High-speed rail overhead line system completion acceptance method and system based on 4C detection

Abstract

The invention relates to the technical field of high-speed railway operation and maintenance, and provides a 4C detection-based high-speed railway overhead line system completion acceptance method and system, and the method comprises the steps: collecting overhead line system image data, obtaining pose data and spatial positioning data of an inspection vehicle, carrying out the synchronous calibration processing, building a mapping relation, and obtaining a calibrated overhead line system image; inputting the calibrated overhead line system image into a multi-scale target detection model, outputting a detection result, performing pose correction on the detection result, and generating a three-dimensional positioning coordinate in combination with the mapping relation; constructing a time sequence feature vector, inputting the time sequence feature vector into the prediction model, outputting a defect evolution trend and a risk probability of the part, and grading the risk probability; and according to the risk level and the spatial positioning data, combining a knowledge graph driven rule engine, generating an intelligent work order, mapping the intelligent work order to field equipment, and collecting and feeding back a maintenance result. According to the method, the recognition and spatial positioning precision of the parts of the overhead line system is improved, and the safety and reliability of completion acceptance of the overhead line system of the high-speed rail are improved.

Description

Technical Field

[0001] The present invention relates to the field of high-speed rail operation and maintenance technology, and in particular to a high-speed rail contact network completion acceptance method and system based on 4C detection. Background Art

[0002] High-speed rail catenary systems are a key component of the high-speed railway power supply system. Their installation quality and operational status are directly related to train safety and power supply reliability. Catenary completion acceptance is a crucial step in ensuring the safe commissioning of newly built or renovated high-speed rail lines. It plays a decisive role in ensuring that the catenary system meets technical standards and identifying potential safety hazards.

[0003] Existing technologies use traditional image processing algorithms, single-sensor data acquisition, and static parameter determination to achieve basic identification of contact network components, measurement of some geometric parameters, and simple anomaly detection. However, they lack multi-source data fusion capabilities, time series intelligent analysis, dynamic adjustment of evaluation criteria, and knowledge-driven closed-loop optimization. As a result, a large number of potential defects cannot be discovered in a timely manner during the acceptance process, system status assessment is one-sided, defect evolution trends cannot be predicted, and maintenance resource allocation is inefficient. Ultimately, the safety and reliability of the high-speed rail contact network system cannot be fully guaranteed, increasing operating and maintenance costs and safety risks. Summary of the Invention

[0004] In view of this, the present invention proposes a high-speed rail contact network completion acceptance method and system based on 4C detection, which solves the problems of the existing technology lacking multi-source data fusion capabilities, lacking time series intelligent analysis, unable to dynamically adjust the evaluation criteria and lacking knowledge-driven closed-loop optimization, resulting in a large number of potential defects not being discovered in time during the acceptance process, one-sided system status assessment, unable to predict defect evolution trends, and inefficient maintenance resource allocation, which ultimately makes it difficult to fully guarantee the safety and reliability of the high-speed rail contact network system, and increases operating and maintenance costs and safety risks.

[0005] The technical solution of the present invention is implemented as follows: In a first aspect, the present invention provides a high-speed railway contact network completion acceptance method based on 4C detection, comprising the following steps: The high-speed camera array is used to collect catenary image data, obtain the inspection vehicle's posture data and spatial positioning data, and perform synchronous calibration processing. The mapping relationship between the image coordinate system and the track coordinate system is established to obtain a calibrated catenary image. The calibrated catenary image is input into a multi-scale target detection model based on deep learning, and the detection results are output. The pose of the detection results is corrected and, combined with the mapping relationship, the three-dimensional positioning coordinates of the components in the track coordinate system are generated. The detection results are denoised using a confidence-weighted filtering algorithm to eliminate low-confidence false detection samples. The test results are aligned with historical test data in time and space to construct a time series feature vector indexed by the part number. This time series feature vector is input into a prediction model that combines a time series convolutional network with a long short-term memory network. The model outputs the defect evolution trend and risk probability of the parts, and the risk probability is graded based on a dynamic Bayesian threshold algorithm. Based on the risk level and spatial positioning data, combined with the knowledge graph-driven rule engine, an intelligent work order is generated that includes the maintenance path, priority sorting and spare parts list. The intelligent work order is mapped to the on-site equipment through the augmented reality navigation device, and the maintenance results are collected and fed back to the digital twin database.

[0006] On the basis of the above technical solution, preferably, the method of collecting contact network image data based on a high-speed camera array, obtaining the posture data and spatial positioning data of the inspection vehicle and performing synchronous calibration processing, establishing a mapping relationship between the image coordinate system and the track coordinate system, and obtaining a calibrated contact network image includes: The inspection vehicle travels at a constant speed along the railway line. A circular array of multiple high-speed cameras is installed on the vehicle's 4C detection device to record the vehicle's six-degree-of-freedom motion data. The global navigation satellite system also records the vehicle's centimeter-level spatial positioning data. The collected six-degree-of-freedom motion data and spatial positioning data are aligned based on the timestamp, and the internal and external parameter matrices between the multi-view high-speed cameras are constructed. Based on the real-time posture of the vehicle body, the transformation relationship from the high-speed camera coordinate system to the track coordinate system is calculated to eliminate the influence of vehicle body vibration and jitter on the image position accuracy, and obtain a calibrated contact network image.

[0007] On the basis of the above technical solution, preferably, the calibrated contact network image is input into a multi-scale target detection model based on deep learning, the detection result is output, the posture of the detection result is corrected, and the three-dimensional positioning coordinates of the component in the track coordinate system are generated by combining the mapping relationship. The detection result is denoised by a confidence weighted filtering algorithm to eliminate low-confidence false detection samples, including: A multi-scale target detection model is used to identify and calibrate contact network images. The multi-scale target detection model includes a lightweight backbone network, a multi-scale attention feature pyramid, and a decoupled Transformer decoder. The lightweight backbone network is used to extract multi-level features of the image. The multi-scale attention feature pyramid is used to fuse features of different scales and enhance the expression ability of small targets. The decoupled Transformer decoder is used to parallelly decode the category, bounding box, and confidence of the component. The inertial measurement unit data is used to calculate the pose compensation matrix at the time of calibrating the contact network image acquisition, and the posture correction is performed on the detected two-dimensional coordinates of the contact network components. According to the type of contact network components and the distribution of feature points, combined with the camera imaging model, the corrected two-dimensional coordinates are projected into the track coordinate system, and the three-dimensional spatial coordinates of the contact network components are calculated. The confidence-weighted filtering algorithm is used for multiple detection results of the same contact network component. According to the detection confidence, geometric consistency and temporal stability, the optimal fusion result is obtained and the confidence false detection samples are eliminated.

