention pertains. The terms "a," "an," "an," "the," and similar words used in this invention do not indicate quantity limitation and may indicate singular or plural. The terms "comprising," "including," "having," and any variations thereof used in this invention are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or device that includes a series of steps or modules (units) is not limited to the listed steps or units, but may also include steps or units not listed, or may include other steps or units inherent to these processes, methods, products, or devices. The terms "connected," "linked," "coupled," and similar words used in this invention are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. "A plurality" used in this invention refers to two or more. "And / or" describes the relationship between related objects, indicating that three relationships may exist; for example, "A and / or B" can represent: A alone, A and B simultaneously, and B alone. The character " / " generally indicates that the preceding and following objects have an "or" relationship. The terms "first," "second," and "third" used in this invention are merely to distinguish similar objects and do not represent a specific ordering of the objects.

[0029] This invention provides a method and system for identifying the block size of blasted ore in underground mines, applicable to complex underground mining environments. The method includes: acquiring images of the blast pile through an image acquisition unit integrated on the top of a loader; the image acquisition unit includes a camera, a dust cover, an automatic cleaning structure, a shock-absorbing support, and an angle-adjustable gimbal that can be manually adjusted based on the tilt angle of the blast pile; preprocessing the images, including illumination correction and noise reduction for uneven lighting and dark ore conditions in the images acquired in underground mines; constructing an improved YOLOv13 model, introducing a convolutional attention module ConvAttn to the backbone of the YOLOv13 model to replace the original depthwise separable convolution DSConv, and replacing the A2C2f module with the proposed CMLS module; inputting the preprocessed acquired images into the YOLO-C² model, and calculating the block size of each ore based on the segmentation results of the YOLO-C² model, using the bucket in the image as a known size reference, without the need to add additional references, thus eliminating the need for workers to place reference objects on the blast pile. This invention proposes an improved YOLOv13 model and a new CMLS module to improve the accuracy of ore block size identification. It also effectively overcomes interference from underground dust, uneven lighting, and angular distortion, achieving automated, high-precision, and high-efficiency identification of blasted block size and improving engineering safety.

[0030] Figure 1This is a flowchart illustrating an embodiment of a method for identifying the blasting block size in underground mines provided by the present invention. (Refer to...) Figure 1 The method for identifying the blasting block size in underground mines provided by this invention includes: S1. Acquire raw images of the blast pile using an image acquisition unit mounted on top of the loader; wherein the image acquisition unit includes a camera 1, a sealed transparent dust cover 2, an automatic cleaning structure located in front of the lens 3, a shock-absorbing bracket for buffering vibration 4, and an angle-adjustable gimbal 5 for manually adjusting the camera's pitch angle according to the slope angle of the blast pile, such as... Figure 2 As shown.

[0031] S2. Preprocess the original image, including illumination correction and guided filtering-based image denoising, to obtain an optimized image.

[0032] S3. Construct an improved YOLOv13 model. In the backbone of the YOLOv13 model, introduce a convolutional attention module ConvAttn to replace the original depthwise separable convolution DSConv. Then, replace the A2C2f module with the proposed CMLS module to form the YOLO-C² model. Figure 3 and Figure 4 As shown, the preprocessed image is input into the pre-trained improved YOLOv13 model, and the model outputs the bounding box coordinates of the ore target and the complete bounding box coordinates of the bucket target.

[0033] S4. Identify the bucket area in the optimized image, use the known physical size of the bucket as a reference, calculate the ratio coefficient between the image pixels and the actual physical size, and combine it with the maximum diameter of the ore pixels obtained in step S3 to calculate the actual block size of each ore.

[0034] Specifically, in step S1, the method for manually adjusting the angle-adjustable gimbal is as follows: Before image acquisition, the tilt angle of the blast pile slope is obtained by manual measurement. Then, the gimbal is manually adjusted so that the camera optical axis is approximately perpendicular to the surface of the blast pile slope, thereby making the captured image approximately a frontal projection and minimizing perspective distortion. After acquiring a large number of images containing underground mine blast piles and buckets, manual annotation is performed using the Labelme annotation tool. The annotation information includes the bounding box of the ore, its category and instance segmentation mask, and the bounding box of the bucket.

[0035] The image acquisition unit includes a camera, a dust cover, an automatic cleaning structure, a shock-absorbing bracket, and an angle-adjustable gimbal that can be manually adjusted based on the tilt angle of the explosive pile. The automatic cleaning structure includes an integrated device for automatic water spraying and electric dust scraping. When the camera detects that dust covers the image and causes blurring, it automatically starts the water spraying device and a miniature explosion-proof motor drives a silicone scraper to wipe the lens.

