A method and system for unmanned intelligent loading material level detection

Abstract

The application discloses a kind of unmanned intelligent loading material level detection method and system, it is related to material level detection technical field, including obtaining current weather condition, vehicle type, car hop size, using sensor or manual registration or ranging radar to carry out collection, and using information entropy to measure information redundancy and value density in different data sources, automatically judge whether multiple sensor redundancy input is needed;The material level detection method and system of the unmanned intelligent loading of the application, can dynamically evaluate the importance and redundancy degree of each data source in environment and vehicle information, automatically determine whether multiple sensor cooperative collection needs to be enabled, effectively improve the adaptability of system to different working conditions.

Description

Technical Field

[0001] This invention relates to the field of material level detection technology, specifically to a method and system for unmanned intelligent loading material level detection. Background Technology

[0002] In automated loading scenarios for bulk materials in mines, ports, power plants, and other locations, unmanned intelligent loading technology is gradually being widely applied to improve operational efficiency and reduce labor costs. Traditional methods for detecting loading material levels mainly rely on manual observation, fixed-point radar, or simple camera systems to monitor the vehicle's cargo bed.

[0003] These methods often suffer from low accuracy, weak dynamic adaptability, and an inability to adaptively adjust to different vehicle models and loading environments. This leads to underloading or overloading during the loading process, affecting the accuracy of material measurement and the safety of subsequent transportation. In recent years, with the development of deep learning, image recognition, and multi-source information fusion technologies, AI-based intelligent perception and decision-making systems have provided new solutions for unmanned loading. However, most current systems still lack the ability to jointly model complex working conditions such as weather, dust, and humidity with variables such as loading speed and truck bed structure, and they also lack the ability to dynamically adjust loading strategies based on different vehicle models. Summary of the Invention

[0004] The purpose of this invention is to provide a method and system for unmanned intelligent loading material level detection, thereby solving the problems existing in the prior art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a method for material level detection in unmanned intelligent loading, comprising:

[0006] S1. Obtain current weather conditions, vehicle type, and truck bed size by using sensors, manual registration, or ranging radar. Information entropy is used to measure the redundancy and value density of information from different data sources, and the system automatically determines whether multiple sensor redundant inputs are needed.

[0007] S2. Use a ranging radar to measure the material level, take pictures of the truck bed during the loading process, and perform image recognition on the images. Combine the image recognition results to analyze the current dust concentration, and determine the dynamic compensation value of the current ranging radar data based on humidity and dust concentration. Use image entropy to measure the degree of dust interference in the truck bed image, and use image entropy as a weighting coefficient to adjust the compensation ratio of the ranging radar.

[0008] S3. Use the changes in the material level in the bucket of different vehicles during the loading process as a dataset for deep learning, so as to establish the functional relationship between the material level change of different vehicles and the loading speed and loading position. When training the material level change model of different vehicle models, use KL divergence to measure the difference between the predicted distribution and the actual material level change distribution as the loss function.

[0009] S4. During the loading process, move the vehicle to change the loading position and replace the material level. Based on the function relationship obtained from training, determine the material level change of different buckets, and decide the loading speed and whether to stop loading at the next moment. Model the material accumulation caused by the change of loading position, calculate the resultant force direction and resultant force of each loading point, analyze the material flow trend. When the resultant force direction is aligned with the center of the bucket, it is judged as uniform loading. Otherwise, the loading trajectory is corrected by adjusting the vehicle position and direction.

[0010] Preferably, it also includes S5, ending the loading when the initial height of the reset material level is close to the vehicle's allowable loading height threshold, and correcting the parameters of the loading function based on the loading result, using Newton interpolation to locally approximate the function curve between the current material level and the height threshold, estimating when the threshold will be reached, issuing an early warning, and if the final result deviates from the prediction, using error feedback to update the difference quotient and correct the loading function model.

[0011] Preferably, it also includes S6, adding the relationship between the material level change and the vehicle position change during the loading process to the deep learning training set, further training and optimizing the loading speed and function relationship of different vehicle models, and mapping different vehicle model numbers, material level features, and position combinations to different training channels through hash mapping.