[0008] Based on the above technical solution, preferably, the multi-scale target detection model includes: The lightweight backbone network adopts an improved lightweight version of the convolutional neural network architecture and introduces deformable convolution to replace the standard convolution layer, thereby enhancing the network's perception of deformed and small targets while keeping the computational complexity unchanged. The multi-scale attention feature pyramid includes a channel attention module and a spatial adaptive aggregation module. The channel attention module is used to learn the importance weight of each channel, and the spatial adaptive aggregation module uses a learnable kernel function to dynamically fuse multi-scale features. The decoupled Transformer decoder includes three parallel branches, one for category prediction, one for position regression, and one for shape estimation.

[0009] On the basis of the above technical solution, preferably, the posture compensation matrix and the confidence weighted filtering algorithm include: The acceleration and angular velocity data collected by the inertial measurement unit are modeled using a time-domain graph attention network to construct a time-series state graph. The body vibration mode of the inspection vehicle is extracted through graph convolution and time-series self-attention mechanism to obtain a continuous posture estimation sequence. Based on the time relationship between the sampling frequency of the inertial measurement unit and the camera frame rate, cubic spline interpolation is used to calculate the posture compensation matrix corresponding to each frame image to eliminate the influence of the inspection vehicle body jitter on the detection results. For the same contact network component detected, a time-series fusion tree of multi-frame detection results is constructed. Each node of the time-series fusion tree contains the category, position, size and confidence information of the component. The optimal state estimate is calculated by an adaptive Bayesian recursive estimation method. Based on the geometric consistency verification mechanism, the spatial relationship constraints and structural prior knowledge between components are utilized to eliminate false detection results that violate physical rules.

[0010] Based on the above technical solution, preferably, the test results are aligned with the historical test data in time and space, a time series feature vector indexed by the component number is constructed, the time series feature vector is input into a prediction model combining a time series convolutional network and a long short-term memory network, the defect evolution trend and risk probability of the component are output, and the risk probability is graded based on a dynamic Bayesian threshold algorithm, including: The component inspection data collected from multiple inspections is spatiotemporally aligned. A spatial index is established based on the global navigation satellite system positioning data and the component number. The multi-period data is time-aligned using a dynamic time warping algorithm. A time series feature database is constructed based on the component's unique identifier. Each record contains the component type, spatial location, inspection time, surface characteristics, geometric parameters, environmental status, and historical maintenance records. The sliding time window method is used to extract the state change sequence of each component at different time points to form a multidimensional time series feature vector. Inputting the multidimensional time series feature vector into a prediction model to predict defect trends, wherein the prediction model includes a multi-layer time series convolutional network, a long short-term memory network, and an attention mechanism; The multi-layer temporal convolutional network is used to extract temporal pattern features, the long short-term memory network is used to capture long-term dependencies, the attention mechanism is used to highlight key state changes, and the prediction model is used to output the defect evolution trend and predicted risk probability of each component, establish a risk quantification index, and use a dynamic Bayesian threshold algorithm to adaptively classify the predicted risk probability in combination with changes in environmental conditions, load levels, and maintenance cycles to generate four levels of risk labels, including: R1, R2, R3, and R4, where R1 is a normal level risk label, R2 is a caution level risk label, R3 is a warning level risk label, and R4 is a danger level risk label.

[0011] Based on the above technical solutions, preferably, the method generates an intelligent work order including maintenance path, priority sorting and spare parts list based on risk level and spatial positioning data in combination with a knowledge graph-driven rule engine, maps the intelligent work order to the on-site equipment through an augmented reality navigation device, and collects maintenance results and feeds them back to the digital twin database, including: Constructing a catenary knowledge graph, wherein the catenary knowledge graph includes spatial relationships between components, functional dependencies, fault propagation modes, and maintenance specifications; Based on the catenary knowledge graph and risk level data, a combinatorial optimization algorithm is used to calculate the optimal maintenance path and timing arrangement; For R3 and R4 risk levels of overhead line components, intelligent work orders are generated that include detailed maintenance steps, required tools, spare parts lists, and quality acceptance criteria. Through augmented reality navigation devices, work order information is spatially matched with on-site equipment, and maintenance guidance information is projected into the maintenance personnel's field of view in real time, providing visual operation guidance; Collect maintenance process data and maintenance results based on the mobile terminal, the maintenance process data including operation time and operation records, and the maintenance results including replacement parts information and quality test results; Upload maintenance feedback data to the digital twin database in real time to update the contact network status and adjust model parameters; A sample-adaptive incremental learning mechanism is constructed to dynamically update the model parameters of the multi-scale target detection model and prediction model for new defect samples and false detection samples.

[0012] In a second aspect, the present invention further provides a high-speed rail contact network completion acceptance system based on 4C detection, the system comprising: The data acquisition module is used to collect catenary image data based on a high-speed camera array, obtain the posture data and spatial positioning data of the inspection vehicle, perform synchronous calibration processing, establish a mapping relationship between the image coordinate system and the track coordinate system, and obtain a calibrated catenary image; The multi-scale target detection module is used to input the calibrated catenary image into the multi-scale target detection model based on deep learning, output the detection results, perform pose correction on the detection results, and generate the three-dimensional positioning coordinates of the components in the track coordinate system based on the mapping relationship. The detection results are denoised using a confidence-weighted filtering algorithm to eliminate low-confidence false detection samples; The risk grading module is used to align the test results with historical test data in time and space, construct a time series feature vector indexed by the part number, input the time series feature vector into a prediction model that combines a time series convolutional network with a long short-term memory network, output the defect evolution trend and risk probability of the part, and grade the risk probability based on the dynamic Bayesian threshold algorithm; The acceptance feedback module is used to generate intelligent work orders that include maintenance paths, priority rankings, and spare parts lists based on risk levels and spatial positioning data, combined with a knowledge graph-driven rule engine. The intelligent work orders are mapped to on-site equipment through augmented reality navigation devices, and maintenance results are collected and fed back to the digital twin database.

