Vehicle damage correlation analysis method and system based on multi-source data

Abstract

The present application relates to the technical field of vehicle damage analysis, and specifically includes a vehicle damage correlation analysis method and system based on multi-source data. The method includes: obtaining vehicle accident multi-source data, establishing a multi-dimensional feature matrix, identifying vehicle damage dominant factors and effective correlation chains, generating a vehicle damage impact degree ranking, constructing a vehicle damage correlation analysis network, and generating a vehicle damage diagnosis report in combination with the vehicle damage impact degree ranking. The technical problems of insufficient recognition ability for latent chain damage, easy neglect of chain damage caused by secondary factors, and the inability of vehicle damage diagnosis reports to provide comprehensive and accurate decision support for accident handling are solved. The use of forward propagation to determine correlation strength values and identify vehicle damage dominant factors, effective correlation chains, and generate impact degree rankings is achieved. The correlation between impact factors and vehicle damage types is accurately quantified. A vehicle damage correlation analysis network is constructed, and a diagnosis report is generated in combination with the impact degree ranking. The technical effects of effectively improving the comprehensiveness, accuracy, and reliability of vehicle damage diagnosis are achieved.

Description

Technical Field

[0001] This invention relates to the field of vehicle damage analysis technology, specifically to a method and system for vehicle damage correlation analysis based on multi-source data. Background Technology

[0002] Vehicle accidents are complex in their causes, and there are multi-dimensional relationships between vehicle damage types and influencing factors. These relationships involve not only the mechanical effects of the collision but also the vehicle's own attributes and the accident scene environment. Conventional vehicle damage analysis often relies on isolated analysis of individual influencing factors such as collision speed, lacking systematic correlation analysis. It is impossible to quantify the strength of the correlation between influencing factors and vehicle damage types, and often only visible vehicle damage can be judged. For example, in a minor side collision, only visible damage such as door dents is noticed, while the slight deformation of the vehicle body structure caused by the collision affects the normal operation of the vehicle's suspension system, which in turn leads to hidden chain damage such as abnormal tire wear.

[0003] In summary, existing technologies suffer from insufficient ability to identify hidden cascading damage, easily overlook cascading damage caused by secondary factors, and vehicle damage diagnosis reports fail to provide comprehensive and accurate decision support for accident handling. Summary of the Invention

[0004] This application provides a vehicle damage correlation analysis method and system based on multi-source data, aiming to solve the technical problems in the existing technology of insufficient ability to identify hidden chain damage, easy to ignore chain damage caused by secondary factors, and vehicle damage diagnosis reports that cannot provide comprehensive and accurate decision support for accident handling.

[0005] In view of the above problems, the technical solution to achieve the present application is as follows:

[0006] In a first aspect, this application provides a method for vehicle damage correlation analysis based on multi-source data. The method includes: acquiring multi-source vehicle accident data, including accident scene image data, vehicle sensor data, environmental data, and vehicle attribute data, and establishing a multi-dimensional feature matrix; performing correlation analysis based on the multi-dimensional feature matrix, using forward propagation to determine the correlation strength values ​​between multiple influencing factors and various vehicle damage types, identifying the dominant vehicle damage factors and effective correlation chains, and generating a vehicle damage impact ranking; hierarchically dividing the dominant vehicle damage factors as core nodes, using the effective correlation chains as directional connection paths for network nodes, constructing a vehicle damage correlation analysis network with the multiple influencing factors and associated vehicle damage types as network nodes, and using the correlation strength values ​​as edge weights; and generating a vehicle damage diagnosis report based on the vehicle damage correlation analysis network and the vehicle damage impact ranking.

[0007] Preferably, the damage area features are extracted by semantic segmentation of the accident scene image data in the multi-source data of the vehicle accident; temporal features are extracted from the vehicle sensor data in the multi-source data of the vehicle accident to determine the damage time features; and a multi-dimensional feature matrix is ​​obtained based on the damage area features and the damage time features.

