Open pit mine pit detection method and system based on rotor unmanned aerial vehicle

Abstract

The application discloses an open-pit mine pit detection method based on a rotor unmanned aerial vehicle, and comprises the following steps: S1, an unmanned aerial vehicle shoots to obtain a high-resolution open-pit mine pit target image; S2, the high-resolution open-pit mine pit target image is processed to generate an orthographic image of the open-pit mine pit area; S3, the orthographic image of the open-pit mine pit area is segmented through a pre-trained improved DenseNet network to obtain an accurate open-pit mine pit range; S4, engineering vehicles in the accurate open-pit mine pit range are detected and identified through a pre-trained target detection framework based on YOLOv3; S5, the open-pit mine pit area in the accurate open-pit mine pit range is divided into regular grids, the engineering vehicles in the grids are removed, the volume of each grid is calculated and accumulated to obtain a mine pit mining volume; and S6, the calculated mine pit mining volume is compared with a specified value to determine whether over-mining occurs. The application can quickly calculate the area, depth and volume of the open-pit mine pit and perform range checking, avoids the problem of over-mining of the mine pit, and can automatically operate, is simple to operate and has very high overall efficiency.

Description

Technical Field

[0001] This invention relates to the field of mineral engineering, and more specifically to a method and system for open-pit mine exploration based on a rotary-wing unmanned aerial vehicle (UAV), specifically for calculating mining volume and conducting ultra-deep exploration. Background Technology

[0002] Currently, traditional manual monitoring mainly involves regional sampling and measurement of specific points, which is inefficient and lacks overall, macroscopic observation of the entire mining area, resulting in blind spots. Furthermore, the measurement process is cumbersome, requiring constant movement of measuring equipment, which is time-consuming and labor-intensive. Satellite remote sensing technology, with its advantages of low cost and wide coverage, is widely used for large-scale mining area environmental measurement, but its timeliness and flexibility are limited, and its resolution is often only at the meter level, which is insufficient for detailed measurement of mine pits. In recent years, UAV low-altitude digital photogrammetry technology has developed rapidly. With its advantages of high precision, high efficiency, and low cost, it has become a new type of remote sensing technology that complements satellite remote sensing technology. This technology has significant advantages in small-scale areas and areas with difficult field measurement conditions.

[0003] However, current research on mine pits using drone photogrammetry mainly focuses on surface cover or landscape changes, and is mostly based on two-dimensional image features. It lacks automated detection of three-dimensional geometric features of the surface and cannot adequately illustrate the actual mining situation in the mining area. Summary of the Invention

[0004] The main objective of this invention is to provide a method for calculating the mining volume and conducting ultra-deep exploration in open-pit mines based on rotary-wing UAVs, which enables automated monitoring of three-dimensional geometric features of the earth's surface and allows for real-time knowledge of the actual mining situation in the mining area.

[0005] The technical solution adopted in this invention is:

[0006] A method for detecting open-pit mines based on rotary-wing UAVs is provided, comprising the following steps:

[0007] S1. The drone flies along a pre-planned shooting route to capture high-resolution images of the open-pit mine target.

[0008] S2. Perform aerial triangulation and dense matching on high-resolution open-pit target images to obtain fine point cloud results of the area surrounding the open-pit mine, and filter out flying points with abnormal positions to generate orthophotos of the open-pit mine area.

[0009] S3. The orthophoto of the open-pit mine area is segmented using a pre-trained improved DenseNet network to obtain the accurate range of the open-pit mine.

[0010] S4. Use a pre-trained YOLOv3-based target detection framework to accurately detect and identify engineering vehicles within the open-pit mine area.

[0011] S5. Divide the precise open-pit mine area into regular grids, remove engineering vehicles within the grids, calculate the volume of each grid and sum them up to obtain the mining volume of the mine.

[0012] S6. Compare the calculated mining volume of the mine pit with the specified value to determine whether over-mining has occurred.

[0013] Following the above technical solution, the preset planned shooting route is obtained through the following steps:

[0014] By taking pictures around the open-pit mine using drones, images of the open-pit mine with high-precision positioning coordinates are obtained, and then processed to obtain the initial terrain information of the open-pit mine.

[0015] Flight routes are planned based on initial terrain information to obtain close-range drone shooting routes.