[0012] Preferably, in step S3, when training material level change models for different vehicle models, the specific formula for using KL divergence to measure the difference between the predicted distribution and the actual material level change distribution is as follows:

[0013] ;

[0014] in, KL divergence represents the degree of difference between model predictions and actual observations. The actual material level change data in the truck bed is on the first The probability of each state. The material level distribution predicted by the network is in the th... The probability of each state. This is the index of the material level distribution range.

[0015] Preferably, in step S4, the material accumulation caused by the change in the loading position is modeled, and the specific formula for calculating the resultant force direction and resultant force result at each loading point is as follows:

[0016] ;

[0017] in, This represents the sum of the stacking trend vectors at all loading points. Indicates the first Material accumulation trend vector at each loading location Indicates the index number of the loading location point. This indicates the total number of loading points.

[0018] Preferably, step S2 further includes performing a temporal consistency check on multiple consecutive time frames of truck bed images during the image recognition stage to eliminate interference from momentary occlusion or abnormal lighting on image entropy calculation.

[0019] Preferably, S3 further includes introducing a model memory mechanism during deep learning training to identify and recall recurring loading behavior patterns of specific vehicle models, thereby reducing real-time computation.

[0020] Preferably, step S4 further includes intelligent path filtering of the relative displacement between the current loading position and the previous loading point, retaining only the displacement direction that can bring material level optimization benefits.

[0021] Preferably, in step S6, the specific formula for mapping different vehicle model numbers, material location features, and location combinations to different training channels via hashing is as follows:

[0022] ;

[0023] in, Indicates the input feature encoding. Indicates the total number of hash buckets. This indicates the training subset number to which the input sample is assigned. This represents the modulo operator.

[0024] A material level detection system for unmanned intelligent loading, comprising the steps of a method for detecting material levels during unmanned intelligent loading, the system including:

[0025] The information acquisition module is used to acquire information on weather, vehicle type, and truck bed size. It uses sensors, manual registration, or ranging radar to collect data and determines whether to enable multi-sensor redundant input based on information entropy.

[0026] The image and radar fusion module is used to acquire images of the truck bed and identify dust concentration during the loading process, and to perform dynamic compensation processing on the ranging radar data by combining humidity and image entropy.

[0027] The model training module is used to build a deep learning model based on the material level changes during the loading process of different vehicles, and optimizes the functional relationship between material level and loading speed and loading position through KL divergence.

[0028] The control decision module is used to determine the current material level status based on the model, control the vehicle position adjustment and update the loading speed, analyze the material flow uniformity based on the resultant force direction of each loading point and dynamically correct the loading trajectory.

[0029] As can be seen from the above technical solution, the present invention has the following beneficial effects:

[0030] This unmanned intelligent loading material level detection method and system, by introducing an information entropy evaluation mechanism, can dynamically assess the importance and redundancy of various data sources in the environment and vehicle information, automatically deciding whether to activate multi-sensor collaborative acquisition, effectively improving the system's adaptability to different working conditions. It integrates image recognition and ranging radar, combining dust concentration and image entropy weights to dynamically compensate radar data, improving the system's ranging accuracy in complex environments such as dust and humidity, avoiding material level estimation deviations caused by single-sensor errors. Through deep learning, it establishes a mapping relationship between material level changes and loading speed and position, and optimizes the prediction model using KL divergence. It can accurately reflect the differences in loading behavior of various vehicle types and structures, improve the generalization ability of loading strategies, adjust vehicle position and loading speed based on real-time judgment results, and analyze material flow trends through force modeling to ensure uniform distribution of materials in the truck bed, avoid overloading or uneven loading caused by loading deviation, improve loading quality and transportation safety. Compared with the traditional method of relying on fixed-point radar or manual observation, this invention realizes intelligent integration of perception-judgment-control closed loop, has stronger autonomous loading decision-making ability, and is suitable for the deployment of unmanned loading systems in bulk material scenarios such as mines, ports, and power plants. It has good practicality and promotion value. Attached Figure Description

[0031] Figure 1 This is a flowchart of the method of the present invention;

[0032] Figure 2 This is a system connection diagram of the present invention. Detailed Implementation

[0033] 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.

[0034] like Figure 1 As shown, the present invention provides a technical solution: a method for material level detection in unmanned intelligent loading, comprising:

[0035] S1. Obtain current weather conditions, vehicle type, and truck bed size by using sensors, manual registration, or ranging radar. Information entropy is used to measure the redundancy and value density of information from different data sources, and the system automatically determines whether multiple sensor redundant inputs are needed.