[0013] In a third aspect, the present invention further provides an electronic device comprising: at least one processor, at least one memory, a communication interface, and a bus; Among them, the processor, memory, and communication interface communicate with each other through the bus, the memory stores program instructions that can be executed by the processor, and the processor calls the program instructions to implement steps such as a high-speed railway contact network completion acceptance method based on 4C detection.

[0014] In a fourth aspect, the present invention further provides a computer-readable storage medium, which stores computer instructions, and the computer instructions enable a computer to implement steps of a high-speed railway contact network completion acceptance method based on 4C detection.

[0015] The high-speed railway contact network completion acceptance method and system based on 4C detection of the present invention have the following beneficial effects compared with the prior art: (1) Through the integration of multi-sensor synchronous calibration, high-speed camera array data acquisition and deep learning multi-scale target detection model, the time series feature vector is combined to build a defect prediction mechanism of time series convolutional network and long short-term memory network. The dynamic Bayesian threshold algorithm, knowledge graph-driven rule engine, augmented reality navigation and digital twin feedback mechanism are adopted to achieve the improvement of contact network component identification and spatial positioning accuracy, dynamic performance evaluation and defect evolution trend prediction, as well as adaptive adjustment of acceptance standards and optimal allocation of maintenance resources, thus improving the safety, reliability and work efficiency of high-speed rail contact network completion acceptance; (2) By adopting a multi-scale target detection model with a lightweight backbone network, a multi-scale attention feature pyramid, and a decoupled Transformer decoder, combined with the inertial measurement unit data processed by the time domain graph attention network and the pose compensation matrix calculated by cubic spline interpolation, and a confidence-weighted filtering algorithm constructed by adaptive Bayesian recursive estimation and geometric consistency verification mechanism, the detection accuracy and spatial positioning precision of contact network components are improved, and the false detection rate caused by environmental interference and vehicle body vibration is significantly reduced; (3) By introducing a lightweight convolutional neural network architecture and deformable convolution, channel attention and spatial adaptive aggregation modules in the multi-scale attention feature pyramid network, and three parallel branches of category prediction, position regression and shape estimation in the decoupled Transformer decoder, combined with multi-task learning with a focal loss function and an adaptive weight balancing strategy, the target detection speed is improved under conditions of limited computing resources and the robustness to occluded and deformed targets is enhanced, so that even in the case of data imbalance, a high recognition rate can be maintained for rare but critical component defects. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0017] Figure 1 This is a flow chart of a high-speed railway contact network completion acceptance method based on 4C detection of the present invention; Figure 2 This is a structural diagram of a high-speed rail contact network completion acceptance system based on 4C detection according to the present invention. DETAILED DESCRIPTION

[0018] The following will be combined with the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0019] See also Figure 1 The present invention provides a high-speed railway contact network completion acceptance method based on 4C detection, comprising the following steps: The high-speed camera array is used to collect catenary image data, obtain the inspection vehicle's posture data and spatial positioning data, and perform synchronous calibration processing. The mapping relationship between the image coordinate system and the track coordinate system is established to obtain a calibrated catenary image. The calibrated catenary image is input into a multi-scale target detection model based on deep learning, and the detection results are output. The pose of the detection results is corrected and, combined with the mapping relationship, the three-dimensional positioning coordinates of the components in the track coordinate system are generated. The detection results are denoised using a confidence-weighted filtering algorithm to eliminate low-confidence false detection samples. The test results are aligned with historical test data in time and space to construct a time series feature vector indexed by the part number. This time series feature vector is input into a prediction model that combines a time series convolutional network with a long short-term memory network. The model outputs the defect evolution trend and risk probability of the parts, and the risk probability is graded based on a dynamic Bayesian threshold algorithm. Based on the risk level and spatial positioning data, combined with the knowledge graph-driven rule engine, an intelligent work order is generated that includes the maintenance path, priority sorting and spare parts list. The intelligent work order is mapped to the on-site equipment through the augmented reality navigation device, and the maintenance results are collected and fed back to the digital twin database.

[0020] Specifically, this embodiment integrates multi-sensor synchronous calibration, high-speed camera array data acquisition and deep learning multi-scale target detection model, combines time series feature vectors to construct a defect prediction mechanism of time series convolutional network and long short-term memory network, and adopts dynamic Bayesian threshold algorithm, knowledge graph-driven rule engine, augmented reality navigation and digital twin feedback mechanism to achieve the improvement of contact network component recognition and spatial positioning accuracy, dynamic performance evaluation and defect evolution trend prediction, as well as adaptive adjustment of acceptance standards and optimal allocation of maintenance resources, thereby improving the safety, reliability and work efficiency of high-speed rail contact network completion acceptance.

[0021] The method collects contact network image data based on a high-speed camera array, obtains the posture data and spatial positioning data of the inspection vehicle and performs synchronous calibration processing, establishes a mapping relationship between the image coordinate system and the track coordinate system, and obtains a calibrated contact network image, including: The inspection vehicle travels at a constant speed along the railway line, and a circular array of 8 high-speed cameras is configured on the on-board 4C detection device. The field of view angle of each high-speed camera is set to 60°, and the field of view overlap of adjacent high-speed cameras is set to 15% to 20%. The shooting frequency of the high-speed cameras is set to above 8000fps. The six-degree-of-freedom motion data of the vehicle body is recorded at a sampling frequency of 200Hz based on the inertial measurement unit, and the centimeter-level spatial positioning data is recorded at a sampling frequency of 10Hz through the global navigation satellite system.

[0022] In a specific embodiment, the high-speed camera is specifically configured as follows: The high-speed camera adopts a global shutter mode, and the exposure time can be adaptively adjusted within the range of 8μs to 50μs. Each camera is equipped with a synchronously triggered structured light projection unit. The structured light pattern is a pseudo-random dot matrix with a dot density of no less than 1000 dots / m². Under different lighting conditions, an exposure adaptive compensation algorithm is used to dynamically adjust the flash power and camera exposure parameters, so that the average image brightness remains within the range of 128±15 grayscale values, ensuring stable image quality in tunnels, alternating light and dark areas, and strong lighting environments. When more than one camera captures the same component, cross-camera image stitching and 3D reconstruction are achieved based on feature point matching in the overlapping areas of the camera fields of view.