[0016] Following the above technical solution, the specific construction process of the improved DenseNet network is as follows:

[0017] Construct an improved deep learning framework for DenseNet based on multi-scale feature fusion and enhancement;

[0018] Training sample sets were obtained by annotating the open-pit mine images collected at different times and in different areas.

[0019] The improved DenseNet deep learning framework was trained using a training sample set.

[0020] Following the above technical solution, the specific construction process of the YOLOv3-based object detection framework is as follows:

[0021] Construct an object detection framework based on YOLOv3;

[0022] A training sample set was obtained by annotating engineering vehicle data based on images of open-pit mines collected at different times and in different areas.

[0023] The YOLOv3-based object detection framework was trained using a training sample set.

[0024] Following the above technical solution, in step S5, if an engineering vehicle exists within the grid, the depth of that grid is determined by the average depth of the surrounding grids.

[0025] Following the above technical solution, the method further includes the following steps:

[0026] S7. Calculate the average depth of all grids within the open-pit mine area and determine whether it exceeds the specified depth.

[0027] Following the above technical solution, the specific calculation process for the average depth of the grid is as follows: calculate the depth value of each point in the grid based on the elevation value extracted from the point cloud data, and calculate the average depth value of all points in the grid as the average depth of the grid.

[0028] Following the above technical solution, in step S3, the improved DenseNet network specifically uses convolutional kernels of different sizes to obtain features at different scales of the orthophoto of the open-pit mine area, performs global pooling operation on the obtained feature maps and their sub-regions, and finally connects the obtained multi-scale feature maps to obtain the accurate range of the open-pit mine.

[0029] The present invention also provides an open-pit mine detection system based on a rotary-wing unmanned aerial vehicle, comprising:

[0030] The drone photography module is used to capture high-resolution images of open-pit mine targets by flying the drone along a pre-planned shooting route.

[0031] The mine pit image processing module is used to perform aerial triangulation and dense matching on high-resolution open-pit target images to obtain fine point cloud results of the area around the open-pit mine, filter out flying points with abnormal positions, and generate orthophotos of the open-pit mine area; and segment the orthophotos of the open-pit mine area through a pre-trained improved DenseNet network to obtain the accurate range of the open-pit mine.

[0032] The engineering vehicle recognition module is used to detect and recognize engineering vehicles within a precise open-pit mine area using a pre-trained YOLOv3-based target detection framework.

[0033] The mine volume calculation module is used to divide the mine area within a precise open-pit mine into regular grids, remove engineering vehicles within the grids, calculate the volume of each grid and sum them up to obtain the mine volume.

[0034] The judgment module is used to compare the calculated mining volume of the mine pit with the specified value to determine whether over-extraction has occurred.

[0035] The present invention also provides a computer storage medium storing a computer program executable by a processor, the computer program implementing the open-pit mine exploration method for rotary-wing UAVs described in the above technical solution.

[0036] The beneficial effects of this invention are as follows: The open-pit mine detection method based on rotary-wing UAVs can quickly calculate the area, depth, and volume of open-pit mines and perform range checks, avoiding over-extraction. It is also automated, easy to operate, and highly efficient overall. Furthermore, it detects numerous construction vehicles in the mine area as targets, and then removes their interference during the subsequent calculation of the mine's volume and depth, thereby improving calculation accuracy.

[0037] Furthermore, for large-scale open-pit mines, convolutional kernels of different sizes are used to obtain features at different scales, and global pooling is performed on the feature maps and their sub-regions. Finally, the obtained multi-scale features are connected, thereby realizing feature extraction of large-scale targets in open-pit mines. Attached Figure Description

[0038] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0039] Figure 1(a) is a flowchart of the open-pit mine exploration method based on a rotary-wing UAV according to an embodiment of the present invention.

[0040] Figure 1(b) is a flowchart of another embodiment of the present invention, which is a method for detecting open-pit mines based on a rotary-wing UAV.

[0041] Figure 1(c) is a flowchart of the open-pit mine exploration method based on a rotary-wing UAV according to the third embodiment of the present invention.

[0042] Figure 2 This is a schematic diagram of the initial terrain of the open-pit mine in an embodiment of the present invention.

[0043] Figure 3 This is a three-dimensional flight path map of the initial terrain and flight path planning for the open-pit mine in this embodiment of the invention.

