Mining area watering cart running control method based on animal avoidance real-time re-planning

Abstract

The invention discloses a mining area watering cart driving control method based on animal avoidance real-time re-planning, and the method achieves the intelligent obstacle avoidance through multi-modal sensor fusion detection, deep learning driven target recognition, three-dimensional dynamic path planning, a multistage acousto-optic early warning system and adaptive speed adjustment. And a radar, a camera, a bionic compound eye vision module and the like are adopted, so that the detection response time is shortened, and the false alarm rate is reduced. Constraint conditions such as terrain gradient and soil humidity are integrated through an RRT algorithm, continuous processing of path curvature is achieved, the watering cart can complete path reconstruction without complete stop in an animal penetration scene, a graded early warning mechanism triggers three-level acousto-optic warning according to the animal approaching degree, and a dynamic speed adjusting model is matched to adjust the speed of the watering cart. And the spraying operation continuity is ensured while the safety allowance is ensured. According to the invention, through the environment interference compensation, the geological disaster early warning interface and the driver behavior simulation unit, the operation safety and the operation efficiency in the complex environment of the mining area are obviously improved.

Description

Technical Field

[0001] This invention relates to the field of autonomous driving and control applications, and in particular to a driving control method for a mine sprinkler truck based on real-time replanning for animal avoidance. Background Technology

[0002] With the development of intelligent mining technology, the application of unmanned vehicles in mining areas is becoming increasingly widespread, improving mining efficiency and safety. However, the sudden intrusion of animals into the driving path poses a challenge to the safety and stability of autonomous vehicles.

[0003] Common autonomous vehicles typically follow preset routes, using sensors to detect obstacles and then taking emergency braking or obstacle avoidance measures. Existing obstacle avoidance measures are slow to react when faced with dynamic animal avoidance and struggle to replan paths in real time, impacting the efficiency and safety of unmanned sprinkler trucks in mining areas and failing to meet the requirements of autonomous driving and control applications. Therefore, a driving control method for sprinkler trucks in mining areas based on real-time replanning for animal avoidance is proposed. Summary of the Invention

[0004] This invention provides the following technical solution: a method for controlling the driving of a water sprinkler truck in a mining area based on real-time replanning for animal avoidance, comprising: S1 multimodal sensor fusion detection: First, an animal detection sensor group is constructed by combining radar, camera, bionic compound eye vision module, polarization light sensor array and infrared imaging equipment with multi-source information fusion algorithm. Then, the animal detection sensor group is used to detect and identify dynamic obstacles in the mining environment in real time. In the process of real-time detection and identification, the environmental interference compensation mechanism set in the multi-source information fusion algorithm is used to perform signal denoising processing for interference factors such as dust and strong light in the mining area. S2 Deep Learning-Driven Target Recognition: This invention employs the YOLOv7 deep learning algorithm to train models for specific animal species in mining areas, including blue sheep, foxes, and birds. It also utilizes transfer learning to improve the accuracy of target recognition in small samples. Furthermore, the built-in attention mechanism of the YOLOv7 algorithm enhances the ability to predict animal trajectories. By fusing multi-source data from radar, cameras, a bionic compound eye vision module, a polarized light sensor array, and infrared imaging equipment, combined with the animal trajectory prediction capabilities of the YOLOv7 algorithm, early detection and accurate identification of animal targets in mining areas are achieved. Compared to traditional single-sensor solutions, this invention shortens animal detection response time and reduces false alarm rates, providing a crucial time window for subsequent obstacle avoidance decisions. S3 3D Dynamic Path Planning: A real-time path planning module based on the RRT algorithm is employed. This module receives map data, real-time animal location information, and sprinkler truck dynamic parameters. Based on the acquired data, it constructs a multi-constraint optimization model incorporating terrain slope, soil moisture, and work area priority. This model then generates a new path that meets vehicle kinematic constraints, satisfying minimum turning radius and maximum gradient. Furthermore, a path smoothing optimizer and its internal Bézier curves are used to achieve curvature continuity in the planned path. By integrating mine-specific constraints such as terrain slope and soil moisture, the RRT-based real-time path planning module can quickly generate new paths that meet vehicle kinematic requirements. This overcomes the limitations of traditional preset route driving, allowing sprinkler trucks to reconstruct paths without complete stopping in animal attack scenarios. This improves path replanning efficiency, reduces downtime, and increases overall work efficiency. S4 Multi-level Audio-Visual Warning System: The system employs a tiered warning mechanism, including primary, intermediate, and advanced warnings, which triggers an alarm when an animal is detected approaching during operation. Through the tiered warning system and environmental adaptive adjustment function, the system ensures effective animal removal while avoiding the stress response that may be caused by traditional emergency braking. Furthermore, the dynamic speed adjustment model precisely controls the vehicle speed, allowing the sprinkler truck to maintain a reasonable operating speed while reserving sufficient safety margin during obstacle avoidance. This mechanism reduces the risk of conflict between people, vehicles, and animals, while ensuring the continuity of spraying operations. S5 Adaptive Speed ​​Adjustment: A dynamic speed adjustment model was established to adjust the speed of the sprinkler truck in real time based on the animal's danger level and the tortuosity of the path.