[0008] Preferably, by using the cross features of environmental data and vehicle attribute data in the multi-source data of vehicle accidents, a coupling factor between road friction coefficient and vehicle weight, an adaptation index between visibility and speed limit, and a dynamic response coefficient between vehicle wheelbase and steering angle are generated; the damage area features, damage time features, and cross features are standardized and the multi-dimensional feature matrix is ​​obtained according to the matrix structure of accident sample × feature dimension.

[0009] Preferably, the forward propagation process employs a multilayer perceptron, with the input layer being the multidimensional feature matrix and the hidden layer undergoing nonlinear transformation using the ReLU activation function; when analyzing the correlation strength between multiple influencing factors and various vehicle damage types, a weighted fusion of Pearson correlation coefficient and mutual information entropy is introduced.

[0010] Preferably, hierarchical clustering is used to group the vehicles based on the dominant factors of vehicle damage, and the potential association chains are screened by the association strength filtering threshold; the potential association chains are then verified by time series to obtain the effective association chains.

[0011] Preferably, based on the dominant factors of vehicle damage, the inter-class distance is determined, and multiple factor components and multiple factor groups are obtained by dividing the data according to the inter-class distance; path analysis is performed based on the multiple factor components and multiple factor groups, and the multilayer perceptron is traversed by depth-first search. When the association strength values ​​of M consecutive network nodes in the vehicle damage propagation process all meet the association strength filtering threshold, they are marked as potential association chains, where M is greater than or equal to 2.

[0012] Preferably, on the potential associated chain, the contribution of the on-chain network node is obtained; based on the contribution of the on-chain network node, the cumulative value of the product of the time difference and the contribution of adjacent on-chain network nodes in the potential associated chain is obtained; when the cumulative value of the product of the time difference and the contribution meets the timing verification threshold, and the occurrence order of the on-chain nodes conforms to the accident mechanics transmission logic, the timing verification is determined to be passed.

[0013] Preferably, if the cumulative value of the product of the time difference and the contribution does not meet the time series verification threshold, a two-way tracing mechanism is used to deduce the extended potential association chain; the extended potential association chain is then time-tested by superimposing the cumulative value corresponding to the extended potential association chain with the time series compensation value and the time series verification threshold. If it still does not meet the time series verification threshold, the association chain is split and reorganized: the extended potential association chain is split into multiple sub-chains according to the peak of the time difference, and multiple sub-chains that pass the time series verification are retained; for multiple sub-chains that fail the time series verification, a secondary verification is performed in combination with the spatial damage correlation of the accident scene image data.

[0014] Preferably, the preceding influencing factors of the first node of the chain are traced forward to determine the correlation strength compensation value with the first node of the chain; the derived damage of the last node of the chain is deduced backward to generate the potential damage propagation coefficient; and the correlation strength compensation value and the potential damage propagation coefficient are weighted and fused to obtain the time-series compensation value.

[0015] In a second aspect, this application provides a vehicle damage correlation analysis system based on multi-source data. The system includes: a multi-dimensional feature matrix establishment module: acquiring multi-source vehicle accident data including accident scene image data, vehicle sensor data, environmental data, and vehicle attribute data, and establishing a multi-dimensional feature matrix; a correlation analysis module: performing correlation analysis based on the multi-dimensional feature matrix, using forward propagation to determine the correlation strength values ​​between multiple influencing factors and various vehicle damage types, identifying the dominant vehicle damage factors and effective correlation chains, and generating a vehicle damage impact ranking; a hierarchical division module: using the dominant vehicle damage factors as core nodes for hierarchical division, using the effective correlation chains as directional connection paths for network nodes, constructing a vehicle damage correlation analysis network with the multiple influencing factors and associated vehicle damage types as network nodes, and using the correlation strength values ​​as edge weights; and a vehicle damage diagnosis report generation module: generating a vehicle damage diagnosis report based on the vehicle damage correlation analysis network and the vehicle damage impact ranking.