[0036] For example, when the camera detects that dust cover in the image causes blurring, that is, when the camera detects that the average gradient of the image is <40 (dust cover causes blurring), the micro explosion-proof motor (12V) is automatically started to drive the silicone scraper (hardness 60 Shore A) to scrape the lens, each scraping lasts for 1.2 seconds, and after the bucket unloads the material (triggered by ECU signal), the dust is forcibly scraped once.

[0037] In step S2, image preprocessing includes the following steps: S21. Illumination Inhomogeneity and Shadow Suppression: The Multi-Scale Retinex Algorithm (MSR) is employed. This algorithm uses a Gaussian wrap function to estimate illumination components at small, medium, and large scales.

[0038] Small scale: preserves the best details, but is prone to creating halos.

[0039] Mesoscale: The core balancing scale. This scale has a moderate response function radius, effectively suppressing excessive halos caused by small scales while preserving richer texture information than large scales. It is key to eliminating medium-range shadows (such as the shadow between two adjacent mineral blocks), significantly improving the naturalness and visual comfort of the image, and providing high-quality features with minimal geometric distortion for subsequent segmentation.

[0040] Large scale: It can best maintain color fidelity and eliminate large areas of shadow, but some details will be lost.

[0041] Weighted fusion: The results from the three scales are merged to suppress the effects of large-area shadows and dark light while preserving the texture and edge details of the mineral as much as possible and balancing the halo effect.

[0042] S22. Contrast Enhancement and Post-processing: The overall contrast of the image after MSR processing may be too flat. Limiting Contrast Adaptive Histogram Equalization (CLAHE) is used to enhance the local contrast of the image, further highlighting the boundaries between the ore and the background, as well as between the ore, providing a clearer basis for segmentation.

[0043] S23. Image Denoising: After illumination correction and enhancement, guided filtering is applied for noise reduction. This filter can smooth dust noise while strictly preserving the enhanced ore edges from the previous step, avoiding edge blurring that could affect the accuracy of block size calculation.

[0044] In step S3, in the YOLO-C² model, as follows Figure 3 and Figure 4As shown, a convolutional attention module, ConvAttn, is introduced into the backbone of the YOLOv13 model to replace the original depthwise separable convolution DSConv. ConvAttn simulates the long-range modeling capability of self-attention and instance dependency weighting through convolution, significantly reducing the computational overhead and memory usage in block-level image super-resolution tasks. Furthermore, a lightweight module, CMLS, is proposed in the YOLO-C² model, such as... Figure 4 As shown, the CMLS module incorporates the depthwise separable convolution DSConv, the MECS module, and the LS convolution. The MECS module combines channel attention and spatial attention mechanisms, enabling it to capture and fuse features at different scales. The LS (Large-Small) convolution combines large kernel perception with small kernel aggregation, efficiently capturing a wide range of perceptual information and achieving precise feature aggregation for dynamic and complex visual representations, thus enabling skillful processing of visual information.

[0045] Specifically, such as Figure 4 As shown, the CMLS module in this embodiment includes a first convolution Con, a second convolution Con, a depth-separable convolution DSConv, a MECS module, a first feedforward neural network FNN, an LS convolution and a second feedforward neural network FNN. It can improve the accuracy of ore block size recognition and overcome interference from underground dust, uneven lighting and viewing angle distortion, so as to realize automated, high-precision and high-efficiency recognition of blasted block size.

[0046] The acquired block size training data is input into the initial block size recognition network, i.e., the YOLO-C² model. The parameters of the initial block size recognition network are optimized by gradient descent based on the adaptive matrix estimation optimizer. The iterative optimization process yields a fully trained block size recognition network.

[0047] In step S4, the block size (maximum size) of each ore is calculated, specifically including: S31. Based on the YOLO-C² model, identify the bucket region in the image and obtain the pixel length; S32. Calculate the pixel-to-actual-length ratio K based on the actual length of the bucket: K = actual length / pixel size; S33. For each detected block, based on its instance segmentation mask, use the contour-based caliper diameter algorithm to calculate its maximum pixel diameter D. S34. Calculate the actual block size using the scaling factor K: Actual block size = pixel size × K.

[0048] This method eliminates the need for additional dedicated markers on-site. It uses the known physical dimensions of the bucket as a reference. Its innovation lies in directly utilizing existing, known-sized engineering machinery parts on-site as visual benchmarks, thus avoiding the setup, maintenance, and errors introduced by additional markers and reducing the danger and workload for workers in blasting areas.