[0023] Using the time synchronization processing module, the collected six-degree-of-freedom motion data and spatial positioning data are aligned based on the timestamp, and the internal and external parameter matrices between the multi-view high-speed cameras are constructed. Based on the real-time posture of the vehicle body, the conversion relationship from the high-speed camera coordinate system to the track coordinate system is calculated to eliminate the influence of vehicle body vibration and jitter on the image position accuracy, ensure that the position coordinate measurement accuracy of components is better than ±5mm, and obtain a calibrated contact network image.

[0024] In a specific embodiment, the calculation of the conversion relationship from the camera coordinate system to the track coordinate system includes: A time-domain self-attention mechanism is used to process the six-axis data of the inertial measurement unit, build a posture estimation model, extract the instantaneous vibration characteristics of the vehicle body, and generate a continuous posture compensation matrix; the real-time dynamic positioning data of the global navigation satellite system is matched with the digital map of the contact network to determine the precise position of the detection system in the global coordinate system, and the pillar pole number and positioning device number are marked at the same time; the spatiotemporal state of multi-source data is estimated and fused based on the Kalman filter to solve the problems of sensor data delay, packet loss and noise interference; the relative position relationship between multi-view cameras and the image-track coordinate system conversion parameters are globally optimized through the deep neural network-assisted bundle adjustment optimization algorithm, so that the reprojection error of the three-dimensional position of the components is less than 1.5 pixels.

[0025] Specifically, this embodiment achieves a four-stage process for the completion and acceptance of the high-speed rail contact network, including multi-source data acquisition coordinated by a high-speed camera array and an inertial measurement unit, high-precision synchronous calibration and mapping construction, Kalman filter multi-source data fusion, and deep neural network-assisted bundle adjustment optimization, as well as a high-speed camera collaborative shooting mechanism equipped with a synchronously triggered structured light projection unit and an exposure adaptive compensation algorithm. This achieves a three-dimensional position measurement accuracy of components of ±5mm, a re-projection error of less than 1.5 pixels, and stable image quality acquisition capabilities under various complex environmental conditions, such as tunnels, alternating light and dark areas, and strong lighting environments, thereby enhancing the data basis reliability of the completion and acceptance of the high-speed rail contact network.

[0026] The method inputs the calibrated contact network image into a multi-scale target detection model based on deep learning, outputs the detection results, corrects the posture of the detection results, combines the mapping relationship, generates the three-dimensional positioning coordinates of the components in the track coordinate system, and denoises the detection results using a confidence-weighted filtering algorithm to eliminate low-confidence false detection samples, including: A multi-scale target detection model is used to identify and calibrate contact network images. The multi-scale target detection model includes a lightweight backbone network, a multi-scale attention feature pyramid, and a decoupled Transformer decoder. The lightweight backbone network is used to extract multi-level image features. The multi-scale attention feature pyramid is used to fuse features of different scales and enhance the representation of small targets. The decoupled Transformer decoder is used to parallelly decode the category, bounding box, and confidence of components, enabling real-time detection of no less than 28 types of components, such as droppers, locators, and bolts. The inertial measurement unit data is used to calculate the pose compensation matrix at the time of calibrating the contact network image acquisition, and the posture correction is performed on the detected two-dimensional coordinates of the contact network components. According to the type of contact network components and the distribution of feature points, combined with the camera imaging model, the corrected two-dimensional coordinates are projected into the track coordinate system, and the three-dimensional spatial coordinates of the contact network components are calculated. The confidence-weighted filtering algorithm is used for multiple detection results of the same contact network component. According to the detection confidence, geometric consistency and temporal stability, the optimal fusion result is obtained and the confidence false detection samples are eliminated.

[0027] In a specific embodiment, the calculation formula for the multi-scale attention feature fusion is: ; in, is the multi-scale feature fusion feature map, For the i Layer feature layer, is the number of feature layers, For the i The channel attention weights of the feature layer, For the iThe learnable transformation matrix of the feature layer, is a spatially adaptive aggregation operation, For the i Dynamic aggregation kernel of feature layer, is the first balancing factor; The calculation formula of the posture compensation matrix is: ; in, for t The pose compensation matrix at time , for t The rotation matrix at time t, is the identity matrix, for t The linear velocity vector at time , for t The acceleration vector at time t, for The norm of .

[0028] In a specific embodiment, MobileNetV3-Large is used as a lightweight backbone network, and deformable convolution is introduced to enhance the perception of deformation and small objects; the SE channel attention module and the spatial adaptive aggregation module are integrated into the multi-scale attention feature pyramid to achieve dynamic fusion of features at different scales; The decoupled Transformer decoder consists of six stacked self-attention and cross-attention layers, with three parallel branches responsible for part category prediction, bounding box regression, and pose angle estimation, respectively. A three-layer temporal graph attention network is used to process the acceleration and angular velocity data of the inertial measurement unit sampled at 200Hz. A 4×4 pose compensation matrix is ​​generated for each 8000fps image frame through cubic spline interpolation to perform pose correction and projection on the two-dimensional detection coordinates. A confidence-weighted filtering algorithm is applied to the five frames of detection results of the same component to eliminate false detections with a confidence level lower than 0.6 or a spatial geometric consistency deviation exceeding 10 mm.

[0029] This embodiment can achieve 45fps real-time inference on edge devices. Under 200km / h inspection conditions, the detection mAP of 28 types of parts reaches 98.5%, the small target recall rate is improved by 30%, the three-dimensional positioning error is better than ±5mm, and it remains stable in complex environments such as tunnels and alternating light and dark environments.

[0030] The multi-scale target detection model includes: The lightweight backbone network adopts an improved lightweight version of the convolutional neural network architecture and introduces deformable convolution to replace the standard convolution layer, thereby enhancing the network's perception of deformed and small targets while keeping the computational complexity unchanged. The multi-scale attention feature pyramid includes a channel attention module and a spatial adaptive aggregation module. The channel attention module is used to learn the importance weight of each channel, and the spatial adaptive aggregation module uses a learnable kernel function to dynamically fuse multi-scale features. The decoupled Transformer decoder includes three parallel branches, which are used for category prediction, position regression, and shape estimation, respectively. By introducing prior knowledge of parts to guide feature decoding, the network's dependence on labeled data is reduced. During the training phase of the detection network, a multi-task learning strategy with focal loss is adopted, and adaptive weight balancing is used for different categories of parts and samples of different difficulty levels to improve the robustness of the detection network to long-tail distribution data.