[0016] In summary, one or more technical solutions provided in this application achieve the technical effect of using forward propagation to determine the correlation strength value, identify the dominant factors of vehicle damage, effective correlation chains, and generate influence ranking, accurately quantify the correlation between influencing factors and vehicle damage types, construct a vehicle damage correlation analysis network, and generate a diagnostic report by combining influence ranking, effectively improving the comprehensiveness, accuracy, and reliability of vehicle damage diagnosis. Attached Figure Description

[0017] Figure 1 This application provides a flowchart illustrating the vehicle damage correlation analysis method based on multi-source data.

[0018] Figure 2 This application provides a schematic diagram of the structure of a vehicle damage correlation analysis system based on multi-source data.

[0019] Figure labeling: Multi-dimensional feature matrix establishment module M100, correlation analysis module M200, hierarchical division module M300, vehicle damage diagnosis report generation module M400. Detailed Implementation

[0020] Example 1: The present application will be described in detail below with reference to the accompanying drawings, as follows... Figure 1 As shown, this application provides a vehicle damage correlation analysis method based on multi-source data, wherein the method includes:

[0021] S1: Acquire multi-source vehicle accident data, including accident scene image data, vehicle sensor data, environmental data, and vehicle attribute data, and establish a multi-dimensional feature matrix.

[0022] Specifically, accident scene image data refers to image information captured by devices such as cameras after an accident, including visual content such as damaged vehicle parts, collision marks, and the scene environment; vehicle sensor data is data recorded by various sensors on the vehicle, such as acceleration sensors, angular velocity sensors, and tire pressure sensors, on vehicle operation and the time of the accident, such as acceleration at the moment of collision and changes in vehicle speed; environmental data covers weather conditions, road conditions, and lighting conditions at the time of the accident; vehicle attribute data includes inherent attributes such as vehicle brand, model, year, weight, and wheelbase; establishing a multi-dimensional feature matrix involves processing the above multi-source data and integrating it according to different data feature dimensions to form a matrix structure that can comprehensively reflect all aspects of vehicle accident information.

[0023] Execution steps: Acquire accident scene image data through image acquisition equipment, acquire vehicle-mounted sensor data through vehicle-mounted sensors, acquire environmental data through meteorological departments, traffic management departments, and other channels, and acquire vehicle attribute data from vehicle registration information, etc.; Preprocess this data, such as preliminary processing of accident scene image data such as damage area identification, and outlier removal of vehicle-mounted sensor data, etc. Integrate the processed data according to feature dimensions to construct a multi-dimensional feature matrix. The above steps are the foundation of the entire vehicle damage correlation analysis, providing comprehensive and systematic data support for subsequent correlation analysis, avoiding the one-sidedness of analysis caused by single or isolated data, and making the subsequent analysis of the correlation between influencing factors and vehicle damage types more accurate and reliable.

[0024] S2: Based on the multi-dimensional feature matrix, perform correlation analysis, use forward propagation to determine the correlation strength values ​​between multiple influencing factors and various vehicle damage types, identify the dominant factors of vehicle damage and effective correlation chains, and generate a ranking of vehicle damage impact.

[0025] Specifically, correlation analysis refers to exploring the relationship between different influencing factors and vehicle damage types in a multi-dimensional feature matrix to find the inherent connections. Multiple influencing factors include collision speed and road surface friction coefficient, while vehicle damage types include sheet metal deformation and glass breakage. Forward propagation is the data propagation method in a multilayer perceptron, where input data is processed from the input layer through the hidden layer and then transmitted to the output layer. The correlation strength value quantifies the degree of correlation between multiple influencing factors and various vehicle damage types; a higher value indicates a closer correlation. The dominant factor in vehicle damage refers to the factor that plays a major role among the many factors causing vehicle damage. An effective correlation chain refers to a real, logically connected chain between influencing factors and vehicle damage types, such as heavy rain → increased braking distance → rear-end collision → rear structure damage. The vehicle damage impact ranking is the order in which influencing factors are arranged according to the degree of their influence on vehicle damage types.

