Model training method, intelligent spraying method and system of mine watering truck

Abstract

This invention relates to the field of monitoring water trucks in open-pit mines. The process involves data acquisition, obtaining video data of coal burning and dust monitoring; processing the coal burning and dust video data into image data of coal burning and dust; labeling the coal burning and dust image data to form a dataset; building a YOLOv6 model and importing the dataset processed in step three; training the model to obtain a model with the ability to identify coal burning and dust; specifically, it involves model training methods, intelligent spraying methods and systems for mine water trucks; this invention provides a training method for a coal burning and dust identification model, a scheduling method for optimal spraying paths, and a training method for a dust identification model, obtaining a model with the ability to identify coal burning and dust, at least solving the problem of low efficiency in manual identification; simultaneously, it can rationally plan the spraying path based on the identified coal burning and dust location information.

Description

Technical Field

[0001] This invention relates to the field of monitoring water sprinkler trucks in open-pit mines, specifically to model training methods, intelligent spraying methods and systems for mine water sprinkler trucks. Background Technology

[0002] With the continuous expansion of the coal mining industry, large open-pit mines now mostly use heavy-duty dump trucks for transportation. The transport roads are generally unpaved dirt roads with poor surface smoothness, making dust pollution unavoidable. This not only pollutes the surrounding environment and harms the health of workers, but also shortens the lifespan of vehicles. Furthermore, dust can impair the driver's vision, affecting driving safety and reducing production efficiency. Simultaneously, after coal mining, an open face is formed, which, upon reaching its ignition point, can spontaneously combust, causing coal resource loss, ground subsidence, and the emission of large amounts of harmful gases such as carbon monoxide and sulfur dioxide into the air, resulting in significant environmental pollution. Therefore, it is essential to use water spraying to suppress coal combustion and dust in mining areas.

[0003] Currently, the common method for dust control in open-pit mines involving coal combustion is to continuously spray water on the road surface using water trucks. While this method can achieve a certain dust control effect, continuous water spraying wastes water resources. Furthermore, the road conditions in open-pit mines are complex, requiring reasonable route planning. Therefore, how to quickly and accurately obtain road surface information has become the key to road maintenance work. Summary of the Invention

[0004] This invention provides a training method for a coal and dust identification model, a scheduling method for optimal water spraying paths, and a training method for a dust identification model, thereby obtaining a model with coal and dust identification capabilities, which at least solves the problem of low efficiency in manual identification; at the same time, it can rationally plan the water spraying path based on the identified coal and dust location information.

[0005] The technical solution used in this invention is as follows: A training method for a coal and dust identification model, comprising the following steps:

[0006] Step 1: Data acquisition, obtaining video data of coal burning and dust monitoring;

[0007] Step two: Process the video data of coal burning and dust into image data of coal burning and dust.

[0008] Step 3: Tag the image data of coal and dust to form a dataset;

[0009] Step 4: Build a YOLOv6 model and import the dataset processed in Step 3; train the model to obtain a model with the ability to identify coal fires and dust.

[0010] Furthermore, the specific method for step three is as follows: use labelme software to outline the areas of coal burning and dust in the dust image data with rectangles and assign labels, convert the labeled images into txt format, and form a dataset by matching the image data of coal burning and dust with their labels.

[0011] Furthermore, the dataset is further processed as follows: using a Python automated script, the dataset is sorted into a training set: validation set: test set ratio of 7:2:1; and the processed images and their labels are stored on the server.

[0012] Specifically, in step four, multiple parameter adjustments are performed during the training process.

[0013] This application also proposes an intelligent spraying method for mining sprinkler trucks, including the following steps.

[0014] Step S1: Real-time video data acquisition;

[0015] Step S2: The data is transmitted wirelessly to DeepStream for processing; the coal and dust recognition model trained in Example 1 is deployed on DeepStream; coal and dust are recognized and detected.

[0016] Step S3: Locate the final coal dust position using a laser gimbal;

[0017] Step S4: The water truck arrives at the designated location and sprays water.

[0018] Specifically, step S4 involves monitoring the water level, flow rate, and location inside the water truck before it departs.

[0019] Preferably, step S4 specifically involves planning the optimal path for the sprinkler truck; the sprinkler truck then sprays water along the optimal path.