[0005] Preferably, the animal detection sensor group in step S1 also integrates a self-cleaning and maintenance module. The self-cleaning and maintenance module is composed of a micro air pump and a retractable brush assembly. When the amount of mineral dust attached to the surface of the polarized light sensor array, the bionic compound eye vision module, and the camera reaches a preset threshold, the cleaning program is automatically started. Pulsed airflow combined with flexible bristles is used to achieve non-destructive cleaning. The self-cleaning and maintenance module can automatically clean the mineral dust on the sensor surface, ensuring that it works normally in a dusty environment, improving detection reliability, and providing accurate data for obstacle avoidance decisions.

[0006] Preferably, the YOLOv7 deep learning algorithm in step S2 is configured with an online incremental learning and data closed-loop optimization mechanism. This mechanism is implemented through edge computing nodes. When an unknown dynamic target is detected or a vehicle is manually taken over due to abnormal driving conditions, multimodal sensor data is automatically captured, homomorphically encrypted, and then transmitted back to the cloud-based manual annotation platform. When an existing type of animal is detected, the onboard GPU computing unit is immediately used to start the model fine-tuning process, and the target detection network is locally trained through transfer learning. The trained model parameters are automatically updated after security verification, and the feature vectors are encrypted and uploaded to the cloud-based federated learning system for global model aggregation. Through the online incremental learning mechanism, the model can be automatically fine-tuned when new animals or abnormal behaviors are detected, improving recognition accuracy, adapting to changes in animals in the mining area, and shortening recognition time.

[0007] Preferably, the real-time path planning module in step S3 integrates a terrain risk warning interface. The terrain risk warning interface is linked with the terrain scanning radar through Internet of Things technology and acquires terrain humidity and ground cracking data in real time. When the monitoring detects that the water pits and mud formed by the convergence of terrain, or the dry cracks caused by heavy rolling on dry ground exceed the safety threshold, the emergency avoidance mode is immediately activated.

[0008] Preferably, the graded early warning mechanism in step S4 sets response levels based on the degree of animal approach. The primary early warning is based on a medium distance threshold of 100-100 meters, triggering a low-frequency sound wave pulse to alert the animal in a non-disturbing manner. The intermediate early warning activates a high-frequency sound and light combination signal at a close distance of 30-60 meters to enhance the warning intensity. The advanced early warning activates a strong light flashing and a maximum volume alarm within a critical range of less than 30 meters. By setting response levels according to the degree of animal approach through the graded early warning mechanism, accurate early warnings are provided, animal stress reactions are avoided, animals are effectively driven away, and driving and operation safety is ensured.

[0009] Preferably, the dynamic speed adjustment model in step S5 integrates a tire adhesion evaluation submodule. This submodule acquires four-wheel speed, torque output, and steering angle data via the vehicle's CAN bus, combines these with real-time calculated vehicle longitudinal acceleration and yaw rate, and uses a fuzzy logic algorithm to evaluate the current road surface adhesion coefficient. When a slippery road surface with an adhesion coefficient below 0.4 is detected, the maximum permissible vehicle speed is automatically reduced, and the electronic brake force distribution system is activated during braking. By evaluating the road surface adhesion coefficient through the tire adhesion evaluation submodule, the vehicle speed and braking force are automatically adjusted on slippery roads, improving driving stability and enhancing system safety.

[0010] Preferably, the environmental interference compensation mechanism in step S1 integrates a thermal radiation suppression unit. The thermal radiation suppression unit uses pulse cooling technology to address image noise caused by the high-temperature environment in the mining area. It uses a miniature Stirling refrigerator to periodically pulse-cool the infrared detector. The thermal radiation suppression unit solves the high-temperature noise problem of the infrared imaging equipment through pulse cooling technology, ensuring detection accuracy and providing reliable heat source data for obstacle avoidance decisions.

[0011] Preferably, the path smoothing optimizer in step S3 is equipped with a work quality assurance submodule. This submodule automatically adjusts the nozzle opening and closing sequence when the path curvature exceeds a critical value by mapping the path curvature to the spray width. It also adopts a real-time compensation strategy based on a lookup table method and monitors the vehicle's attitude through an inertial measurement unit to dynamically correct spraying deviations caused by terrain undulations. By adjusting the nozzle opening and closing sequence according to the path curvature and dynamically correcting spraying deviations, the work quality during obstacle avoidance is ensured, and work efficiency is improved.