[0026] Execution steps: Based on the constructed multi-dimensional feature matrix, it is input into the input layer of a multilayer perceptron. The hidden layer performs a nonlinear transformation on the data using the ReLU activation function to achieve forward propagation. When analyzing the correlation strength value, a weighted fusion of Pearson correlation coefficient and mutual information entropy is introduced to obtain the correlation strength values ​​between multiple influencing factors and various vehicle damage types. Based on the dominant factors of vehicle damage, hierarchical clustering is used for grouping, combined with correlation strength filtering thresholds to mine potential correlation chains. The potential correlation chains are then time-series verified to obtain effective correlation chains. Based on the correlation strength value and other information, a vehicle damage impact ranking is generated. In the above steps, by quantifying the correlation strength value, the dominant factors of vehicle damage and effective correlation chains are accurately identified, providing key correlation information for the subsequent construction of the vehicle damage correlation analysis network. At the same time, the vehicle damage impact ranking provides an important reference for the final generation of the vehicle damage diagnosis report, improving the accuracy of vehicle damage analysis.

[0027] S3: The vehicle damage dominant factors are used as core nodes for hierarchical division, the effective correlation chains are used as directional connection paths for network nodes, and the multiple influencing factors and the associated vehicle damage types are used as network nodes. The correlation strength value is used as the edge weight to construct a vehicle damage correlation analysis network.

[0028] Specifically, core nodes are network nodes centered on the dominant factors of vehicle damage, serving as the key hubs of the vehicle damage correlation analysis network. Hierarchical division refers to dividing the dominant factors of vehicle damage into different levels based on their importance and scope of influence, making the network structure more hierarchical. Network nodes include multiple influencing factors and various associated vehicle damage types, forming the basic units of the vehicle damage correlation analysis network. Directed connection paths consist of effective association chains, defining the direction of information transmission between network nodes. Edge weights, or association strength values, are used to represent the tightness of the association between network nodes. Constructing the vehicle damage correlation analysis network involves combining the aforementioned core nodes, network nodes, directed connection paths, and edge weights according to certain logic to form a network model that intuitively reflects the correlation relationships of vehicle damage.

[0029] Execution steps: Based on the influence of the dominant factors of vehicle damage and the degree of correlation with other factors, a hierarchical division is performed to determine the core nodes at different levels; multiple influencing factors and various types of vehicle damage are treated as network nodes, and effective correlation chains are used as directional connection paths between network nodes. The correlation strength value is assigned to the corresponding path as the edge weight. Through this combination, a vehicle damage correlation analysis network is constructed. In the above steps, the scattered correlation information obtained from the analysis is integrated into a systematic network model, which intuitively shows the correlation paths and strengths between influencing factors and vehicle damage types. This provides a structured analytical framework for the subsequent generation of vehicle damage diagnosis reports, enhancing the correlation and systematic nature of vehicle damage analysis.

[0030] S4: Based on the vehicle damage correlation analysis network and the vehicle damage impact ranking, generate a vehicle damage diagnosis report.

[0031] Specifically, the vehicle damage correlation analysis network is a pre-constructed network model containing core nodes, network nodes, directional connection paths, and edge weights, used to comprehensively present the correlations related to vehicle damage; the vehicle damage impact ranking is the result of ranking the degree of influence of influencing factors on vehicle damage types; the vehicle damage diagnosis report is a report generated after integrating the network analysis results and impact ranking, containing information such as vehicle damage status and cause analysis; generating a vehicle damage diagnosis report is the process of extracting key information and forming a standardized report based on the network model and impact ranking.

[0032] Execution Steps: Based on the constructed vehicle damage correlation analysis network and combined with the vehicle damage impact ranking, key information such as the correlation path and correlation strength between each influencing factor and vehicle damage type, as well as the impact range of core nodes at different levels, are extracted from the network. Simultaneously, the main influencing factors and the vehicle damage types of focus are determined with reference to the impact ranking. This information is then organized, analyzed, and summarized to form a vehicle damage diagnostic report containing information such as vehicle damage type, causative factors, correlation relationships, and degree of impact. In the above steps, by integrating the correlation information from the network model and the priority information from the impact ranking, a comprehensive and accurate diagnostic report is generated, providing reliable decision support for accident handling and insurance claims, thus achieving a closed loop in vehicle damage analysis from data integration to effective output.