[0020] Specifically, the optimal path planning method is as follows:

[0021] The roads in the mine are abstracted into a network model, the paths are abstracted into edges in the network model, and the path distances are converted into edge weights.

[0022] The dispatch center and each spraying point are abstracted into N nodes.

[0023] A weighted directed graph with N nodes is represented by a weighted adjacency matrix Cost, where arcs...<Vi,Vj> The weight is represented by Cost[i,j]. If Vi is not connected to V, then Cost[i,j] = ∞.

[0024] Then we introduce the vector Dist, where Dist[i] refers to the minimum path weight from the starting point to the ending point Vi.

[0025] If a vector has an index m in a directed graph, then the value of that vector is set as:

[0026]

[0027] Where V is the set of network nodes.

[0028] Set the destinations of all paths that have found the shortest path from the starting point Vm into a set S, with the initial value of S = {Vm}. m},but:

[0029] (1) Select Vj such that: Where Vj is the endpoint of a shortest path that has already been found starting from Vm, then let: ;

[0030] (2) Modify from vertex V m Start from any vertex V in the set VS k The shortest path length. If: Dist[j] + Cost[j,k] <Dist[K]

[0031] (3) Modify Disk[k] to: Dist[K] = Dist[j] + Cost[j,k]

[0032] Repeat steps (2) and (3) N-1 times to obtain the starting point scheduling center V. m Find the shortest path from a starting point to each vertex in a directed graph.

[0033] The third aspect of this application aims to provide an intelligent spraying system for mining sprinkler trucks, characterized by comprising a high-precision camera, an image processing server, a laser pan-tilt unit, and a sprinkler truck system; the high-precision camera is installed on both sides of the open-pit mine road for real-time video data acquisition, and a communication connection is established between the high-precision camera and the processing server; the image processing server receives and processes the real-time video data, identifies coal burning and dust information, and transmits signals to the laser pan-tilt unit, which is installed in the open-pit mine to receive signals from the image processing server, lock the location of coal burning and dust, and obtain location information; the sprinkler truck system is used to spray water to extinguish dust after being dispatched.

[0034] Furthermore, the water truck system includes a water truck and measuring hardware installed on the water truck, including millimeter-wave radar and flow meter; the millimeter-wave radar is installed in the water tank of the water truck to measure the water level inside the water truck; the flow meter is installed in front of the water valve of the water truck to measure the spraying flow rate of the water truck.

[0035] The beneficial effects achieved by this invention are as follows: This application proposes a training method for a coal and dust identification model. Based on this training method, a road surface system can be established using 3D GIS technology based on video data streams from the open-pit mine site. Furthermore, using a laser pan-tilt unit, the location of coal and dust can be quickly determined, enabling real-time scheduling of water truck spraying. Compared to the current method of manual watering in open-pit mines around the clock, this method effectively saves water resources, reduces labor costs, improves spraying efficiency, and prevents dust problems caused by worker error. Attached Figure Description

[0036] Figure 1 This is a flowchart of the training method for the coal and dust identification model of the present invention.

[0037] Figure 2 This is a schematic diagram of the marking of coal and dust according to the present invention.

[0038] Figure 3 This is a hardware connection block diagram of the intelligent spraying system for mining sprinkler trucks of the present invention.

[0039] Figure 4 This is a flowchart of the intelligent spraying method for mining sprinkler trucks according to the present invention. Implementation

[0040] To facilitate understanding of the present invention by those skilled in the art, specific embodiments of the present invention will be described below with reference to the accompanying drawings.

[0041] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. For those skilled in the art, the specific meaning of the above terms in this invention can be understood through the specific circumstances.

[0042] Example 1

[0043] This invention provides a training method for a coal and dust identification model.

[0044] Includes the following steps:

[0045] like Figure 1 As shown, step one is data acquisition.

[0046] To acquire monitoring video data of coal burning and dust from the Zhahanuoer Coal Industry Co., Ltd. in Inner Mongolia Autonomous Region, high-definition cameras are installed in the open-pit mine, and the monitoring video data of coal burning and dust is collected through these high-definition cameras.

[0047] Step 2: Processing video data of coal burning and dust.

[0048] Import the video data of coal burning and dust pollution monitoring into OpenCV, and convert the video data of coal burning and dust pollution into image data at 25 frames per second;

[0049] Step 3: Processing image data of coal burning and dust.