[0036] S2. Use a ranging radar to measure the material level, take pictures of the truck bed during the loading process, and perform image recognition on the images. Combine the image recognition results to analyze the current dust concentration, and determine the dynamic compensation value of the current ranging radar data based on humidity and dust concentration. Use image entropy to measure the degree of dust interference in the truck bed image, and use image entropy as a weighting coefficient to adjust the compensation ratio of the ranging radar.

[0037] S3. Use the changes in the material level in the bucket of different vehicles during the loading process as a dataset for deep learning, so as to establish the functional relationship between the material level change of different vehicles and the loading speed and loading position. When training the material level change model of different vehicle models, use KL divergence to measure the difference between the predicted distribution and the actual material level change distribution as the loss function.

[0038] S4. During the loading process, move the vehicle to change the loading position and replace the material level. Based on the function relationship obtained from training, determine the material level change of different buckets, and decide the loading speed and whether to stop loading at the next moment. Model the material accumulation caused by the change of loading position, calculate the resultant force direction and resultant force of each loading point, analyze the material flow trend. When the resultant force direction is aligned with the center of the bucket, it is judged as uniform loading. Otherwise, the loading trajectory is corrected by adjusting the vehicle position and direction.

[0039] This implementation method establishes an unmanned intelligent loading system that integrates environmental perception, adaptive compensation, image recognition, and intelligent decision-making to achieve real-time and accurate monitoring and dynamic adjustment of the material loading status in the truck bed. The overall system is built on multimodal data fusion and deep learning, and the key technical paths are as follows:

[0040] Environment and initial parameter perception phase:

[0041] First, key variables such as weather parameters (temperature, humidity, wind speed) and vehicle information (model number, truck bed size) of the current loading environment are acquired through environmental sensors. Data acquisition can be automated (e.g., using temperature and humidity sensors, millimeter-wave radar) or manually entered. The redundancy and effective information density of different input sources are calculated using information entropy. When the redundancy of multi-source data is high and the confidence of a single sensor is insufficient, the system automatically activates a multi-sensor redundancy mechanism to improve the robustness and effectiveness of the input data.

[0042] Real-time material level detection and image information fusion stage:

[0043] During the loading process, ranging radar is used to continuously and dynamically scan the material height (i.e., material level) inside the truck bed; simultaneously, industrial cameras positioned above or to the side of the loading equipment intermittently or continuously acquire panoramic images of the truck bed. The system utilizes image recognition algorithms (such as YOLO and Mask R-CNN) to detect dust distribution characteristics, material pile outline changes, and illumination shifts in the truck bed, and combines this with current humidity data to assess the degree of interference of dust on the radar ranging signal reflection intensity.

[0044] Based on this, image information is quantified into an image entropy index, reflecting the intensity of the impact of interfering elements such as dust on material level identification. The system sets a dynamic weighting coefficient based on image entropy to adjust the compensation ratio of the ranging radar output signal, thereby offsetting signal distortion in areas with high dust concentrations. The compensation strategy can be dynamically calculated based on a linear regression model or a polynomial fitting method.

[0045] Modeling and learning phase of the loading process:

[0046] To enhance intelligent decision-making capabilities, the system collects historical data on material level changes in the truck hopper during the loading process, establishing a dataset that includes multiple vehicle types and loading conditions. Sequence modeling networks such as LSTM and Transformer are used to model the trend of material level changes in the truck hopper over time and position, training to obtain the functional mapping relationship between material level, speed, and position.

[0047] The loss function incorporates KL divergence to measure the difference between the model's predicted material level distribution and the actual observed material level distribution. This metric reflects the model's fit to the actual loading trend and helps adjust the model's structural parameters and learning rate strategy.

[0048] Intelligent path adjustment and stacking modeling stage:

[0049] Based on the trained model, the system analyzes the material level distribution and spatial uniformity within the hopper in real time during the loading process. If material pile shifts or localized high material levels occur, the system will control the vehicle to make small movements (forward, backward, or lateral shifts) to change the loading point.