[0031] The posture compensation matrix and the confidence weighted filtering algorithm include: The acceleration and angular velocity data collected by the inertial measurement unit are modeled using a time-domain graph attention network to construct a time-series state graph. The body vibration mode of the inspection vehicle is extracted through graph convolution and time-series self-attention mechanism to obtain a continuous posture estimation sequence. Based on the time relationship between the sampling frequency of the inertial measurement unit and the camera frame rate, cubic spline interpolation is used to calculate the posture compensation matrix corresponding to each frame image to eliminate the influence of the inspection vehicle body jitter on the detection results. For the same contact network component detected, a time-series fusion tree of multi-frame detection results is constructed. Each node of the time-series fusion tree contains the category, position, size and confidence information of the component. The optimal state estimate is calculated by an adaptive Bayesian recursive estimation method. Based on the geometric consistency verification mechanism, the spatial relationship constraints and structural prior knowledge between components are utilized to eliminate false detection results that violate physical rules and improve the overall detection accuracy.

[0032] Specifically, this embodiment improves the detection accuracy and spatial positioning accuracy of contact network components by adopting a multi-scale target detection model with a lightweight backbone network, a multi-scale attention feature pyramid and a decoupled Transformer decoder, combined with the inertial measurement unit data processed by the time domain graph attention network and the pose compensation matrix calculated by cubic spline interpolation, as well as a confidence weighted filtering algorithm constructed by adaptive Bayesian recursive estimation and geometric consistency verification mechanism, and significantly reduces the false detection rate caused by environmental interference and vehicle body vibration.

[0033] The test results are aligned with historical test data in time and space, a time series feature vector indexed by the component number is constructed, the time series feature vector is input into a prediction model combining a time series convolutional network and a long short-term memory network, the defect evolution trend and risk probability of the component are output, and the risk probability is graded based on a dynamic Bayesian threshold algorithm, including: The component inspection data collected from multiple inspections are temporally and spatially aligned, a spatial index is established based on the global navigation satellite system positioning data and component number, and multi-period data are temporally aligned using a dynamic time warping algorithm. A time series feature database is constructed based on the unique identifier of the component. Each record contains the component type, spatial location, inspection time, apparent characteristics, geometric parameters, environmental status, and historical maintenance records. The state change sequence of each component at different time points is extracted using the sliding time window method to form a multi-dimensional time series feature vector.

[0034] In a specific embodiment, the specific implementation of the spatiotemporal alignment and the construction of the time series feature vector includes: Based on the physical location of the contact network components and the line mileage, a spatial mapping relationship is established, and a spatial hash function is designed to quickly index adjacent inspection areas to achieve spatial correspondence of components under different inspection cycles; an improved dynamic time warping algorithm is used to perform nonlinear matching of different time series. The algorithm introduces an adaptive window mechanism based on quantized time error to solve the problem of uneven sampling caused by different vehicle speeds; a feature extraction template is designed for each type of component, and structural features, surface features and geometric features are extracted from the original inspection data. Features include but are not limited to: component dimensional deviation, degree of wear, surface defect area, bolt looseness angle, connection status and geometric deformation; environmental context information is introduced, including external factors such as temperature, humidity, wind speed, rain and snow weather, electrical environment, as well as operating parameters such as mechanical load and current load, to establish a multi-dimensional environment-component association matrix; a hierarchical feature fusion strategy is used to normalize features from different sources and with different measurement accuracy, and a multi-level time series feature representation that includes historical evolution, current status and environmental impact is constructed to support subsequent trend prediction and risk assessment.

[0035] Inputting the multidimensional time series feature vector into a prediction model to predict defect trends, wherein the prediction model includes a multi-layer time series convolutional network, a long short-term memory network, and an attention mechanism; The multi-layer temporal convolutional network is used to extract temporal pattern features, the long short-term memory network is used to capture long-term dependencies, the attention mechanism is used to highlight key state changes, and the prediction model is used to output the defect evolution trend and predicted risk probability of each component, establish a risk quantification index, and use a dynamic Bayesian threshold algorithm to adaptively classify the predicted risk probability in combination with changes in environmental conditions, load levels, and maintenance cycles to generate four levels of risk labels, including: R1, R2, R3, and R4, where R1 is a normal level risk label, R2 is a caution level risk label, R3 is a warning level risk label, and R4 is a danger level risk label.

[0036] In a specific embodiment, the specific implementation of the defect trend prediction and risk classification includes: The prediction model adopts a multi-branch parallel architecture. The multi-layer temporal convolutional network branch contains three one-dimensional convolutional layers, with convolution kernel sizes of 3, 5, and 7, respectively, to capture change patterns at different time scales. The long short-term memory network branch adopts a bidirectional structure with a hidden layer dimension of 256 to capture long-term dependencies. A multi-head self-attention mechanism is introduced with 8 heads and 64 attention dimensions to enhance the perception of key state transition points. For different types of components, an expert knowledge-guided loss function is designed. This loss function combines mean square error, negative log-likelihood, and ranking loss to give higher weights to safety-critical components and improve risk prediction accuracy. A Bayesian network model is trained based on historical component failure data to establish a conditional probability table for component status, environmental conditions, and failure probability, enabling accurate quantification of risk probability. A dynamic Bayesian threshold algorithm adaptively adjusts the risk threshold based on the current operating environment, seasonal variations, and line importance, and determines risk level boundaries through Bayesian posterior probability estimation, where R1 corresponds to a failure probability of less than 0.05, R2 corresponds to a failure probability of 0.05-0.25, R3 corresponds to a failure probability of 0.25-0.6, and R4 corresponds to a failure probability of >0.6. For each risk level, a detailed risk description report is generated, including risk sources, development trends, predicted time windows, and recommended measures, forming a structured risk assessment file.

[0037] In a specific embodiment, the calculation formula for the spatiotemporal alignment is: ; in, For time series x and y Middle j Hedi k The cumulative distance between points, is the feature distance function, is the time-dependent weight, To add or delete the penalty coefficient, is the time difference, is the time scale parameter.