[0033] Furthermore, the method used in this application to establish a multi-dimensional feature matrix includes:

[0034] The damage area features are extracted by semantic segmentation of the accident scene image data in the multi-source data of the vehicle accident; the temporal features are extracted from the vehicle sensor data in the multi-source data of the vehicle accident to determine the damage time features; and a multi-dimensional feature matrix is ​​obtained based on the damage area features and the damage time features.

[0035] Specifically, semantic segmentation refers to pixel-level classification of accident scene image data, accurately dividing the areas of vehicle damage from non-damaged areas and areas of different damage types in the image, thereby identifying the specific damage location and extent; damage area features are extracted from accident scene images through semantic segmentation, which can describe the shape, size, location, and degree of damage of the damaged area; temporal feature extraction analyzes the changes in vehicle sensor data over time, extracting time-related features, including but not limited to peak collision impact force, rate of change of acceleration, and brake pedal travel curve; damage time features are obtained through temporal feature extraction, reflecting the temporal characteristics of the occurrence and development of vehicle damage, such as the moment the damage occurred and the duration of the damage.

[0036] Execution steps: Semantic segmentation is performed on the accident scene image data. Specifically, a deep learning semantic segmentation model, the U-Net model, is used to segment the vehicle damage areas, including dented door areas and broken window areas, from other regions in the image, thereby extracting damage area features such as shape, size, and location. Temporal features are extracted from onboard sensor data, including acceleration sensor data and vehicle speed sensor data at the moment of collision. By calculating time-series statistics such as mean, variance, and peak time, damage temporal characteristics are determined, such as the precise time of the collision, the time of the acceleration peak, and the temporal correlation between the collision damage. The extracted damage area features and damage temporal features are integrated according to preset feature dimensions to form a multi-dimensional feature matrix. In the above steps, the feature extraction and integration of these two types of key data provide core feature content for the construction of the multi-dimensional feature matrix, enabling the feature matrix to simultaneously cover the spatial and temporal information of vehicle damage. This lays a more accurate data foundation for subsequent analysis of the correlation between influencing factors and vehicle damage types.

[0037] Furthermore, based on the damage area features and damage time features, a multi-dimensional feature matrix is ​​obtained. The method of this application includes:

[0038] By leveraging the cross-features of environmental data and vehicle attribute data from the multi-source vehicle accident data, a coupling factor between road surface friction coefficient and vehicle weight, an adaptation index between visibility and speed limit, and a dynamic response coefficient between vehicle wheelbase and steering angle are generated. The damage area features, damage time features, and cross-features are standardized, and the multi-dimensional feature matrix is ​​obtained according to the matrix structure of accident sample × feature dimension.

[0039] Specifically, cross features refer to new features obtained by combining and analyzing the interrelated features in environmental data and vehicle attribute data, which are used to reflect the synergistic influence between the two types of data; the coupling factor between road friction coefficient and vehicle weight is a parameter obtained by combining the comprehensive road friction coefficient corresponding to environmental data and the vehicle weight corresponding to vehicle attribute data, reflecting the degree of influence of both on vehicle braking and collision consequences; the matching index between visibility and speed limit is an index obtained by combining the visibility corresponding to environmental data and the speed limit corresponding to traffic rules when the vehicle is driving, reflecting the rationality of the matching between visibility and actual speed; the dynamic response coefficient between vehicle wheelbase and steering angle is a coefficient obtained by associating the vehicle wheelbase corresponding to vehicle attribute data and the steering angle corresponding to on-board sensor data, characterizing the dynamic characteristics of the vehicle when turning; standardization processing refers to the process of scaling and normalizing the damage area features, damage time features, and cross features to the [0,1] interval to eliminate the difference in dimensions between different features; in the matrix structure of accident sample × feature dimension, the accident sample refers to a single vehicle accident case, the feature dimension refers to the features extracted from various types, and each element in the matrix represents the specific value of a certain accident sample on a certain feature dimension.