[0050] Image data containing coal burning and dust were manually selected, and the LabelMe software was used to outline and label the areas containing coal burning and dust. Figure 2 As shown, the labeled images are converted into txt format (content includes: label, coordinates of the center point of the rectangle (x, y), length h, width w of the rectangle), and the image data of coal and dust are matched with their labels to form a dataset; using a Python automated script, the dataset is sorted into training set: validation set: test set in a ratio of 7:2:1; the processed images and their labels are stored in the server.

[0051] Step 4: Build the YOLOv6 model.

[0052] Build a YOLOv6 model; import the labeled dataset from the server in step three and train it, performing parameter tuning multiple times, such as: initial learning rate, final learning rate, warm-up learning rate epoch, learning rate warm-up initial momentum, and learning rate warm-up biased learning rate. Through the above steps, a coal and dust recognition model with optimal recognition capabilities is obtained. Example

[0053] This application also proposes an intelligent spraying system for mining sprinkler trucks.

[0054] The hardware components of an intelligent spraying system include: image acquisition hardware (high-precision camera); image processing hardware (image processing server); positioning acquisition hardware (GPS, laser pan-tilt unit, 3D GIS road surface system); and water volume measurement hardware for the sprinkler truck (millimeter-wave radar, flow meter). The communication method between the high-precision camera and the processing server is as follows: based on the existing IP address of the high-precision camera, the image processing server can directly pull video streams from the industrial camera via the RTSP protocol, thereby enabling the management and playback of the video streams on the processing server. The processing server acts as a multi-functional container, handling functions such as multi-channel processing of the video stream, execution of edge detection algorithms, and GPS positioning information. The millimeter-wave radar and flow meter are respectively installed inside the sprinkler truck's water tank and in front of the water valve to measure the water level and flow rate within the sprinkler truck.

[0055] The flow rate obtained from the flow meter, the water volume obtained from the millimeter-wave radar, the positioning information obtained from the laser gimbal, and the path obtained from the 3DGIS road surface system can be transmitted to the computing server via IP relationships within a local area network (5G data transmission can also be used).

[0056] Example 3

[0057] A Smart Spraying Method for Mining Sprinkler Trucks

[0058] Step S1: Real-time video data acquisition in open-pit mine. Several high-definition cameras are installed in the open-pit mine to collect real-time video data.

[0059] Step S2: The data is transmitted wirelessly to DeepStream for processing. The coal and dust recognition model trained in Example 1 is deployed on DeepStream. The coal and dust are identified and detected to determine whether there is coal and dust. If coal and dust are detected, the next step is performed.

[0060] Step S3: Locate the final coal dust position using a laser gimbal;

[0061] By using the angle and acceleration information of a high-precision gyroscope to drive the motor, the camera and laser emitter mounted on the laser pan-tilt platform remain fixed in inertial space, achieving target locking and automatic tracking, thereby locking and tracking the position of coal fires and dust in real time.

[0062] Step S4: The water truck arrives at the designated location and sprays water.

[0063] By calling the 3DGIS road surface system, we can view the road surface conditions and, based on the mining area road management rules, determine which roads are open to traffic and spraying.

[0064] The 3DGIS road surface system digitizes aerial photography, worker-collected terrain data, and existing terrain images, and performs data preprocessing. The raw data is converted into discrete elevation points for storage, simplified or encrypted to meet surveying requirements. Simultaneously, coordinate translation is used to compress point coordinate strings, storing the coordinates of points within each block using relative coordinates and performing data compression. The processed data is then used to construct a digital terrain model (DTM), primarily employing two methods: regular grid (GRID) and irregular triangular network (TIN). Due to the strip-like distribution and large spatial span of the road surface in mining areas, the impact data at the above scales needs to be overlaid with the DEM data to achieve a three-dimensional visualization effect, ultimately forming the final system.

[0065] At the same time, before the water truck departs, it is necessary to monitor the water level, flow rate, and location inside the water truck.

[0066] A millimeter-wave radar level gauge is installed inside the water tank of the sprinkler truck. This gauge features a high signal-to-noise ratio, strong anti-interference capability, millimeter-level accuracy, and is unaffected by temperature. It can accurately measure the water volume in the sprinkler truck's water tank. At the same time, a flow meter is installed in front of the valve to monitor the water flow in the tank in real time. GNSS technology can provide accurate satellite navigation positioning at any time, regardless of geographical location. This almost compensates for the harsh conditions caused by the remote location and complex internal environment of the mine. GPS is installed inside the sprinkler truck to display its location in real time.