[0044] Figure 4 This is a diagram showing the intelligent segmentation result of the open-pit mine area in an embodiment of the present invention.

[0045] Figure 5 This is a diagram showing the target detection results of an engineering vehicle in an open-pit mine according to an embodiment of the present invention.

[0046] Figure 6 The grid distribution (white box) after extracting the mine pit area in this embodiment of the invention. Detailed Implementation

[0047] 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.

[0048] As shown in Figure 1(a), the open-pit mine detection method based on a rotary-wing UAV in this embodiment of the invention mainly includes the following steps:

[0049] S1. The drone flies along a pre-planned shooting route to capture high-resolution images of the open-pit mine target.

[0050] S2. Perform aerial triangulation and dense matching on high-resolution open-pit target images to obtain fine point cloud results of the area surrounding the open-pit mine, and filter out flying points with abnormal positions to generate orthophotos of the open-pit mine area.

[0051] S3. The orthophoto of the open-pit mine area is segmented using a pre-trained improved DenseNet network to obtain the accurate range of the open-pit mine.

[0052] S4. Use a pre-trained YOLOv3-based target detection framework to accurately detect and identify engineering vehicles within the open-pit mine area.

[0053] S5. Divide the precise open-pit mine area into regular grids, remove engineering vehicles within the grids, calculate the volume of each grid and sum them up to obtain the mining volume of the mine.

[0054] S6. Compare the calculated mining volume of the mine pit with the specified value to determine whether over-mining has occurred.

[0055] Furthermore, as shown in Figure 1(b), the method also includes the step: S7, calculating the average depth of all grids within the open-pit mine area and determining whether it exceeds the specified depth.

[0056] In another embodiment of the present invention, as shown in FIG1(c), the open-pit mine detection method based on a rotary-wing UAV mainly includes the following steps:

[0057] Step 1: The drone acquires initial terrain information;

[0058] Operating a rotary-wing drone equipped with RTK, routine photography was conducted around the open-pit mine to obtain images of the mine with high-precision positioning coordinates. These images were then processed to obtain the initial terrain information of the open-pit mine; for example... Figure 2 As shown

[0059] Step 2: Using the initial terrain information of the open-pit mine obtained in Step 1, a flight path is planned to obtain a close-range drone photography flight path, such as... Figure 3As shown;

[0060] Step 3: The drone automatically takes pictures;

[0061] The open-pit mine trajectory information obtained in step 2 is imported into the UAV's flight control system. The UAV takes off and automatically takes pictures to obtain high-resolution images of the open-pit mine target.

[0062] Step 4: Obtaining detailed results;

[0063] Based on the high-resolution image obtained in step 4, aerial triangulation and dense matching are performed to obtain fine point cloud results for the area around the open-pit mine. Flying points with abnormal positions are filtered out, and orthophotos of the area around the mine are generated.

[0064] Step 5: Intelligent segmentation of open-pit mine area based on improved DenseNet network;

[0065] Based on collected images of open-pit mines from different times and regions, data annotation was performed to obtain a sample set suitable for segmentation training. After constructing and training an improved DenseNet deep learning framework based on multi-scale feature fusion and enhancement, orthophotos of the open-pit mine area were segmented to obtain a precise extent of the open-pit mine, such as... Figure 4 As shown, since open-pit mines are generally large in scale, in order to obtain semantic information over a wide range, convolutional kernels of different sizes can be used to obtain features at different scales. Global pooling operations can be performed on the feature map and its sub-regions. For example, the feature map can be divided into n×m sub-regions (such as 2×2 sub-regions), and then average pooling can be performed on each sub-region. Finally, the obtained multi-scale features are concatenated, thereby realizing feature extraction of large-scale targets in open-pit mines.

[0066] Step 6: Target detection of engineering vehicles in the mining area based on YOLOv3;

[0067] Based on open-pit mine images collected at different times and in different areas, engineering vehicle data was labeled to obtain a sample set suitable for segmentation training. After constructing and training a YOLOv3-based object detection framework, engineering vehicle targets within the precise open-pit mine area obtained in step 5 were detected. Figure 5As shown. In a preferred embodiment of the present invention, the main focus is on detecting the numerous construction vehicles in the mining area. These vehicles are then removed as interference during the subsequent calculation of the mining volume and depth. To detect construction vehicles, a corresponding training dataset can be constructed based on previously acquired images. The YOLOv3 object detection framework is then trained to detect these vehicles. Furthermore, to improve the accuracy of the network in recognizing construction vehicles, the original image can be cropped to fit the input size of the YOLOv3 network during actual detection, effectively increasing the number of construction vehicle target pixels in the network input image.