[0012] Preferably, the graded early warning mechanism in step S4 is deeply coupled with the sprinkler truck's operating mode. When an animal is detected approaching, an audible and visual alarm is triggered, and a deceleration command is sent to the spray control system via the vehicle's industrial bus. This creates a non-linear mapping relationship between the spray flow rate and the vehicle speed. A segmented control strategy is adopted: normal operating parameters are maintained during the initial early warning stage; a flow attenuation program is initiated during the intermediate early warning stage; and spraying operations are suspended during the advanced early warning stage until the animal leaves the danger zone, at which point the operating parameters are automatically restored. By coupling the graded early warning mechanism with the operating mode, spray parameters are adjusted when an early warning is triggered, achieving obstacle avoidance and operational coordination, ensuring operational continuity, and improving system adaptability.

[0013] Preferably, the dynamic speed adjustment model in step S5 also integrates an intelligent compliant control unit. The intelligent compliant control unit analyzes the road conditions and work load data of the mining area through deep learning algorithms to construct an adaptive speed adjustment model, so that the sprinkler truck generates a smooth and continuous speed change curve according to the real-time road conditions and animal avoidance needs in automatic mode. At the same time, it sets safety boundary constraints to ensure that the acceleration change rate does not exceed 0.5 meters per second squared while ensuring work efficiency.

[0014] In summary, compared with the prior art, the present invention provides a driving control method for mine sprinkler trucks based on real-time replanning for animal avoidance, which has the following beneficial effects: 1. This invention achieves early detection and accurate identification of animal targets in mining areas by fusing multi-source data from radar, cameras, bionic compound eye vision modules, polarized light sensor arrays, and infrared imaging devices, combined with the animal trajectory prediction capability of the YOLOv7 algorithm. Compared with traditional single-sensor solutions, this invention shortens the animal detection response time and reduces the false alarm rate, providing a critical time window for subsequent obstacle avoidance decisions and fundamentally solving the problem of slow response speed in traditional technologies. At the same time, through a real-time path planning module based on the RRT algorithm, by integrating the unique constraints of mining areas such as terrain slope and soil moisture, it can quickly generate new paths that meet the kinematic requirements of vehicles, thereby breaking through the limitations of traditional preset route driving. This allows the sprinkler truck to complete path reconstruction without completely stopping in animal attack scenarios, improving path replanning efficiency and reducing operation interruption time, thus increasing work efficiency. 2. This invention, through a graded early warning system and environmental adaptive adjustment function, ensures the effectiveness of animal deterrence while avoiding the stress response that may be caused by traditional emergency braking. Furthermore, the dynamic speed adjustment model enables precise speed control of the vehicle, allowing the sprinkler truck to maintain a reasonable operating speed while reserving sufficient safety margin during obstacle avoidance. This mechanism reduces the risk of conflict between people, vehicles, and animals, while ensuring the continuity of spraying operations. At the same time, through an environmental interference compensation mechanism, it strengthens the resistance to interference factors such as dust, strong light, and geological activity in mining areas, and expands the application boundaries of unmanned driving technology. Attached Figure Description