[0067] Example 4

[0068] This invention also includes an optimal path planning method, which is used to plan the optimal path in step S4 above. After planning the optimal path, the sprinkler truck sprays water through the optimal path.

[0069] The optimal water spraying route is found using Dijkstra's algorithm. This model assumes a dispatch center that sends water trucks to various coal dust sources for spraying operations.

[0070] The roads in the mine are abstracted into a network model, the paths are abstracted into edges in the network model, and the path distances are converted into edge weights. In the network model, Dijkstra's algorithm can calculate the shortest path from any node to a connected node. The algorithm generates the shortest paths in an increasing order of weights, and the algorithm is implemented using an adjacency matrix.

[0071] The dispatch center and each spraying point are abstracted into N nodes.

[0072] A weighted directed graph with N nodes is represented by a weighted adjacency matrix Cost, where arcs...<Vi,Vj> The weight is represented by Cost[i,j]. If Vi is not connected to V, then Cost[i,j] = ∞.

[0073] Then, we introduce the vector Dist, where Dist[i] represents the minimum path weight from the starting point to the ending point Vi. Let the index of a vector in the directed graph be m; then the value of that vector is set as:

[0074]

[0075] Where V is the set of network nodes.

[0076] Set the destinations of all paths that have found the shortest path from the starting point Vm into a set S, with the initial value of S = {Vm}. m},but:

[0077] (1) Select Vj such that: Where Vj is the endpoint of a shortest path that has already been found starting from Vm, then let: ;

[0078] (2) Modify from vertex V m Start from any vertex V in the set VS k The shortest path length. If: Dist[j] + Cost[j,k] <Dist[K]

[0079] (3) Modify Disk[k] to: Dist[K] = Dist[j] + Cost[j,k]

[0080] Next, repeat steps (2) and (3) N-1 times to obtain the starting point scheduling center V. m The shortest path from a starting point to each vertex in a directed graph is a sequence of paths whose weights increase in order.

[0081] Based on the water level and flow rate information monitored in step five, determine which sprinkler trucks are most suitable for departure and spraying work, and arrange them to depart and spray according to the path determined by the laser gimbal.

[0082] Install electronic fences on the 3DGIS road surface system to ensure that sprinkler trucks drive within designated areas and issue alarms for sprinkler trucks that leave the electronic fence.

[0083] After all vehicles have departed, their locations are determined. Vehicles are categorized into three states: not yet departed, in operation, and completed. Water tanks are monitored to determine if they need to be refilled. When new coal fires or dust are detected, water trucks that are nearby, not currently in operation, and have sufficient water are prioritized for spraying operations, taking into account the water level and operating status of the trucks.

[0084] In addition to the automatic dispatch module, the sprinkler truck dispatch system combines a manual dispatch module and a voice communication module to achieve real-time vehicle dispatch.

[0085] The automatic scheduling module formulates the spraying task for each water truck based on the actual conditions of the open-pit mine and the watering task. By specifying the coal burning points, dust sources, and water truck locations, the system treats each watering shift as a whole. Before each operation begins, the system is input with the following information: coal burning point, dust source, water volume, and starting point. After scheduling is activated, the system obtains the optimal scheduling plan based on the input information and through system algorithms.

[0086] The manual dispatch module is an auxiliary module to the automatic dispatch module. Managers can use this module to dispatch water trucks according to special circumstances. After the server commands are transmitted, the water truck terminal receives new task commands. After the water truck completes the current task, it will execute the manual dispatch task.

[0087] The voice communication module is the module in the sprinkler truck terminal system that directly serves the driver. Through this module, the driver can receive instructions from the server in a timely manner, avoiding the inconvenience of observing the display while driving and ensuring smooth production.

[0088] The embodiments of the present invention described above do not constitute a limitation on the scope of protection of 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 claims of the present invention.