[0050] Meanwhile, the system models the material flow path and accumulation trend caused by changes in the loading point location based on physical simulation methods (such as the Discrete Element Method, DEM). This model calculates the direction of the local resultant force and the total resultant force vector generated at each loading point and compares them with the geometric center of the truck bed. When the resultant force direction tends towards the centerline, the system determines that the loading is uniform; otherwise, it corrects the trajectory through vehicle fine-tuning.

[0051] This implementation method effectively improves the quality assessment and dynamic compensation capabilities of sensor data by introducing information entropy and image entropy quantification mechanisms during the material level detection process, enhancing the system's adaptability to dust interference and humidity changes in harsh environments. By constructing a material level change function model using deep learning, the accuracy of predicting the loading trends of different vehicle models is significantly improved. Combined with the resultant force modeling of material accumulation, automatic identification and real-time correction of loading uniformity are achieved, effectively reducing safety hazards and material waste caused by uneven loading. The overall system can complete high-precision, intelligent loading monitoring and control without human intervention, possessing high practicality and promotional value.

[0052] It also includes S5, which ends loading when the initial height of the reset material level is close to the vehicle's allowable loading height threshold, and corrects the parameters of the loading function based on the loading result. It uses Newton interpolation to locally approximate the function curve between the current material level and the height threshold, estimates when the threshold will be reached, and issues an early warning. If the final result deviates from the prediction, it uses error feedback to update the difference quotient and correct the loading function model.

[0053] After completing the initial loading stage, the system continuously monitors the real-time material level and compares it with the set safe loading limit for vehicles. When the system detects that the current material level is close to the preset threshold, it immediately triggers the loading termination prediction mechanism. The system calls the established loading function relationship (derived from historical deep learning models or real-time fitting functions), selects loading data points for the current time period, and uses Newton's interpolation method to construct curves and estimate slopes in local areas of the material level change function. It predicts the material level change trend in the short term and, combined with time extrapolation, obtains the estimated time point when the threshold is reached, issuing an early warning signal to the control system or operating interface in advance.

[0054] If there is a deviation between the actual material level at the final loading point and the Newton interpolation prediction result, the system automatically calculates the error value and uses this error as a feedback signal to update the interpolation difference quotient table, dynamically correcting the approximate structure of the local interpolation function. Simultaneously, the system will perform parameter backtracking and weight adjustment on the original loading function model based on the direction and magnitude of the error, improving prediction accuracy and response speed in similar loading environments in the future.

[0055] By introducing local interpolation prediction and error feedback mechanisms when the material level approaches the safe loading height threshold, it is possible not only to effectively prevent overflow or safety hazards caused by overloading, but also to improve the accuracy and response sensitivity of the loading end control. In addition, the adaptive correction capability of the function model improves the system's learning efficiency and generalization capability in dynamic and complex environments, making the intelligent loading process more stable, efficient and controllable.

[0056] It also includes S6, which adds the relationship between material level changes and vehicle position changes during the loading process to the deep learning training set, further trains and optimizes the loading speed and function relationship of different vehicle models, and maps different vehicle model numbers, material level features, and position combinations to different training channels through hash mapping.

[0057] In the existing deep learning training system based on the relationship between material level and loading speed, the vehicle's positional information during the loading process, such as latitude and longitude coordinates, steering angle, and truck bed rotation posture, is further introduced. These spatiotemporal positional parameters are then jointly modeled with real-time material level change data and added to the training set as new feature vectors. These composite features not only reflect the dynamic changes in the loading path but also reveal the loading efficiency characteristics of specific vehicle models under specific positional structures.

[0058] To improve the parallelism and feature discriminative power of the training process, the system combines vehicle model number, material location features, and vehicle position status and inputs them into a hash mapping function. This data is then mapped to multiple independent training channels or neural network branches using a distributed encoding strategy, allowing for the separate training of loading behaviors for different vehicle models in the semantic space. Each channel can independently optimize its specific model parameters, thereby giving the model stronger recognition and generalization capabilities.

[0059] The training process employs a neural network structure based on attention mechanisms or gating units, enabling the model to automatically capture the key dependencies between material level and location, and to grasp the changing trends during the loading process through time series modeling, thereby achieving refined modeling of different vehicle loading modes.

[0060] This implementation method, by introducing the joint features of vehicle spatial location and material level changes, not only improves the model's adaptability to complex loading scenarios, but also enhances its ability to characterize the loading efficiency of different vehicle models. Through hash mapping and channel-specific training, it significantly improves the modeling efficiency and generalization accuracy under multi-vehicle conditions, enabling the same system to maintain high-performance operation when dealing with different types of vehicles and working conditions.