[0049] As described above, this invention uses a pre-designed device on top of a loader to collect image data, performs Retinex illumination compensation and noise reduction on the image data, outputs a pre-processed blasting image, inputs the processed dataset into a trained YOLO-C² model to obtain segmented data of the ore-bearing and loader bounding boxes, and then completes the block size calculation, realizing automated, high-precision, and high-efficiency identification of blasting block size, thus improving the safety of the project.

[0050] Figure 5 An intelligent identification system for blasting block size in underground mines, as exemplified by this invention, comprises: Manual angle measurement module: includes an electronic inclinometer and a data entry terminal, used to acquire and upload the target inclination angle θ of the blast pile slope; Explosion-proof data acquisition and adjustment module: includes an industrial explosion-proof camera, a dual-axis electric adjustment bracket and a MEMS tilt sensor, to realize equipment angle adjustment and data acquisition; The preprocessing module includes an illumination correction unit, a contrast enhancement unit, and an image denoising unit. The illumination correction unit, based on the MSR algorithm, is responsible for eliminating uneven lighting and shadows; the contrast enhancement unit, based on the CLAHE algorithm, is responsible for improving local contrast; and the image denoising unit, based on a guided filtering algorithm, is responsible for suppressing dust noise and preserving edges. Intelligent recognition module: Includes YOLO-C² model, completes block size calculation; Data storage and transmission module: Includes a 2TB explosion-proof SSD and an industrial Ethernet transmission unit, enabling local data storage and remote uploading.

[0051] In summary, this invention provides a method and system for identifying the size of blasted blocks in underground mines. By installing a camera on top of a loader and installing dustproof and shock-absorbing devices, and by measuring the tilt angle of the blast pile in advance and adjusting the camera angle, the influence of distortion of the ore blocks and bucket in the image on the measurement of the true block size is greatly avoided.

[0052] By manually adjusting the pan-tilt head to reduce distortion at the source of data acquisition, and combining this with the innovative method of using the bucket in the same frame as a natural reference, residual perspective errors are effectively offset. This eliminates the need for complex software distortion correction algorithms, reducing system computational complexity while ensuring the accuracy of block size measurements. The system as a whole is more robust and better suited to the actual engineering environment of underground mines.

[0053] By using a cascaded preprocessing workflow of MSR + CLAHE + guided filtering, the problems of extreme uneven illumination, shadows, and low contrast caused by downhole auxiliary light sources are innovatively solved, providing high-quality input images for the YOLO-C² model, which is a key prerequisite for achieving high-precision recognition.

[0054] The YOLOv13 model's backbone introduces a convolutional attention module, ConvAttn, to replace the original depthwise separable convolution DSConv. ConvAttn simulates the long-range modeling capability of self-attention and instance dependency weighting through convolution, significantly reducing computational overhead and memory usage in block-level image super-resolution tasks. Furthermore, a lightweight module, CMLS, is proposed to replace the original A2C2f module. CMLS incorporates depthwise separable convolution DSConv, MECS, and LS convolutions. The MECS module combines channel attention and spatial attention mechanisms, enabling the capture and fusion of features at different scales. LS (Large-Small) convolutions combine large kernel perception with small kernel aggregation, efficiently capturing broad perceptual information and achieving precise feature aggregation for dynamic and complex visual representations, thus enabling skillful processing of visual information.

[0055] Finally, the innovative use of the bucket itself as a dimensional reference completely eliminates the need for manually pre-placed, easily damaged, and contaminated special calibration objects (such as calibration plates). This not only simplifies the operation process and eliminates systematic errors caused by calibration object displacement or damage, but also avoids frequent entry of measurement personnel into the hazardous area of ​​the explosive pile, greatly improving operational safety and making fully automated online measurement possible.

[0056] It should be noted that, depending on the implementation needs, the various steps / components described in this invention can be broken down into more steps / components, or two or more steps / components or parts of the operation of steps / components can be combined into new steps / components to achieve the purpose of this invention.