[0015] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0016] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0017] Please see Figure 1 This invention provides a technical solution for a method of controlling the driving of a water sprinkler truck in a mining area based on real-time replanning for animal avoidance, comprising the following steps: S1 multimodal sensor fusion detection: First, an animal detection sensor group is constructed by combining radar, camera, bionic compound eye vision module, polarization light sensor array and infrared imaging equipment with multi-source information fusion algorithm. Then, the animal detection sensor group is used to detect and identify dynamic obstacles in the mining environment in real time. In the process of real-time detection and identification, the environmental interference compensation mechanism set in the multi-source information fusion algorithm is used to perform signal denoising processing for interference factors such as dust and strong light in the mining area. The animal detection sensor group also integrates a self-cleaning and maintenance module. The self-cleaning and maintenance module consists of a micro air pump and a retractable brush assembly. When the amount of mineral dust attached to the polarized light sensor array, the bionic compound eye vision module and the camera surface reaches a preset threshold, the cleaning program is automatically started. Pulsed airflow combined with flexible bristles achieves non-destructive cleaning. The environmental interference compensation mechanism integrates a thermal radiation suppression unit. The thermal radiation suppression unit uses pulse cooling technology to address image noise generated by the infrared imaging equipment due to the high temperature environment in the mining area. It uses a miniature Stirling refrigerator to periodically pulse cool the infrared detector. The specific implementation process of the above method is as follows: First, a composite sensing array is constructed, consisting of radar, cameras, a bionic compound eye vision module, a polarized light sensor array, and an infrared imaging device. The sensors are arranged in a complementary manner to cover the vehicle: the radar is deployed around the vehicle to form a horizontal 360-degree detection ring, the cameras adopt a binocular stereo vision solution and are installed at the front and rear of the vehicle, the bionic compound eye vision module is distributed around the vehicle in a hexagonal array, and the polarized light sensor array and the infrared imaging device form a vertical detection group, which focuses on covering the three-dimensional space in front of the sprinkler truck's driving path. During the data acquisition phase, all sensors enter a collaborative working mode. The radar continuously scans dynamic obstacles around the vehicle by emitting millimeter waves and receiving reflected signals, focusing on capturing point cloud data with motion characteristics; the camera system simultaneously acquires visible light image streams and performs preliminary target segmentation through convolutional neural networks; the bionic compound eye vision module utilizes its high temporal resolution to track the trajectory of fast-moving objects; the polarization light sensor array enhances animal outline extraction in rainy and foggy weather by analyzing the polarization characteristics of light reflected from the object's surface; and the infrared imaging equipment focuses on differences in thermal radiation to detect animals hidden in grass or low-light environments. After entering the information fusion center, the multi-source data first undergoes spatiotemporal alignment processing. The system uses a Kalman filter algorithm to synchronize radar point clouds and visual images spatiotemporally, ensuring strict correspondence between data from different sensors in the time dimension and spatial coordinate system. Then, it enters the feature extraction stage. A deep learning network extracts animal morphology features from the visual data, while the radar data processing module generates a distance-velocity vector map of obstacles. Polarized light and infrared data are transformed into intermediate features that can be used for target association through a feature mapping network. In the target association stage, the system employs a multi-hypothesis tracking algorithm to correlate data from potential targets detected by different sensors. This algorithm establishes a consistency hypothesis regarding target motion, eliminating false targets caused by sensor errors or environmental interference. For confirmed animal targets, the system further utilizes the YOLOv7 target recognition model for fine-grained classification to determine the target type and its movement trend. To address the unique environmental interference in mining areas, the system employs a dual compensation mechanism. For dust suppression, when the polarization sensor or vision module detects a decrease in visibility, it automatically activates the wavelet transform denoising module. This module performs multi-scale decomposition of the sensor signal to filter out noise components caused by dust particle scattering. For strong light interference, the system uses a dynamic exposure control strategy. It adjusts the exposure parameters of the camera and infrared imaging equipment in real time based on the light intensity, and simultaneously uses the reflected light angle information obtained from the polarization sensor to intelligently compensate for overexposed areas caused by strong light. When the amount of mineral dust adhering to the sensor surface reaches a preset threshold, the self-cleaning maintenance module automatically initiates the cleaning program. This module generates pulsed airflow through a miniature air pump, which, in conjunction with a retractable flexible brush, performs non-destructive cleaning of the optical sensor surface. The cleaning process employs a segmented control strategy, prioritizing the cleaning of the core sensing areas that have the greatest impact on detection accuracy. After cleaning, a built-in calibration board is used for rapid calibration to ensure that sensor performance is restored promptly. To address the impact of the high-temperature environment in the mining area on infrared imaging equipment, the thermal radiation suppression unit employs a miniature Stirling refrigerator for periodic pulse cooling. This unit adjusts its cooling power in real time based on the operating temperature of the infrared detector, minimizing the impact of thermal noise on image quality while ensuring normal equipment operation. The cooling process is synchronized with the water truck's operating cycle, providing deep cooling during vehicle downtime for resupply, while maintaining a basic temperature control mode during operation. S2 Deep Learning-Driven Target Recognition: The YOLOv7 deep learning algorithm was used to train models for specific animal species in the mining area, including argali, foxes and birds. At the same time, the accuracy of target recognition in small samples was improved through transfer learning mechanism, and the attention mechanism unit built into the YOLOv7 deep learning algorithm was used to focus on strengthening the ability to predict animal movement trajectories. The YOLOv7 deep learning algorithm is equipped with an online incremental learning and data closed-loop optimization mechanism. This mechanism is implemented through edge computing nodes. When an unknown dynamic target is detected or a vehicle is triggered to take over manually due to abnormal driving conditions, multimodal sensor data is automatically captured, homomorphically encrypted, and then sent back to the cloud-based manual annotation platform. When an existing animal category is detected, the onboard GPU computing unit is immediately used to start the model fine-tuning process, and the target detection network is trained locally through transfer learning. The trained model parameters are automatically updated after security verification, and the feature vectors are encrypted and uploaded to the cloud-based federated learning system for global model aggregation. The specific implementation process of the above method is as follows: First, a deep learning framework based on the YOLOv7 algorithm was constructed, and the model was trained for animal species unique to the mining area, such as blue sheep, foxes, and birds. The training dataset covers animal behavior samples under different seasons, times, and weather conditions. Data augmentation techniques were used to generate virtual samples covering various animal postures, such as standing, running, and jumping, to ensure the model's robustness to changes in animal morphology. The model training phase employs a transfer learning strategy. First, it is pre-trained on a publicly available animal detection dataset to enable the network to initially grasp the ability to extract general animal features. Then, it is fine-tuned on a mining-specific dataset by freezing the parameters of the lower convolutional layers and fine-tuning the parameters of the higher fully connected layers, allowing the model to quickly adapt to the specific mining environment. To address the high cost of labeling animal samples in mining areas, the system introduces a semi-supervised learning mechanism. Unlabeled data undergoes knowledge distillation using a teacher-student model, automatically generating pseudo-labels to participate in model iteration. In the model optimization phase, the attention mechanism unit plays a crucial role. This module embeds spatial attention and channel attention sub-networks, enabling the network to focus on key parts of the animal target, such as the head and limbs. During the feature extraction stage, the spatial attention module generates pixel-level weight maps, enhancing the feature response to the animal's outline region; the channel attention module, through a compression-excitation network structure, dynamically adjusts the contribution of different feature channels. This dual attention mechanism allows the model to accurately locate animal targets even in complex backgrounds. To address the need for animal motion trajectory prediction, the system constructs a spatiotemporal feature fusion network. In the temporal dimension, recurrent neural network units are introduced to model the changes in animal position between consecutive frames; in the spatial dimension, graph convolutional networks are used to capture the motion correlations of animal limb joints. Finally, a feature fusion layer jointly encodes the spatiotemporal information to generate motion trajectory prediction results containing parameters such as velocity, direction, and acceleration. When a new animal species is detected or an abnormal behavior pattern is observed in an existing animal, the online incremental learning mechanism is automatically activated. Edge computing nodes first isolate and store abnormal samples. When the accumulated amount of similar samples reaches a preset threshold, the model fine-tuning process is triggered. The fine-tuning process is performed locally on the vehicle's GPU computing unit, using knowledge distillation technology to keep the basic model parameters unchanged, updating only the parameters of the classifier corresponding to the newly added category. To protect data privacy, after feature extraction is completed locally, only the gradient update is uploaded to the cloud-based federated learning system for secure aggregation