[0061] In S3, when training material level change models for different vehicle models, the KL divergence is used to measure the difference between the predicted distribution and the actual material level change distribution. The specific formula is as follows:

[0062] ;

[0063] in, KL divergence represents the degree of difference between model predictions and actual observations. The actual material level change data in the truck bed is on the first The probability of each state. The material level distribution predicted by the network is in the th... The probability of each state. This is the index of the material level distribution range.

[0064] To achieve high-precision fitting of the material level prediction model, this implementation introduces KL divergence as a loss function during the model training phase. After collecting a large amount of material level change data for different vehicle models during the loading process, the system first discretizes the continuous material level height into several state intervals and statistically analyzes the frequency of occurrence in each state to form a true probability distribution. Simultaneously, the currently constructed neural network model, after inputting loading speed, time series, and vehicle parameters, outputs the predicted material level distribution. KL divergence measures the degree of deviation between the two and serves as the objective function during the training process.

[0065] Through backpropagation, if the model output deviates significantly from the actual distribution, the corresponding KL divergence value increases. The optimizer then adjusts the network weight parameters accordingly, causing the predicted distribution to gradually approximate the actual material level change pattern. This process can be repeated iteratively until the model converges, ensuring that the model accurately represents the dynamic characteristics of vehicle material level changes.

[0066] Introducing KL divergence as a loss function enhances the model's ability to capture the true material distribution characteristics, enabling the neural network to learn not only the overall trend but also the micro-distribution features, effectively improving the model's prediction accuracy and robustness. It is suitable for intelligent judgment and real-time feedback under different vehicle models and complex loading paths.

[0067] In S4, the material accumulation caused by changes in the loading position is modeled, and the specific formulas for calculating the resultant force direction and resultant force at each loading point are as follows:

[0068] ;

[0069] in, This represents the sum of the stacking trend vectors at all loading points. Indicates the first Material accumulation trend vector at each loading location Indicates the index number of the loading location point. This indicates the total number of loading points.

[0070] During loading, the vehicle's dynamic position adjustments (such as moving forward, backward, or turning) cause spatial shifts in the loading points, resulting in changes in the material's accumulation direction and distribution trend in the truck bed. To accurately determine whether the current loading status is uniform and to guide subsequent loading strategies, the system models the accumulation trend of each loading point in three-dimensional space.

[0071] Stacking trend vector at each loading point The direction of the total force can be calculated from the material flow direction, the mass loaded per unit time, the stacking angle, and the direction of inertia at that location. This is typically achieved through centroid offset modeling or discrete element method (DEM) simulation. Based on this, the trend vectors of all loading points are summed to obtain the direction of the total resultant force. .

[0072] like If the load is aligned with the center axis of the truck bed, it indicates that the loading is uniform. If it deviates from the center axis, the system will identify it as an off-center loading state and control the vehicle's movement direction or the angle of the loading arm accordingly to correct the loading trajectory and achieve dynamic control of the material accumulation trend.

[0073] By introducing a loading trend resultant force vector model, the system can perceive the loading offset trend in real time during the loading process, avoid the shift of the vehicle's center of gravity caused by unilateral loading, and improve vehicle driving stability. At the same time, it also improves the space utilization efficiency and dynamic control capability of loading, providing key decision-making basis for the unmanned intelligent loading process.

[0074] S2 also includes performing time sequence consistency checks on truck bed images of multiple consecutive time frames during the image recognition stage to eliminate interference from momentary occlusion or abnormal lighting on image entropy calculation.

[0075] In material level detection systems, the truck bed image is a crucial basis for evaluating dust interference and ranging compensation accuracy. To improve the accuracy of image entropy calculation, a temporal consistency verification mechanism is introduced in the image recognition stage. The specific operation process is as follows:

[0076] The system continuously acquires a sequence of images of the truck bed at a set frame rate (e.g., 5 frames per second), and ensures the consistency of the input images through an image preprocessing module (including grayscale conversion, normalization, edge enhancement, etc.). Subsequently, the system compares the brightness, contrast, texture boundary, and other features of pixel regions at the same location within multiple time frames, and uses dynamic time warping or sliding window correlation analysis to determine the consistency of the sequence in terms of visual features.