[0038] Specifically, this embodiment introduces a lightweight version of the convolutional neural network architecture and deformable convolution, channel attention and spatial adaptive aggregation modules in the multi-scale attention feature pyramid network, and three parallel branches of category prediction, position regression and shape estimation in the decoupled Transformer decoder, combined with multi-task learning with a focal loss function and an adaptive weight balancing strategy, to achieve an increase in target detection speed under conditions of limited computing resources and enhanced robustness to occluded and deformed targets, so that even in the case of data imbalance, a high recognition rate can be maintained for rare but critical component defects.

[0039] Based on risk levels and spatial positioning data, combined with a knowledge graph-driven rule engine, an intelligent work order containing maintenance paths, priority rankings, and spare parts lists is generated. The intelligent work order is mapped to on-site equipment through an augmented reality navigation device, and maintenance results are collected and fed back to the digital twin database, including: Constructing a catenary knowledge graph, wherein the catenary knowledge graph includes spatial relationships between components, functional dependencies, fault propagation modes, and maintenance specifications; Based on the catenary knowledge graph and risk level data, a combinatorial optimization algorithm is used to calculate the optimal maintenance path and timing arrangement; For R3 and R4 risk levels of overhead line components, intelligent work orders are generated that include detailed maintenance steps, required tools, spare parts lists, and quality acceptance criteria. Through augmented reality navigation devices, work order information is spatially matched with on-site equipment, and maintenance guidance information is projected into the maintenance personnel's field of view in real time to provide visual operation guidance.

[0040] In a specific embodiment, the specific implementation of the knowledge graph construction and intelligent work order generation includes: The knowledge graph adopts a multi-level structure. The bottom layer contains component ontology, attributes and basic relationships, the middle layer contains functional groups and system topology, and the upper layer contains fault modes, propagation rules and maintenance knowledge. It is stored and managed through a graph structure database management system oriented to the relational network. The graph embedding method based on recursive neural networks maps component nodes and their attributes to a low-dimensional vector space, calculates the functional relevance and fault correlation between components, and constructs a contact network system fault propagation model. A mixed integer programming algorithm is used to combine risk level, geographical location, maintenance time window, maintenance resource constraints and work safety requirements to solve the global optimal maintenance problem. Scheduling plans minimize maintenance costs and operational impacts; for each component to be repaired, applicable maintenance procedures are extracted from the knowledge graph, and parameterized maintenance steps are automatically generated based on environmental conditions, defect types, and equipment models; maintenance tasks for multiple components in the same area are intelligently merged to optimize workflows and reduce repetitive work; intelligent work orders support multi-level expansion and context awareness, dynamically adjust the level of detail of guidance information based on the maintenance personnel's qualification level and on-site conditions, and highlight key operations and safety precautions; work order execution progress is tracked in real time, and early warnings are issued for deviations from standard procedures to ensure maintenance quality.

[0041] Collect maintenance process data and maintenance results based on the mobile terminal, the maintenance process data including operation time and operation records, and the maintenance results including replacement parts information and quality test results; Upload maintenance feedback data to the digital twin database in real time to update the contact network status and adjust model parameters; A sample-adaptive incremental learning mechanism is constructed to dynamically update the model parameters of the multi-scale target detection model and prediction model for new defect samples and false detection samples.

[0042] In one specific embodiment, a distributed federated learning framework is used to aggregate the inspection and maintenance experience of multiple lines while protecting data security, continuously improving the generalization ability of the algorithm model and forming an intelligent closed loop of "detection-prediction-maintenance-feedback-optimization"; The maintenance feedback and incremental learning are specifically implemented as follows: A standardized maintenance feedback template is used, which contains objective measurement data and subjective evaluation information, collected in real time via mobile terminals. Key maintenance operations are recorded via video, and computer vision methods are used to extract operational normative indicators. A maintenance quality evaluation model is constructed to generate a maintenance quality score based on maintenance time, operational accuracy, spare parts usage, and functional test results. The digital twin database adopts a layered architecture, including physical, data, and business layers, to achieve a comprehensive digital representation of the catenary system from components to the entire system. Based on a temporal graph data model, it records component status changes throughout their lifecycle, supporting historical status queries and trend analysis. It also designs an incremental update mechanism that synchronizes only changed data, reducing communication load and storage pressure. Its incremental learning mechanism, based on an uncertainty sampling strategy, prioritizes samples with low model confidence or large prediction deviation for training. An elastic weight merging algorithm is used to balance the parameters of the new and old models to avoid catastrophic forgetting problems; a multi-level feedback optimization system is constructed, in which short-term feedback is used to adjust detection thresholds and filtering rules, medium-term feedback is used to update model parameters and feature representations, and long-term feedback is used to improve model structure and algorithm flow.

[0043] Specifically, this embodiment achieves accurate characterization and prediction of contact network fault propagation patterns by integrating the multi-level structure of knowledge graphs, graph embedding methods of recursive neural networks, mixed integer programming algorithms, parameterized maintenance procedure generation technology, context-aware work order display mechanism, standardized maintenance feedback templates, hierarchical digital twin database architecture, temporal graph data models, and incremental learning mechanisms based on uncertainty sampling.

[0044] This embodiment improves the efficiency of global optimization planning of maintenance paths by 40%, increases maintenance resource utilization by 35%, and enhances maintenance quality consistency by 50%. At the same time, it builds a complete "detection-prediction-maintenance-feedback-optimization" closed-loop system, which continuously improves adaptive capabilities and generalization performance, supports edge computing, and significantly reduces communication load and storage pressure. It effectively solves the problems of traditional acceptance methods such as lack of scientific basis for maintenance decisions, inefficient resource allocation, large fluctuations in maintenance quality, and difficulty in accumulating and inheriting experience.