[0057] Those skilled in the art will readily understand that the above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method of identifying blast fragmentation in an underground mine, characterised by, The method comprises: obtaining an original image containing a muck pile and a shovel bucket through an image acquisition unit installed on the top of the shovel; preprocessing the original image to obtain an optimized image; inputting the optimized image into a YOLO-C2 model to output a boundary box of the ore and the shovel bucket; wherein the YOLO-C2 model is improved from a YOLOv13 model, and a convolutional attention module ConvAttn is introduced in the Backbone part of the YOLOv13 model to replace a depth separable convolution DSConv, and an A2C2f module is replaced by a CMLS module, and the CMLS module sequentially comprises a first convolution Con, a second convolution Con, a depth separable convolution DSConv, an MECS module, a first feedforward neural network FNN, an LS convolution, and a second feedforward neural network FNN; based on the boundary box of the shovel bucket and the known physical size of the shovel bucket, a proportionality coefficient of image pixels to actual physical size is calculated, and combined with the boundary box of the ore, the actual size of each ore is calculated.

2. The underground mine fragmentation identification method of claim 1, wherein, The image acquisition unit comprises a camera, a sealed transparent dustproof cover, an automatic cleaning structure, a shock-absorbing support, and an angle-adjustable holder for manually adjusting the pitch angle of the camera based on the slope angle of the muck pile; before image acquisition, the inclination angle of the muck pile slope is obtained, and the holder is manually adjusted so that the camera optical axis is approximately perpendicular to the surface of the muck pile slope.

3. A method of blast fragmentation sizing in an underground mine as claimed in claim 2, characterised in that, The automatic cleaning structure comprises an automatic water spraying device and an electric dust scraping device, which are started to perform cleaning when it is detected that the average image gradient is less than or equal to a preset threshold.

4. The underground mine fragmentation identification method of claim 3, wherein, The preset threshold is 40.

5. The underground mine fragmentation identification method of claim 1, wherein, Image preprocessing includes: using a multi-scale Retinex algorithm MSR to suppress uneven illumination and shadows of the original image; using a limited contrast adaptive histogram equalization CLAHE to enhance the local contrast of the image processed by the MSR algorithm; and finally applying guided filtering to the enhanced image for noise reduction processing.

6. The underground mine fragmentation identification method of claim 1, wherein, In the YOLO-C2 model, the convolutional attention module ConvAttn simulates the long-range modeling capability and instance-dependent weighting of self-attention through convolution, the MECS module combines channel attention and spatial attention mechanisms to capture and fuse features at different scales, and the LS convolution combines large kernel perception and small kernel aggregation to efficiently capture extensive perception information and accurately aggregate dynamic and complex visual representations.

7. The underground mine fragmentation identification method of claim 1, wherein, The YOLO-C2 model is trained based on an adaptive matrix estimator optimizer.

8. The underground mine fragmentation identification method of claim 1, wherein, The image is labeled using the Labelme labeling tool, and the labeling information includes the boundary box, class, and instance segmentation mask of the ore, and the boundary box of the shovel.

9. A method of fragmentation sizing for underground blasting according to claim 8, characterised in that, Based on the boundary box of the shovel bucket and the known physical size of the shovel bucket, a proportionality coefficient of image pixels to actual physical size is calculated, and combined with the boundary box of the ore, the actual size of each ore is calculated, including: based on the boundary box of the shovel bucket, the pixel size of the shovel in the image is obtained; based on the known physical size of the shovel, the proportionality coefficient K of image pixels to actual physical size is calculated; for each ore, based on the corresponding instance segmentation mask, the maximum pixel diameter D is calculated using a contour-based caliper diameter algorithm; According to the maximum pixel diameter D and the proportional coefficient K, the actual size of each ore is calculated.

10. An underground mine blast fragmentation identification system characterized by, The system comprises the underground mine blasting size identification method according to any one of claims 1 to 9. The artificial angle measurement module comprises an electronic inclinometer and a data entry terminal, and is used for acquiring and uploading the slope inclination of the blast pile; The explosion-proof collection and adjustment module comprises an image collection unit, which is used for adjusting the pitch angle of the camera to make the optical axis of the camera approximately perpendicular to the surface of the blast pile, and collecting the original image containing the blast pile and the shovel of the shovel truck; The preprocessing module comprises an illumination correction unit, a contrast enhancement unit and an image noise reduction unit, the illumination correction unit is used for suppressing uneven illumination and shadow of the original image by using a multi-scale Retinex algorithm MSR, the contrast enhancement unit is used for enhancing the local contrast of the image processed by the MSR algorithm by using a limited contrast adaptive histogram equalization CLAHE, and the image noise reduction unit is used for applying guided filtering to the enhanced image for noise reduction processing; The intelligent identification module is used for completing size calculation based on a YOLO-C2 model; The data storage and transmission module is used for realizing data storage and remote uploading.

Images