[0077] If a frame exhibits a significant deviation (such as a sudden increase in brightness or a sudden appearance of an occluded object's outline), the system marks it as an abnormal frame and removes it from the image entropy calculation to avoid the transient interference misleading the dust weight correction coefficient. Furthermore, within each time window, only image frames that have passed consistency verification are retained for subsequent weighted averaging of image entropy, ensuring that the compensation mechanism is based on stable and reliable image features.

[0078] By introducing temporal consistency analysis of image frames, image fluctuations caused by unstructured factors such as instantaneous occlusion, strong light irradiation, and dynamic interference can be effectively filtered out, improving the robustness of image entropy in actual working conditions. This, in turn, makes the ranging radar compensation mechanism more accurate, avoids misjudgment of material level caused by abnormal images, and improves the stability and environmental adaptability of the entire loading system.

[0079] S3 also includes a model memory mechanism introduced during deep learning training to identify and recall recurring loading behavior patterns of specific vehicle models, reducing real-time computation.

[0080] In practical applications, certain vehicle models exhibit highly repetitive loading behavior patterns under similar working conditions, such as stable loading speed, consistent truck bed structure, and similar stress responses. To improve system processing efficiency, this implementation method embeds a model memory mechanism into the deep learning training architecture. By identifying and storing the loading behavior characteristics of typical vehicle models, it achieves rapid response and low computational load inference for repetitive scenarios.

[0081] In the specific implementation process, the system first constructs model memory units (such as Key-Value Memory) during the training phase. Key features such as vehicle number, loading path characteristics, and material level change trends are used as retrieval keys, and the corresponding pre-trained loading prediction model parameters or state vectors are cached as values ​​in the model memory. When the system subsequently identifies a newly entering work vehicle whose basic parameters have a high matching degree with a certain key in the known model (such as Euclidean distance or cosine similarity within the threshold), it skips the complete neural network forward propagation process and directly calls the matching value for fast inference and prediction.

[0082] Furthermore, to prevent memory redundancy and failure, a memory update strategy is employed: parameters of frequently used memory units are dynamically refreshed, and units that have not been matched for a long time are reclaimed to ensure optimal allocation of memory resources. This mechanism is particularly suitable for fixed vehicle models or enterprise-owned fleets with frequent repetitive tasks, improving inference efficiency and reducing hardware consumption.

[0083] By introducing a model memory mechanism, the system can identify and reuse historical high-frequency loading patterns, avoiding repeated model training and high-overhead inference, effectively reducing real-time computing load, and improving model response speed and resource utilization efficiency, especially showing significant advantages in edge computing environments. It also improves the system's accuracy and stability when dealing with known vehicle models.

[0084] S4 also includes intelligent path filtering of the relative displacement between the current loading position and the previous loading point, retaining only the displacement direction that can bring material level optimization benefits.

[0085] During unmanned intelligent loading, the adjustment of the vehicle or loading device position will directly affect the material accumulation structure and material level distribution in the truck bed. To avoid ineffective movement or uneven loading due to improper path planning, this implementation introduces an "intelligent path filtering" mechanism in the S4 loading adjustment stage. This mechanism analyzes and filters the relative displacement directions between consecutive loading points, allowing only paths that have a positive optimization effect on the material level distribution to be executed.

[0086] First, after each loading operation, the system records the current loading position coordinates (e.g., relative X and Y displacement values ​​with the center of the hopper as the reference origin) and the corresponding material level changes (e.g., local accumulation height, slope distribution, central symmetry index, etc.). Before proceeding to the next loading operation, the system calculates the displacement vector between the current position and the previous position, and establishes a dynamic prediction model based on the current material level distribution to evaluate whether continuing loading in this direction can effectively improve the material level center offset, reduce edge over-piling, and improve overall flatness.

[0087] If the prediction results show that the direction can bring benefits to material level optimization (such as the resultant force direction after simulated accumulation approaching the center of gravity of the truck bed), then the path is retained as a candidate movement direction; otherwise, it is regarded as an inefficient path and filtered out to avoid resource waste and accumulation deterioration. The system finally selects the optimal direction from several candidate paths based on a multi-objective optimization strategy to perform displacement adjustment and loading actions.