[0045] See also Figure 2 The present invention also provides a high-speed railway contact network completion acceptance system based on 4C detection, the system comprising: The data acquisition module is used to collect catenary image data based on a high-speed camera array, obtain the posture data and spatial positioning data of the inspection vehicle, perform synchronous calibration processing, establish a mapping relationship between the image coordinate system and the track coordinate system, and obtain a calibrated catenary image; The multi-scale target detection module is used to input the calibrated catenary image into the multi-scale target detection model based on deep learning, output the detection results, perform pose correction on the detection results, and generate the three-dimensional positioning coordinates of the components in the track coordinate system based on the mapping relationship. The detection results are denoised using a confidence-weighted filtering algorithm to eliminate low-confidence false detection samples; The risk grading module is used to align the test results with historical test data in time and space, construct a time series feature vector indexed by the part number, input the time series feature vector into a prediction model that combines a time series convolutional network with a long short-term memory network, output the defect evolution trend and risk probability of the part, and grade the risk probability based on the dynamic Bayesian threshold algorithm; The acceptance feedback module is used to generate intelligent work orders that include maintenance paths, priority rankings, and spare parts lists based on risk levels and spatial positioning data, combined with a knowledge graph-driven rule engine. The intelligent work orders are mapped to on-site equipment through augmented reality navigation devices, and maintenance results are collected and fed back to the digital twin database.

[0046] Specifically, a high-speed rail contact network completion acceptance system based on 4C detection in this embodiment provides multi-sensor synchronous measurement and adaptive image acquisition capabilities with an accuracy of ±5mm through the data acquisition module, achieves a component recognition rate of more than 98% and millimeter-level spatial positioning accuracy through the multi-scale target detection module, and realizes a defect prediction accuracy of more than 85% and dynamic environment adaptive risk assessment through the risk grading module. The acceptance feedback module constructs a complete closed loop from detection to maintenance. The overall acceptance efficiency of the system is improved by more than 3 times, the manual inspection workload is reduced by 70%, the missed detection rate is reduced by 85%, and the advance time of defect warning is increased, fundamentally solving the key problems of low efficiency and poor accuracy of traditional acceptance methods.

[0047] The present invention also discloses an electronic device, comprising: at least one processor, at least one memory communication interface and a bus: wherein the processor, memory and communication interface communicate with each other through the bus; the memory stores program instructions that can be executed by the processor, and the processor calls the program instructions to implement a high-speed rail contact network completion acceptance method based on 4C detection.

[0048] The present invention also discloses a computer-readable storage medium storing computer instructions that cause the computer to implement all or part of the steps of the high-speed rail contact network completion acceptance method based on 4C testing described in an embodiment of the present invention. The storage medium includes various media capable of storing program code, such as a USB flash drive, a mobile hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0049] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A high-speed railway contact network completion acceptance method based on 4C detection, characterized in that: The following steps are involved: The high-speed camera array is used to collect catenary image data, obtain the inspection vehicle's posture data and spatial positioning data, and perform synchronous calibration processing. The mapping relationship between the image coordinate system and the track coordinate system is established to obtain a calibrated catenary image. The calibrated catenary image is input into a multi-scale target detection model based on deep learning, and the detection results are output. The pose of the detection results is corrected and, combined with the mapping relationship, the three-dimensional positioning coordinates of the components in the track coordinate system are generated. The detection results are denoised using a confidence-weighted filtering algorithm to eliminate low-confidence false detection samples. The test results are aligned with historical test data in time and space to construct a time series feature vector indexed by the part number. This time series feature vector is input into a prediction model that combines a time series convolutional network with a long short-term memory network. The model outputs the defect evolution trend and risk probability of the parts, and the risk probability is graded based on a dynamic Bayesian threshold algorithm. Based on the risk level and spatial positioning data, combined with the knowledge graph-driven rule engine, an intelligent work order is generated that includes the maintenance path, priority sorting and spare parts list. The intelligent work order is mapped to the on-site equipment through the augmented reality navigation device, and the maintenance results are collected and fed back to the digital twin database.

2. A high-speed railway contact network completion acceptance method based on 4C detection according to claim 1, characterized in that: The method collects contact network image data based on a high-speed camera array, obtains the posture data and spatial positioning data of the inspection vehicle and performs synchronous calibration processing, establishes a mapping relationship between the image coordinate system and the track coordinate system, and obtains a calibrated contact network image, including: The inspection vehicle travels at a constant speed along the railway line. A circular array of multiple high-speed cameras is installed on the vehicle's 4C detection device to record the vehicle's six-degree-of-freedom motion data. The global navigation satellite system also records the vehicle's centimeter-level spatial positioning data. The collected six-degree-of-freedom motion data and spatial positioning data are aligned based on the timestamp, and the internal and external parameter matrices between the multi-view high-speed cameras are constructed. Based on the real-time posture of the vehicle body, the transformation relationship from the high-speed camera coordinate system to the track coordinate system is calculated to eliminate the influence of vehicle body vibration and jitter on the image position accuracy, and obtain a calibrated contact network image.

3. A high-speed railway contact network completion acceptance method based on 4C detection as claimed in claim 2, characterized in that: The method inputs the calibrated contact network image into a multi-scale target detection model based on deep learning, outputs the detection results, corrects the posture of the detection results, combines the mapping relationship, generates the three-dimensional positioning coordinates of the components in the track coordinate system, and denoises the detection results using a confidence-weighted filtering algorithm to eliminate low-confidence false detection samples, including: A multi-scale target detection model is used to identify and calibrate contact network images. The multi-scale target detection model includes a lightweight backbone network, a multi-scale attention feature pyramid, and a decoupled Transformer decoder. The lightweight backbone network is used to extract multi-level features of the image. The multi-scale attention feature pyramid is used to fuse features of different scales and enhance the expression ability of small targets. The decoupled Transformer decoder is used to parallelly decode the category, bounding box, and confidence of the component. The inertial measurement unit data is used to calculate the pose compensation matrix at the time of calibrating the contact network image acquisition, and the posture correction is performed on the detected two-dimensional coordinates of the contact network components. According to the type of contact network components and the distribution of feature points, combined with the camera imaging model, the corrected two-dimensional coordinates are projected into the track coordinate system, and the three-dimensional spatial coordinates of the contact network components are calculated. The confidence-weighted filtering algorithm is used for multiple detection results of the same contact network component. According to the detection confidence, geometric consistency and temporal stability, the optimal fusion result is obtained and the confidence false detection samples are eliminated.