[0068] Step 7: Calculation and verification of open-pit mining volume based on fine point cloud;

[0069] Based on the refined point cloud results obtained in step 4 and the precise extent of the mine pit obtained in step 5, the mine pit area is divided into a regular grid. For each grid within the mine pit area, the volume of each grid can be calculated by combining the average depth and area of ​​the grid. Specifically, for each point within the grid, the depth value can be calculated based on its elevation. The average depth of the grid is obtained by averaging the depth values ​​of all points within the grid. Based on the engineering vehicle target detected in step 6, if an engineering vehicle is present within a grid, the depth of that grid is determined by the average depth of the surrounding grids, thus eliminating the influence of the engineering vehicle. The mining volume of the mine pit is obtained by summing all the grid values. The calculated mining volume is compared with the value specified in the permit document to check for over-mining. In this embodiment of the invention, because the mining area is not large, the grid length and width are set to 20 meters. Figure 6 As shown, after removing the engineering vehicle targets detected in step 6, the volumes of all grids are summed to obtain the volume of the entire open-pit mine. The initial volume of the mine is calculated to be 1.71 million cubic meters, which meets the requirements of the permit document and there is no over-extraction.

[0070] Step 8: Ultra-deep exploration of open-pit mines based on fine point clouds;

[0071] Based on all the grids within the open-pit mine area obtained in step 7, the average depth of all grids is calculated to check if any grid has a depth exceeding the range specified in the permit document. Since the average depth information is calculated for each grid, these average depth information are statistically analyzed and arranged in descending order. The largest value is the deepest depth among all grids. In this embodiment, the deepest depth is 45 meters, and the average depth is 31 meters, both of which meet the requirements of the permit document.

[0072] In the above technical solution, step 5, the specific method for intelligent segmentation of the open-pit mine area based on the improved DenseNet network is as follows:

[0073] Since the area captured by drones is generally larger than the actual area of ​​an open-pit mine, the actual area of ​​the open-pit mine can be extracted using image segmentation methods. Based on open-pit mine images collected at different times and in different areas, data annotation is performed to obtain a sample set suitable for segmentation training. Training types include mine areas, vegetation areas, and soil areas, resulting in a training set, a validation set, and a test set.

[0074] By adjusting the number of nonlinear combination functions in the dense blocks and the growth rate parameter of the dense blocks in the original DenseNet network, the computational efficiency of the DenseNet network is improved. Furthermore, an attention-based semantic enhancement mechanism is introduced into the decoder, thereby improving the accuracy of semantic segmentation. Using the improved DenseNet network, semantic segmentation of the orthophoto generated in step 4 can yield a precise determination of the open-pit mine's extent.

[0075] In summary, the present invention has the following advantages:

[0076] 1) This invention can accurately extract the range of open-pit mines and calculate the volume of the mines. Moreover, this method is automated, easy to operate, and has very high overall efficiency.

[0077] 2) This invention can automatically detect engineering vehicles in the mine pit, thereby eliminating the influence of engineering vehicles in volume calculation.

[0078] 3) This invention can perform mining volume calculation and ultra-deep detection, and compare the results with the values ​​specified in the license document, thereby automatically realizing mine mining inspection and avoiding the problem of over-mining.

[0079] It should be understood that those skilled in the art can make improvements or modifications based on the above description, and all such improvements and modifications should fall within the protection scope of the appended claims.