[0088] By introducing a path filtering mechanism, the loading path is shifted from "blind traversal" to "target-driven", which effectively improves loading efficiency and stacking uniformity, and reduces ineffective actions and energy consumption during the loading process. At the same time, it improves the system's decision-making intelligence and resource scheduling efficiency, and is especially suitable for large mining trucks or engineering vehicles with high load balancing requirements.

[0089] In S6, different vehicle model numbers, material location features, and location combinations are hashed to different training channels using the following formula:

[0090] ;

[0091] in, Indicates the input feature encoding. Indicates the total number of hash buckets. This indicates the training subset number to which the input sample is assigned. This represents the modulo operator.

[0092] When processing large-scale, multi-type vehicle material location data, directly inputting all samples into the same training network channel can easily lead to model overfitting, decreased generalization ability, and training resource conflicts. Therefore, this implementation introduces a hash mapping mechanism in stage S6 to project multi-dimensional features onto different training channels for distributed modeling.

[0093] The specific method involves encoding the feature groups of each input sample (such as vehicle model number, loading position features, and real-time material level change) into integer values. Through hash function Map it to numbers 0 to The training subset. Total number of hash buckets. It can be set according to the system hardware capabilities and data scale. For example, when deployed in a GPU cluster environment, it can be set to the number of parallel channels.

[0094] This hash-based splitting mechanism assigns similar feature samples to the same channel for local modeling, ensuring that different channels do not interfere with each other and each maintains its corresponding model weights. This achieves highly parallel and discriminative feature learning. Simultaneously, the uniformity of the hash function helps ensure a balanced distribution of training samples across subsets, preventing model failure due to sparse channel data.

[0095] Using hash mapping to group training channels can significantly improve training efficiency and model scalability, reduce resource conflicts and memory consumption, and enhance the model's generalization ability and portability in scenarios with multiple vehicle models and material locations. It is suitable for large-scale deployment and accelerated training on heterogeneous platforms.

[0096] like Figure 2 As shown, a material level detection system for unmanned intelligent loading is also provided, comprising the steps of a method for detecting material level during unmanned intelligent loading. The system includes:

[0097] The information acquisition module is used to acquire information on weather, vehicle type, and truck bed size. It uses sensors, manual registration, or ranging radar to collect data and determines whether to enable multi-sensor redundant input based on information entropy.

[0098] The image and radar fusion module is used to acquire images of the truck bed and identify dust concentration during the loading process, and to perform dynamic compensation processing on the ranging radar data by combining humidity and image entropy.

[0099] The model training module is used to build a deep learning model based on the material level changes during the loading process of different vehicles, and optimizes the functional relationship between material level and loading speed and loading position through KL divergence.

[0100] The control decision module is used to determine the current material level status based on the model, control the vehicle position adjustment and update the loading speed, analyze the material flow uniformity based on the resultant force direction of each loading point and dynamically correct the loading trajectory.

[0101] This system employs a modular structure to achieve a closed-loop process encompassing multi-source heterogeneous information perception, fusion analysis, predictive modeling, and intelligent control. By guiding input optimization and compensation strategies through information entropy and image entropy, combined with KL divergence-driven deep model training, the material level detection not only possesses high accuracy and robustness but also enables real-time responses to dynamic loading conditions. The control decision module, centered on a resultant force model, analyzes material accumulation trends and adaptively adjusts the path to ensure the loading operation remains safe, efficient, and uniform.

[0102] This system can automate the entire process from data acquisition, intelligent analysis, model prediction to real-time decision-making, significantly improving the accuracy of material level detection and the intelligence level of loading operations. Its modular design is suitable for distributed deployment, making it easy to promote and apply in large and complex operating environments such as mines and ports. The system has learning and optimization capabilities, and can continuously iterate to improve performance, meeting the needs of long-term high-intensity unmanned operations.