with other mining vehicles. During the model deployment phase, the system employs dynamic quantization technology to compress the YOLOv7 model, reducing computational resource consumption while maintaining detection accuracy. The inference process is optimized using the TensorRT acceleration engine, leveraging GPU parallel computing capabilities for real-time detection. Detection results are output in a structured data format, including animal species, confidence level, bounding box coordinates, and trajectory prediction parameters, providing crucial input for subsequent obstacle avoidance decisions. S3 3D Dynamic Path Planning: A real-time path planning module based on the RRT algorithm is adopted. The module receives map data, real-time animal location information and sprinkler truck dynamic parameters. Based on the data obtained, a multi-constraint optimization model including terrain slope, soil moisture and operation area priority is constructed. Then, a new path that meets the vehicle kinematic constraints of minimum turning radius and maximum climbing slope is generated. The planned path is then processed for curvature continuity using a path smoothing optimizer and its internal Bézier curve. The real-time path planning module integrates a terrain risk warning interface. The terrain risk warning interface is linked with the terrain scanning radar through IoT technology and acquires terrain humidity and ground cracking data in real time. When the monitoring detects that the water pits and mud formed by the convergence of terrain, or the dry cracks caused by heavy rolling on dry ground exceed the safety threshold, the emergency avoidance mode is immediately activated. The path smoothing optimizer has an internal operation quality assurance submodule. The operation quality assurance submodule automatically adjusts the nozzle opening and closing sequence when the path curvature exceeds the critical value by mapping the path curvature and spray width. It also adopts a real-time compensation strategy based on the lookup table method. At the same time, it monitors the vehicle attitude through the inertial measurement unit and dynamically corrects the spraying deviation caused by terrain undulation. The specific implementation process of the above method is as follows: During the data reception phase, the module simultaneously acquires three types of core information: high-precision map data including a 3D terrain model of the mining area, work area division, and historical path records; real-time obstacle location coordinates and predicted motion trajectories output by the animal detection sensor group; and vehicle dynamics parameters covering current vehicle speed, steering angle, and drive system status. This data is transmitted to the planning module via in-vehicle Ethernet, forming the foundational dataset for path calculation. The path generation stage employs an improved RRT algorithm, which introduces a dynamic weight offset strategy based on the traditional random sampling mechanism. When sampling in open areas, the algorithm prioritizes areas with gentle slopes and low soil moisture; when animal activity is detected, the sampling point automatically shifts to the area opposite to the direction of obstacle movement. During path search, the algorithm continuously verifies whether candidate paths meet vehicle kinematic constraints, including minimum turning radius, maximum gradeability, and suspension system travel limits, ensuring the physical feasibility of the planned path. The multi-constraint optimization model is the core decision engine for path planning. This model integrates four main categories of constraints: terrain constraints, which assess slope variations and surface roughness using a digital elevation model; environmental constraints, which combine soil moisture sensor data to avoid muddy sections; operational constraints, which set path preferences based on regional priorities to ensure coverage density in high-value operational areas; and safety constraints, which transform animal trajectory predictions into dynamic buffer zones to maintain a safe distance from obstacles. The model employs a hierarchical decision-making mechanism, prioritizing efficiency during routine operations and immediately switching to a safety-first mode upon detecting animals. The path smoothing optimizer refines the initial path. Its built-in Bézier curve fitting module interpolates key points of the path using cubic Bézier curves to generate a smooth trajectory with continuously changing curvature. The optimization process employs a progressive correction strategy, first replacing large-angle turning points of the path with arcs, then performing curvature transition processing on adjacent line segments, ultimately forming an executable path that meets the minimum turning radius requirements of vehicles. The geological disaster early warning interface is deeply integrated with the mining area geological monitoring system through IoT technology, receiving real-time monitoring data from rock displacement sensors and groundwater level gauges. When the monitored value exceeds a preset safety threshold, the system immediately activates the emergency avoidance mode, elevating the geological risk coefficient to the primary consideration in path planning. At this time, the Dijkstra algorithm is activated to calculate safe evacuation routes. This algorithm constructs a cost map based on geological stability parameters, prioritizing areas with intact rock strata and low groundwater levels to generate escape routes. The operation quality assurance submodule ensures that operation quality is unaffected by path changes through dynamic coupling between path curvature and the spraying system. This module establishes a mapping relationship between path curvature and sprinkler opening / closing sequence. When a sharp turn is detected, the corresponding sprinkler on that side is automatically shut off to prevent uneven water distribution caused by centrifugal force. Simultaneously, the inertial measurement unit monitors the vehicle's attitude in real time, obtaining the corresponding spray compensation amount for the current roll and pitch angles using a lookup table method. This dynamically adjusts the pump pressure and sprinkler angle to eliminate spraying deviations caused by terrain undulations. S4 Multi-level Audio-Visual Warning System: A tiered early warning mechanism, including primary, intermediate, and advanced warnings, is set up so that an alarm is triggered when an animal is detected approaching during the journey. The graded early warning mechanism sets response levels based on the degree of animal approach. The primary warning is based on a medium distance threshold of 100-100 meters, triggering a low-frequency sound wave pulse to alert the animal in a non-disturbing manner. The intermediate warning activates a high-frequency sound and light combination signal at a close distance of 30-60 meters to enhance the warning intensity. The advanced warning activates a strong light flashing and maximum volume alarm within a critical range of less than 30 meters. The graded early warning mechanism is deeply coupled with the operation mode of the sprinkler truck. When an animal is detected approaching, an audible and visual alarm is triggered, and a deceleration command is sent to the spray control system via the vehicle industrial bus. This creates a non-linear mapping relationship between the spray flow rate and the vehicle speed. A segmented control strategy is adopted to maintain normal operating parameters during the initial warning stage, activate the flow attenuation program during the intermediate warning stage, and suspend spraying operations during the high warning stage until the animal leaves the danger zone and the operating parameters are automatically restored. S5 Adaptive Speed ​​Adjustment: Establish a dynamic speed adjustment model to adjust the speed of the sprinkler truck in real time based on the animal's danger level and the tortuosity of the path; The dynamic speed adjustment model integrates a tire adhesion assessment submodule. The tire adhesion assessment submodule obtains four-wheel speed, torque output and steering angle data through the vehicle CAN bus, combines the real-time calculated vehicle longitudinal acceleration and yaw rate, and uses a fuzzy logic algorithm to assess the current road surface adhesion coefficient. When a wet road surface with a road surface adhesion coefficient of less than 0.4 is detected, the maximum permissible vehicle speed is automatically reduced, and the electronic brake force distribution system is activated during braking. The dynamic speed adjustment model also integrates an intelligent compliant control unit. The intelligent compliant control unit analyzes the road conditions and work load data in the mining area through deep learning algorithms to build an adaptive speed adjustment model. This enables the sprinkler truck to generate a smooth and continuous speed change curve in automatic mode based on real-time road conditions and animal avoidance needs. At the same time, it sets safety boundary constraints to ensure that the rate of acceleration change does not exceed 0.5 meters per second squared while ensuring work efficiency. The specific implementation process of the above method is as follows: The model integrates a tire adhesion assessment submodule, which works as follows: it continuously acquires data such as four-wheel speed, torque output, and steering angle via the vehicle's communication bus, and combines this with real-time calculated vehicle longitudinal acceleration and yaw rate. A special logic algorithm is then used to comprehensively assess the current road surface adhesion. Once the system detects that the road surface adhesion coefficient is below a specific wet / slippery standard, it immediately and automatically reduces the maximum permissible speed and activates the electronic brake-force distribution system during braking to ensure the vehicle's stability and safety on wet / slippery surfaces. In addition, the dynamic speed regulation model also includes an intelligent compliant control unit. This unit uses deep learning algorithms to conduct in-depth analysis of road conditions and workload data in the mining area, thereby constructing an adaptive speed regulation model. With this model, the water truck in automatic mode can generate a smooth and continuous speed change curve based on real-time road conditions, such as road slope and mud level, as well as animal avoidance needs. To ensure safety while maintaining operational efficiency, the system also sets safety boundary constraints to ensure that the rate of acceleration change does not exceed a specified value. The entire adaptive speed adjustment process is dynamic and intelligent. It continuously receives data from various sources, analyzes and calculates it in real time, and then makes corresponding speed adjustment decisions. For example, when an animal is detected approaching, the system will gradually adjust the vehicle speed according to the animal's danger level. In low-risk situations, it maintains a normal speed; in medium-risk situations, it initiates gradual deceleration; and in high-risk situations, it immediately performs emergency braking. At the same time, it will further optimize the speed adjustment strategy based on the curvature of the path to ensure that the sprinkler truck can maintain a reasonable operating speed while avoiding obstacles, thus ensuring the continuity and stability of the spraying operation.