[0040] Execution steps: Extract features such as road surface friction coefficient and visibility from environmental data, and extract features such as vehicle weight and wheelbase from vehicle attribute data. Calculate cross-features using a preset algorithm: For example, based on the road surface friction coefficient μ and vehicle weight m, calculate the coupling factor formula (μ×m) / k, where k is a constant. The coupling factor reflects the change in impact force during a vehicle collision under the combined action of the two factors. Combine visibility v and the speed limit s of the road segment, obtain the adaptation index by (v / s)×100%. If the index is below 60%, it indicates low visibility but high speed, posing a risk. Based on the vehicle wheelbase L and steering angle θ, obtain the dynamic response coefficient by (L×sinθ) / t, where t is the steering response time. The larger the coefficient, the more significant the change in vehicle posture during steering.

[0041] The damage area features, including damage area and depth, the damage time features, including damage duration and occurrence time, and cross-features are standardized. For example, the min-max normalization formula is used to transform the data to a unified interval. Each accident case is treated as a row of accident samples, and all features, including damage area features, damage time features, and cross-features, are used as columns. The number of columns is the feature dimension, thus constructing a multi-dimensional feature matrix of accident samples × feature dimensions. In the above steps, the generation of cross-features enriches the feature dimensions and compensates for the limitations of single data features. The standardization process provides a consistent basis for the fusion analysis of multi-source features. The resulting multi-dimensional feature matrix can more comprehensively and accurately depict the entire accident, providing high-quality data support for subsequent correlation strength calculation and network construction.

[0042] Furthermore, based on the aforementioned multi-dimensional feature matrix, correlation analysis is performed, and forward propagation is used to determine the correlation strength values ​​between multiple influencing factors and various vehicle damage types. The method of this application includes:

[0043] The forward propagation process employs a multilayer perceptron, with the input layer being the multidimensional feature matrix and the hidden layer undergoing nonlinear transformation using the ReLU activation function. When analyzing the correlation strength between multiple influencing factors and various vehicle damage types, a weighted fusion of Pearson correlation coefficient and mutual information entropy is introduced.

[0044] Specifically, a multilayer perceptron is an artificial neural network model that includes an input layer, hidden layers, and an output layer. It can process complex data relationships through multilayer nonlinear transformations. The ReLU activation function is a commonly used nonlinear activation function, expressed as f(x) = max(0,x), which can effectively solve the gradient vanishing problem in neural networks and enhance the nonlinear fitting ability of the model. The Pearson correlation coefficient is a statistic used to measure the degree of linear correlation between two variables, with a value range of [-1,1]. The closer the absolute value is to 1, the stronger the linear correlation. Mutual information entropy is an indicator that measures the dependency between two random variables. It is not limited by the linear relationship between variables and can capture nonlinear associations. Weighted fusion refers to assigning different weights to the Pearson correlation coefficient and mutual information entropy and then performing a comprehensive calculation to take into account the measurement results of both linear and nonlinear associations.

[0045] Execution steps: The multi-dimensional feature matrix is ​​input into the input layer of the multilayer perceptron, with the number of neurons in the input layer matching the feature dimensions of the feature matrix. Data is then passed to the hidden layers, where neurons perform a non-linear transformation on the input data using the ReLU activation function. Specifically, a positive input value is directly output, and a negative value is output as 0. Through this multi-layered processing, complex non-linear relationships within the feature matrix are extracted. When calculating the correlation strength between influencing factors and vehicle damage type, the Pearson correlation coefficient and mutual information entropy are first calculated separately. Then, based on the importance of linear and non-linear correlations in the actual scenario, specifically, the collision speed and vehicle dent depth are compared. The linear correlation degree is used as the Pearson correlation coefficient, and the nonlinear correlation between the road friction coefficient and the tire wear type is used as the mutual information entropy. Weights are assigned to both, for example, the linear correlation weight is 0.6 and the nonlinear correlation weight is 0.4. The correlation strength value is obtained by weighted summation. In the above steps, the forward propagation of the multilayer perceptron is used to achieve in-depth mining of complex data features. Combined with the weighted fusion correlation strength calculation method, both linear and nonlinear correlations are considered, making the quantification of the correlation strength value more accurate. This provides a reliable numerical basis for subsequent identification of the dominant factors of vehicle damage and effective correlation chains, and further improves the scientific nature of vehicle damage analysis.