Claims

1. A method for detecting open-pit mines based on rotary-wing unmanned aerial vehicles (UAVs), characterized in that, Includes the following steps: S1. The drone flies along a pre-planned shooting route to capture high-resolution images of the open-pit mine target. S2. Perform aerial triangulation and dense matching on high-resolution open-pit target images to obtain fine point cloud results of the area surrounding the open-pit mine, and filter out flying points with abnormal positions to generate orthophotos of the open-pit mine area. S3. The orthophoto of the open-pit mine area is segmented using a pre-trained improved DenseNet network to obtain the accurate range of the open-pit mine. The improved DenseNet network specifically uses convolutional kernels of different sizes to obtain features at different scales and performs global pooling operations on the feature maps and their sub-regions. S4. Use a pre-trained YOLOv3-based target detection framework to accurately detect and identify engineering vehicles within the open-pit mine area. S5. Divide the precise open-pit mine area into regular grids, remove engineering vehicles within the grids, calculate the volume of each grid and sum them up to obtain the mining volume of the mine. S6. Compare the calculated mining volume of the mine pit with the specified value to determine whether over-mining has occurred.

2. The method for open-pit mine exploration based on a rotary-wing UAV according to claim 1, characterized in that, The pre-planned shooting route is obtained through the following steps: By taking pictures around the open-pit mine using drones, images of the open-pit mine with high-precision positioning coordinates are obtained, and then processed to obtain the initial terrain information of the open-pit mine. Flight routes are planned based on initial terrain information to obtain close-range drone shooting routes.

3. The method for open-pit mine exploration based on a rotary-wing UAV according to claim 1, characterized in that, The specific construction process of the improved DenseNet network is as follows: Construct an improved deep learning framework for DenseNet based on multi-scale feature fusion and enhancement; Training sample sets were obtained by annotating the open-pit mine images collected at different times and in different areas. The improved DenseNet deep learning framework was trained using a training sample set.

4. The method for open-pit mine exploration based on a rotary-wing UAV according to claim 1, characterized in that, The specific construction process of the YOLOv3-based object detection framework is as follows: Construct an object detection framework based on YOLOv3; A training sample set was obtained by annotating engineering vehicle data based on images of open-pit mines collected at different times and in different areas. The YOLOv3-based object detection framework was trained using a training sample set.

5. The method for open-pit mine exploration based on a rotary-wing UAV according to claim 1, characterized in that, In step S5, if an engineering vehicle exists within the grid, the depth of that grid is determined by the average depth of the surrounding grids.

6. The method for open-pit mine exploration based on a rotary-wing unmanned aerial vehicle according to any one of claims 1-5, characterized in that, The method also includes the following steps: S7. Calculate the average depth of all grids within the open-pit mine area and determine whether it exceeds the specified depth.

7. The method for open-pit mine exploration based on a rotary-wing UAV according to claim 6, characterized in that, The specific calculation process for the average depth of the grid is as follows: calculate the depth value of each point in the grid based on the elevation value extracted from the point cloud data, and calculate the average depth value of all points in the grid as the average depth of the grid.

8. The method for open-pit mine exploration based on a rotary-wing UAV according to claim 6, characterized in that, In step S3, the improved DenseNet network specifically uses convolutional kernels of different sizes to obtain features at different scales of the orthophoto of the open-pit mine area, performs global pooling on the obtained feature maps and their sub-regions, and finally connects the obtained multi-scale feature maps to obtain the accurate range of the open-pit mine.

9. An open-pit mine detection system based on a rotary-wing unmanned aerial vehicle (UAV), characterized in that, include: The drone photography module is used to capture high-resolution images of open-pit mine targets by flying the drone along a pre-planned shooting route. The mine pit image processing module is used to perform aerial triangulation and dense matching on high-resolution open-pit target images to obtain fine point cloud results of the area around the open-pit mine, filter out flying points with abnormal positions, and generate orthophotos of the open-pit mine area; and segment the orthophotos of the open-pit mine area through a pre-trained improved DenseNet network to obtain the accurate range of the open-pit mine. The engineering vehicle recognition module is used to accurately detect and recognize engineering vehicles within the open-pit mine using a pre-trained YOLOv3-based target detection framework. The improved DenseNet network specifically utilizes convolutional kernels of different sizes to obtain features at different scales and performs global pooling operations on the feature maps and their sub-regions. The mine volume calculation module is used to divide the mine area within a precise open-pit mine into regular grids, remove engineering vehicles within the grids, calculate the volume of each grid and sum them up to obtain the mine volume. The judgment module is used to compare the calculated mining volume of the mine pit with the specified value to determine whether over-extraction has occurred.

10. A computer storage medium, characterized in that, It contains a computer program that can be executed by a processor, which implements the open-pit mine exploration method for rotary-wing UAVs as described in claim 6.

Images