Claims

1. A smart spraying method for mining sprinkler trucks, characterized in that, Includes the following steps: Step S1: Real-time video data acquisition; Step S2: The data is transmitted wirelessly to DeepStream for processing; a pre-trained coal and dust recognition model is deployed on DeepStream to identify and detect coal and dust. The coal burning and dust identification model is trained using the following method, specifically including the following steps: Step 1: Data acquisition, obtaining video data of coal burning and dust monitoring; Step two: Process the video data of coal burning and dust into image data of coal burning and dust. Step 3: Tag the image data of coal and dust to form a dataset; Step 4: Build a YOLOv6 model and import the dataset processed in Step 3; train the model to obtain a model with the ability to identify coal fires and dust. Step S3: Locate the final location of the coal fire and dust using a laser gimbal. Specifically, the angle and acceleration information of a high-precision gyroscope are used to drive a motor, so that the camera and laser emitter mounted on the laser gimbal remain fixed in the inertial space, achieving target locking and automatic tracking, thereby locking and tracking the location of the coal fire and dust in real time. Step S4: Plan the optimal route for the water truck; the water truck travels along the optimal route to the designated location to spray water.

2. The intelligent spraying method for a mining sprinkler truck according to claim 1, characterized in that: The specific method for step three is to use the labelme software to outline the areas of coal burning and dust in the dust image data with rectangles and assign them labels. The labeled images are then converted into txt format, and the image data of coal burning and dust and their labels are matched to form a dataset.

3. The intelligent spraying method for a mining sprinkler truck according to claim 2, characterized in that: The dataset is further processed as follows: using a Python automated script, the dataset is sorted into a training set: validation set: test set ratio of 7:2:1; the processed images and their labels are then stored on the server.

4. The intelligent spraying method for a mining sprinkler truck according to claim 1, characterized in that: Step four involves multiple parameter adjustments during the training process.

5. The intelligent spraying method for a mining sprinkler truck according to claim 1, characterized in that: Step S4 specifically involves monitoring the water level, flow rate, and location inside the water truck before it departs.

6. The intelligent spraying method for a mining sprinkler truck according to claim 1, characterized in that: The optimal path planning method is as follows: The roads in the mine are abstracted into a network model, the paths are abstracted into edges in the network model, and the path distances are converted into edge weights. The dispatch center and each spraying point are abstracted into N nodes. A weighted directed graph with N nodes is represented by a weighted adjacency matrix Cost, where arcs...<Vi,Vj> The weight is represented by Cost[i,j]. If Vi is not connected to V, then Cost[i,j] = ∞. Then we introduce the vector Dist, where Dist[i] refers to the minimum path weight from the starting point to the ending point Vi; If a vector has an index m in a directed graph, then the value of that vector is set as: Where V is the set of network nodes; Set the destinations of all paths that have found the shortest path from the starting point Vm into a set S, with the initial value of S = {Vm}. m },but: (1) Select Vj such that: Where Vj is the endpoint of a shortest path that has already been found starting from Vm, then let: ; (2) Modify from vertex V m Start from any vertex V in the set VS k The shortest path length; if: Dist[j] + Cost[j,k] <Dist[K] (3) Modify Disk[k] to: Dist[K] = Dist[j] + Cost[j,k] Repeat steps (2) and (3) N-1 times to obtain the starting point scheduling center V. m Find the shortest path from a starting point to each vertex in a directed graph.

7. A smart spraying system for a mining sprinkler truck, employing the smart spraying method for a mining sprinkler truck as described in claim 1, characterized in that, The system includes high-precision cameras, an image processing server, a laser pan-tilt unit, and a water sprinkler system. The high-precision cameras are installed on both sides of the open-pit mine roads for real-time video data acquisition. The image processing server establishes a communication connection between the high-precision cameras and the image processing server. The image processing server receives and processes the real-time video data, identifies coal burning and dust, and transmits signals to the laser pan-tilt unit. The laser pan-tilt unit, located in the open-pit mine, receives signals from the image processing server, locks onto the location of coal burning and dust, and obtains location information. The water sprinkler system is used to spray water and suppress dust after deployment.

8. The intelligent spraying system for a mining sprinkler truck according to claim 7, characterized in that: The water sprinkler truck system includes the water sprinkler truck and the measuring hardware installed on the water sprinkler truck, including millimeter-wave radar and flow meter; the millimeter-wave radar is installed in the water tank of the water sprinkler truck to measure the water level in the water sprinkler truck; the flow meter is installed in front of the water valve of the water sprinkler truck to measure the spray flow rate of the water sprinkler truck.

Images