[0046] Furthermore, to identify the dominant factors and effective correlation chains of vehicle damage, the method of this application includes:

[0047] Hierarchical clustering is used to group the vehicle damage-dominant factors, and a correlation strength filtering threshold is used to screen potential correlation chains. The potential correlation chains are then subjected to time-series verification to obtain the effective correlation chains.

[0048] Specifically, hierarchical clustering is a method that aggregates or splits data objects layer by layer according to their similarity to form different hierarchical clusters. It is used to divide multiple factor groups based on the correlation characteristics of the dominant factors of vehicle damage. The correlation strength filtering threshold is a preset critical value used to screen correlation relationships. When the correlation strength value reaches or exceeds the threshold, the corresponding correlation relationship is considered to be significant. Potential correlation chains refer to the correlation chains between potential but unverified influencing factors and vehicle damage types obtained through screening. Time-series verification is to test the rationality of potential correlation chains in combination with the time dimension and judge whether the occurrence order of each node in the chain conforms to the time logic of accident development.

[0049] Execution steps: 1. Perform hierarchical clustering based on the dominant factors of vehicle damage: Calculate the inter-class distance between different dominant factors, such as using Euclidean distance or Manhattan distance. Divide the dominant factors into multiple factor components and factor groups according to the size of the inter-class distance, so that factors within a group are closely related and factors between groups are less related. 2. Filter potential association chains by combining association strength filtering threshold: Perform path analysis on the divided factor components and groups. Traverse the multilayer perceptron through depth-first search. When the association strength values ​​of M consecutive network nodes in the vehicle damage propagation process all reach or exceed the association strength filtering threshold, the path is marked as a potential association chain, where M≥2. For example, in a collision accident, if the association strength values ​​of the three nodes—collision speed, side impact force, and door deformation—all exceed the threshold of 0.7, then a potential association chain is formed.

[0050] Temporal verification of potential association chains is performed: the contribution of each network node in the chain is obtained, and the cumulative value of the product of the time difference and contribution of adjacent nodes is calculated. When the cumulative value meets the temporal verification threshold and the order of node occurrence conforms to the accident mechanics transmission logic, the temporal verification is deemed successful, and a valid association chain is obtained. Specifically, the accident mechanics transmission logic refers to the sequence and mechanism of force generation, transmission, action, and damage caused by a vehicle during an accident, following the laws of physics and mechanics. It covers the complete process from the external impact force acting on the vehicle, to the force being transmitted step by step in the vehicle body structure, and then to the deformation and damage of different components due to the force. Each link conforms to the force transmission path and energy dissipation law in mechanics. In the above steps, clustering simplifies the association analysis of complex factors, filtering thresholds reduce the interference of invalid associations, and temporal verification ensures the temporal logic rationality of the association chain. The final valid association chain provides a core directional connection path for constructing the vehicle damage association analysis network, significantly improving the logic and reliability of vehicle damage association analysis.

[0051] Furthermore, based on the aforementioned dominant factors of vehicle damage, hierarchical clustering is employed for grouping, combined with correlation strength filtering thresholds for screening, to uncover potential correlation chains. The method of this application includes:

[0052] Based on the dominant factors of vehicle damage, the inter-class distance is determined, and multiple factor components and multiple factor groups are obtained by dividing the data according to the inter-class distance. Path analysis is performed based on the multiple factor components and multiple factor groups. The multilayer perceptron is traversed by depth-first search. When the association strength values ​​of M consecutive network nodes in the vehicle damage propagation process all meet the association strength filtering threshold, they are marked as potential association chains, where M is greater than or equal to 2.