4. A high-speed railway contact network completion acceptance method based on 4C detection as claimed in claim 3, characterized in that: The multi-scale target detection model includes: The lightweight backbone network adopts an improved lightweight version of the convolutional neural network architecture and introduces deformable convolution to replace the standard convolution layer, thereby enhancing the network's perception of deformed and small targets while keeping the computational complexity unchanged. The multi-scale attention feature pyramid includes a channel attention module and a spatial adaptive aggregation module. The channel attention module is used to learn the importance weight of each channel, and the spatial adaptive aggregation module uses a learnable kernel function to dynamically fuse multi-scale features. The decoupled Transformer decoder includes three parallel branches, one for category prediction, one for position regression, and one for shape estimation.

5. The high-speed railway contact network completion acceptance method based on 4C detection according to claim 3 is characterized in that: The posture compensation matrix and the confidence weighted filtering algorithm include: The acceleration and angular velocity data collected by the inertial measurement unit are modeled using a time-domain graph attention network to construct a time-series state graph. The body vibration mode of the inspection vehicle is extracted through graph convolution and time-series self-attention mechanism to obtain a continuous posture estimation sequence. Based on the time relationship between the sampling frequency of the inertial measurement unit and the camera frame rate, cubic spline interpolation is used to calculate the posture compensation matrix corresponding to each frame image to eliminate the influence of the inspection vehicle body jitter on the detection results. For the same contact network component detected, a time-series fusion tree of multi-frame detection results is constructed. Each node of the time-series fusion tree contains the category, position, size and confidence information of the component. The optimal state estimate is calculated by an adaptive Bayesian recursive estimation method. Based on the geometric consistency verification mechanism, the spatial relationship constraints and structural prior knowledge between components are utilized to eliminate false detection results that violate physical rules.

6. The high-speed railway contact network completion acceptance method based on 4C detection according to claim 1 is characterized in that: The test results are aligned with historical test data in time and space, a time series feature vector indexed by the component number is constructed, the time series feature vector is input into a prediction model combining a time series convolutional network and a long short-term memory network, the defect evolution trend and risk probability of the component are output, and the risk probability is graded based on a dynamic Bayesian threshold algorithm, including: The component inspection data collected from multiple inspections is spatiotemporally aligned. A spatial index is established based on the global navigation satellite system positioning data and the component number. The multi-period data is time-aligned using a dynamic time warping algorithm. A time series feature database is constructed based on the component's unique identifier. Each record contains the component type, spatial location, inspection time, surface characteristics, geometric parameters, environmental status, and historical maintenance records. The sliding time window method is used to extract the state change sequence of each component at different time points to form a multidimensional time series feature vector. Inputting the multidimensional time series feature vector into a prediction model to predict defect trends, wherein the prediction model includes a multi-layer time series convolutional network, a long short-term memory network, and an attention mechanism; The multi-layer temporal convolutional network is used to extract temporal pattern features, the long short-term memory network is used to capture long-term dependencies, the attention mechanism is used to highlight key state changes, and the prediction model is used to output the defect evolution trend and predicted risk probability of each component, establish a risk quantification index, and use a dynamic Bayesian threshold algorithm to adaptively classify the predicted risk probability in combination with changes in environmental conditions, load levels, and maintenance cycles to generate four levels of risk labels, including: R1, R2, R3, and R4, where R1 is a normal level risk label, R2 is a caution level risk label, R3 is a warning level risk label, and R4 is a danger level risk label.

7. A high-speed railway contact network completion acceptance method based on 4C detection according to claim 6, characterized in that: Based on risk levels and spatial positioning data, combined with a knowledge graph-driven rule engine, an intelligent work order containing maintenance paths, priority rankings, and spare parts lists is generated. The intelligent work order is mapped to on-site equipment through an augmented reality navigation device, and maintenance results are collected and fed back to the digital twin database, including: Constructing a catenary knowledge graph, wherein the catenary knowledge graph includes spatial relationships between components, functional dependencies, fault propagation modes, and maintenance specifications; Based on the catenary knowledge graph and risk level data, a combinatorial optimization algorithm is used to calculate the optimal maintenance path and timing arrangement; For R3 and R4 risk levels of overhead line components, intelligent work orders are generated that include detailed maintenance steps, required tools, spare parts lists, and quality acceptance criteria. Through augmented reality navigation devices, work order information is spatially matched with on-site equipment, and maintenance guidance information is projected into the maintenance personnel's field of view in real time, providing visual operation guidance; Collect maintenance process data and maintenance results based on the mobile terminal, the maintenance process data including operation time and operation records, and the maintenance results including replacement parts information and quality test results; Upload maintenance feedback data to the digital twin database in real time to update the contact network status and adjust model parameters; A sample-adaptive incremental learning mechanism is constructed to dynamically update the model parameters of the multi-scale target detection model and prediction model for new defect samples and false detection samples.

8. A high-speed railway contact network completion acceptance system based on 4C detection, used to execute a high-speed railway contact network completion acceptance method based on 4C detection according to any one of claims 1 to 7, characterized in that: The system comprises: The data acquisition module is used to collect catenary image data based on a high-speed camera array, obtain the posture data and spatial positioning data of the inspection vehicle, perform synchronous calibration processing, establish a mapping relationship between the image coordinate system and the track coordinate system, and obtain a calibrated catenary image; The multi-scale target detection module is used to input the calibrated catenary image into the multi-scale target detection model based on deep learning, output the detection results, perform pose correction on the detection results, and generate the three-dimensional positioning coordinates of the components in the track coordinate system based on the mapping relationship. The detection results are denoised using a confidence-weighted filtering algorithm to eliminate low-confidence false detection samples; The risk grading module is used to align the test results with historical test data in time and space, construct a time series feature vector indexed by the part number, input the time series feature vector into a prediction model that combines a time series convolutional network with a long short-term memory network, output the defect evolution trend and risk probability of the part, and grade the risk probability based on the dynamic Bayesian threshold algorithm; The acceptance feedback module is used to generate intelligent work orders that include maintenance paths, priority rankings, and spare parts lists based on risk levels and spatial positioning data, combined with a knowledge graph-driven rule engine. The intelligent work orders are mapped to on-site equipment through augmented reality navigation devices, and maintenance results are collected and fed back to the digital twin database.

9. An electronic device, characterized in that: include: at least one processor, at least one memory, a communication interface, and a bus; The processor, memory, and communication interface communicate with each other through the bus, the memory stores program instructions that can be executed by the processor, and the processor calls the program instructions to implement the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores computer instructions, and the computer instructions enable a computer to implement the method according to any one of claims 1 to 7.

Images