[0103] 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 detecting the level of an unmanned intelligent car loading, characterized in that, The method comprises the following steps: S1, acquiring the current weather condition, vehicle type and hopper size, collecting by sensors or manual registration or ranging radar, and measuring the information redundancy and value density in different data sources by information entropy to automatically determine whether multiple sensor redundant inputs are needed; S2, measuring the material level by the ranging radar, taking pictures of the hopper during loading, and performing image recognition on the pictures, analyzing the current dust concentration according to the image recognition result, and determining the dynamic compensation value of the ranging radar data according to the humidity and dust concentration, and measuring the dust interference degree in the hopper image by image entropy, and using the image entropy as a weight coefficient to adjust the compensation ratio of the ranging radar; S3, taking the hopper material level change of different vehicles during loading as a data set for deep learning, thereby establishing a functional relationship between the hopper material level change and the loading speed and loading position of different vehicles, and using KL divergence to measure the difference between the predicted distribution and the actual hopper material level change distribution as a loss function when training the hopper material level change model of different vehicle types; S4, changing the loading position and replacing the material level by moving the vehicle during loading, determining the hopper material level change according to the functional relationship obtained by training, and deciding the loading speed and whether to stop loading at the next moment, modeling the material accumulation caused by the change of the loading position, calculating the resultant force direction and resultant force of each loading point, analyzing the material flow trend, and aligning the resultant force direction with the hopper center to determine whether the loading is uniform, otherwise adjusting the vehicle position and direction to correct the loading trajectory; Further comprising S5, ending the loading when the initial height of the reset material level approaches the vehicle allowable loading height threshold, and correcting the parameters of the loading function based on the loading result, locally approximating the functional curve between the current material level and the height threshold by Newton interpolation to estimate when the threshold will be reached, and issuing an early warning, and if the final result deviates from the prediction, using error feedback to update the difference quotient and correct the loading function model; Further comprising S6, adding the change relationship between the hopper material level change and the vehicle position during loading to the training set of deep learning, and further training and optimizing the loading speed and functional relationship of different vehicle types, and mapping different vehicle type numbers, material level characteristics and position combinations to different training channels through hash mapping.

2. The method of claim 1, wherein: The specific formula of the KL divergence used to measure the difference between the predicted distribution and the actual hopper material level change distribution in S3 is: ； wherein, is the KL divergence, indicating the difference between the model prediction and the actual observation, is the probability of the actual hopper level change data in the state, is the probability of the hopper level distribution predicted by the network in the state, is the index of the hopper level distribution interval.

3. The method of claim 1, wherein: The specific formula of modeling the material accumulation caused by the change of the loading position, calculating the resultant force direction and resultant force of each loading point in S4 is: ； wherein, represents the sum of the accumulation tendency vectors of all charging points, represents the accumulation tendency vector at the charging position, represents the charging position point index number, represents the total number of charging points.

4. The method of claim 1, wherein: S2 further comprises time sequence consistency verification of the hopper images of multiple consecutive time frames in the image recognition stage, which is used to exclude the interference of instantaneous occlusion or abnormal light on image entropy calculation.

5. The method of claim 1, wherein: S3 further comprises introducing a model memory mechanism in the deep learning training process to identify and remember specific vehicle loading behavior patterns that repeatedly occur, thereby reducing real-time calculation.

6. The method of claim 1, wherein: S4 further comprises intelligent path filtering of the relative displacement between the current loading position and the previous loading point, and only retaining the displacement direction that can bring material level optimization benefits.

7. The method of claim 1, wherein: The S6 will be different vehicle model number, material level characteristics, position combination through hash mapping to different training channel specific formula is: ； wherein, denotes an input feature encoding, denotes the total number of hash buckets, denotes the training subset number to which the input sample is assigned, denotes a modulo operator.

8. A system for detecting the level of a load for unmanned intelligent loading, for implementing the steps of the method for detecting the level of a load for unmanned intelligent loading according to any one of claims 1-7, characterized in that, The system comprises: An information acquisition module is configured to acquire weather, vehicle model, and hopper size information, collect data by using sensors, manual registration, or ranging radar, and determine whether to enable multi-sensor redundant input based on information entropy; An image and radar fusion module is configured to acquire hopper images and identify dust concentration during the loading process, and perform dynamic compensation processing on the ranging radar data in combination with humidity and image entropy; A model training module is configured to construct a deep learning model based on the material level changes during the loading process of different vehicles, and optimize the functional relationship between the material level and the loading speed and loading position by KL divergence; A control decision module is configured to determine the current material level state according to the model, control the vehicle position adjustment and update the loading speed, analyze the material flow uniformity according to the resultant force direction of each loading point, and dynamically correct the loading trajectory.

Images