[0018] This solution achieves early detection and accurate identification of animal targets in mining areas by fusing multi-source data from radar, cameras, bionic compound eye vision modules, polarized light sensor arrays, and infrared imaging equipment, combined with the animal trajectory prediction capabilities of the YOLOv7 algorithm. Compared to traditional single-sensor solutions, this invention shortens animal detection response time and reduces false alarm rates, providing a critical time window for subsequent obstacle avoidance decisions and fundamentally solving the problem of slow response speed in traditional technologies. Furthermore, through a real-time path planning module based on the RRT algorithm, which integrates unique constraints in mining areas such as terrain slope and soil moisture, it can quickly generate new paths that meet vehicle kinematic requirements. This overcomes the limitations of traditional preset route driving, allowing water trucks to reconstruct paths without completely stopping in animal-prone scenarios, improving path replanning efficiency, reducing downtime, and increasing work efficiency.

[0019] This solution, through a tiered early warning system and environmental adaptive adjustment function, ensures effective animal deterrence while avoiding the stress response that may be triggered by traditional emergency braking. Furthermore, the dynamic speed adjustment model precisely controls vehicle speed, allowing the water truck to maintain a reasonable operating speed while reserving sufficient safety margin during obstacle avoidance. This mechanism reduces the risk of conflict between people, vehicles, and animals, while ensuring the continuity of spraying operations. Additionally, through an environmental interference compensation mechanism, it enhances resistance to interference factors such as dust, strong light, and geological activity in mining areas, and expands the application boundaries of autonomous driving technology.

[0020] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0021] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for controlling the driving of a water sprinkler truck in a mining area based on real-time replanning for animal avoidance, characterized in that: Includes the following steps: S1 multimodal sensor fusion detection: First, an animal detection sensor group is constructed by combining radar, camera, bionic compound eye vision module, polarization light sensor array and infrared imaging equipment with multi-source information fusion algorithm. Then, the animal detection sensor group is used to detect and identify dynamic obstacles in the mining environment in real time. In the process of real-time detection and identification, the environmental interference compensation mechanism set in the multi-source information fusion algorithm is used to perform signal denoising processing for interference factors such as dust and strong light in the mining area. S2 Deep Learning-Driven Target Recognition: The YOLOv7 deep learning algorithm was used to train models for specific animal species in the mining area, including argali, foxes and birds. At the same time, the accuracy of target recognition in small samples was improved through transfer learning mechanism, and the attention mechanism unit built into the YOLOv7 deep learning algorithm was used to focus on strengthening the ability to predict animal movement trajectories. S3 3D Dynamic Path Planning: A real-time path planning module based on the RRT algorithm is employed. This module receives map data, real-time animal location information, and sprinkler truck dynamic parameters. Based on the acquired data, a multi-constraint optimization model is constructed, incorporating terrain slope, soil moisture, and work area priority. Subsequently, a new path is generated that satisfies vehicle kinematic constraints of minimum turning radius and maximum gradient. Furthermore, a path smoothing optimizer and its internal Bézier curves are used to perform curvature continuity processing on the planned path. S4 Multi-level Audio-Visual Warning System: A tiered early warning mechanism, including primary, intermediate, and advanced warnings, is set up so that an alarm is triggered when an animal is detected approaching during the journey. S5 Adaptive Speed ​​Adjustment: A dynamic speed adjustment model was established to adjust the speed of the sprinkler truck in real time based on the animal's danger level and the tortuosity of the path.

2. The method for controlling the driving of a water sprinkler truck in a mining area based on real-time replanning for animal avoidance, as described in claim 1, is characterized in that: The animal detection sensor group in step S1 also integrates a self-cleaning and maintenance module. The self-cleaning and maintenance module is composed of a micro air pump and a retractable brush assembly. When the amount of mineral dust attached to the surface of the polarized light sensor array, the bionic compound eye vision module and the camera reaches a preset threshold, the cleaning program is automatically started, and pulsed airflow is used in conjunction with flexible bristles to achieve non-destructive cleaning.

3. The method for controlling the driving of a water sprinkler truck in a mining area based on real-time replanning for animal avoidance, as described in claim 1, is characterized in that: The YOLOv7 deep learning algorithm in step S2 is configured with an online incremental learning and data closed-loop optimization mechanism. This mechanism is implemented through edge computing nodes. When an unknown dynamic target is detected or a vehicle is triggered to take over manually due to abnormal driving conditions, multimodal sensor data is automatically captured, homomorphically encrypted, and then transmitted back to the cloud-based manual annotation platform. When an existing type of animal is detected, the onboard GPU computing unit is immediately used to start the model fine-tuning process, and the target detection network is trained locally through transfer learning. The trained model parameters are automatically updated after security verification, and the feature vectors are encrypted and uploaded to the cloud-based federated learning system for global model aggregation.

4. The method for controlling the driving of a water sprinkler truck in a mining area based on real-time replanning for animal avoidance, as described in claim 1, is characterized in that: In step S3, the real-time path planning module integrates a terrain risk warning interface. The terrain risk warning interface is linked with the terrain scanning radar through Internet of Things technology and acquires terrain humidity and ground cracking data in real time. When the monitoring detects that the water pits and mud pools formed by the convergence of terrain, or the dry cracks caused by heavy rolling on dry ground exceed the safety threshold, the emergency avoidance mode is immediately activated.

5. The method for controlling the driving of a water sprinkler truck in a mining area based on real-time replanning for animal avoidance, as described in claim 1, is characterized in that: The graded early warning mechanism in step S4 sets response levels based on the degree of animal approach. The primary warning is based on a medium distance threshold of 100-100 meters, triggering a low-frequency sound wave pulse to alert the animal in a non-disturbing manner. The intermediate warning activates a high-frequency sound and light combination signal at a close distance of 30-60 meters. The advanced warning activates a strong light flashing and maximum volume alarm within a critical range of less than 30 meters.

6. The method for controlling the driving of a water sprinkler truck in a mining area based on real-time replanning for animal avoidance, as described in claim 1, is characterized in that: The dynamic speed adjustment model in step S5 integrates a tire adhesion evaluation submodule. The tire adhesion evaluation submodule obtains four-wheel speed, torque output and steering angle data through the vehicle CAN bus, combines the real-time calculated vehicle longitudinal acceleration and yaw rate, and uses a fuzzy logic algorithm to evaluate the current road surface adhesion coefficient. When a wet road surface with a road surface adhesion coefficient lower than 0.4 is detected, the maximum allowable speed is automatically reduced, and the electronic brake force distribution system is activated during braking.

7. The method for controlling the driving of a water sprinkler truck in a mining area based on real-time replanning for animal avoidance, as described in claim 1, is characterized in that: The environmental interference compensation mechanism in step S1 integrates a thermal radiation suppression unit. The thermal radiation suppression unit uses pulse cooling technology to periodically pulse-cool the infrared detector by using a miniature Stirling refrigerator to address image noise generated by the high-temperature environment in the mining area.

8. The method for controlling the driving of a water sprinkler truck in a mining area based on real-time replanning for animal avoidance, as described in claim 1, is characterized in that: The path smoothing optimizer in step S3 is equipped with an operation quality assurance submodule. The operation quality assurance submodule automatically adjusts the nozzle opening and closing sequence when the path curvature exceeds the critical value by mapping the path curvature and spray width. It also adopts a real-time compensation strategy based on the lookup table method and monitors the vehicle attitude through the inertial measurement unit to dynamically correct the spraying deviation caused by terrain undulation.

9. The method for controlling the driving of a water sprinkler truck in a mining area based on real-time replanning for animal avoidance, as described in claim 1, is characterized in that: The graded early warning mechanism in step S4 is deeply coupled with the sprinkler truck's operation mode. When an animal is detected approaching, an audible and visual alarm is triggered, and a deceleration command is sent to the spray control system via the vehicle's industrial bus. This creates a non-linear mapping relationship between the spray flow rate and the vehicle speed. A segmented control strategy is adopted, maintaining normal operating parameters during the initial early warning stage, initiating a flow attenuation program during the intermediate early warning stage, and suspending spraying operations during the advanced early warning stage, until the operating parameters are automatically restored after the animal leaves the danger zone.

10. The method for controlling the driving of a water sprinkler truck in a mining area based on real-time replanning for animal avoidance, as described in claim 1, is characterized in that: The dynamic speed adjustment model in step S5 also integrates an intelligent compliant control unit. The intelligent compliant control unit analyzes the road conditions and work load data of the mining area through deep learning algorithms to construct an adaptive speed adjustment model, so that the sprinkler truck generates a smooth and continuous speed change curve according to the real-time road conditions and animal avoidance needs in automatic mode.

Images