[0053] Specifically, inter-class distance refers to the similarity measure between the categories to which different dominant factors of vehicle damage belong, used to reflect the differences between dominant factors of vehicle damage in different groups. Commonly used calculation methods include Euclidean distance and Manhattan distance. Factor components and factor groups are different sets obtained by partitioning through inter-class distance. Factor components can be regarded as subdivision units under groups, while factor groups are higher-level classification sets. Together, they realize the structured grouping of dominant factors of vehicle damage. Depth-first search is an algorithm for traversing graphs or tree structures, used to traverse network nodes in a multilayer perceptron in depth-first order to explore potential association paths. Association strength filtering threshold is a critical value for judging whether the association strength value is significant. When the association strength value reaches or exceeds the threshold, the association between nodes is considered to have analytical value. Potential association chain refers to an association path that satisfies the requirement that the association strength values ​​of M consecutive nodes all meet the filtering threshold, and has not yet been verified by time series or other methods.

[0054] Execution steps: Based on the dominant factors of vehicle damage, including collision speed, road friction coefficient, and vehicle weight, Euclidean distance is selected to determine the inter-class distance. For example, the inter-class distance between collision speed and road friction coefficient, and the inter-class distance between vehicle weight and collision angle are obtained. Clustering is performed based on the magnitude of the inter-class distance to obtain multiple factor components and multiple factor groups. The multiple factor components include sub-components such as high-speed collision-related factors and low-friction road surface-related factors. The multiple factor groups include external environmental influence groups and vehicle self-attribute groups. Based on these factor components and groups, path analysis is performed. A depth-first search algorithm is used to traverse the multilayer perceptron. Starting from the starting node, the next level node directly associated with it is explored first, and then the next level node associated with that node is explored.

[0055] Simultaneously, it determines in real time whether the association strength values ​​of consecutive nodes meet the preset association strength filtering threshold. When the collision speed → vehicle body impact force → vehicle body structural damage satisfy the condition that the association strength values ​​of two or more consecutive nodes are all greater than the association strength filtering threshold, the path is marked as a potential association chain. In the above steps, clustering and grouping achieve orderly management of complex dominant factors, reducing the blindness of path search. Depth-first search ensures that the potential associations between nodes can be explored in depth, and the application of the association strength filtering threshold initially screens out significant association paths, providing targeted analysis objects for subsequent time-series verification of potential association chains.

[0056] Furthermore, the method of this application for temporal verification of the potential related chains includes:

[0057] On the potential associated chain, the contribution of the on-chain network nodes is obtained; based on the contribution of the on-chain network nodes, the cumulative value of the product of the time difference and the contribution of adjacent on-chain network nodes in the potential associated chain is obtained; when the cumulative value of the product of the time difference and the contribution meets the timing verification threshold, and the occurrence order of the on-chain nodes conforms to the accident mechanics transmission logic, the timing verification is determined to be passed.

[0058] Specifically, the contribution of on-chain network nodes refers to the degree to which each node in a potential associated chain influences subsequent nodes. It is an indicator that measures the magnitude of a node's role in the associated chain. Each node in a potential associated chain includes influencing factors and vehicle damage types. The time difference is the time interval between two adjacent network nodes in a potential associated chain. The cumulative value of the product of time difference and contribution is obtained by multiplying the time difference of adjacent nodes by the contribution of the corresponding nodes and then summing all the product results. It is used to quantify the overall effectiveness of the associated chain in the time dimension. The time sequence verification threshold is the critical value for judging whether the cumulative value conforms to the time sequence logic. The accident mechanics transmission logic refers to the physical laws followed by the transmission of forces and the development of damage during a vehicle accident, such as the collision force acting on the outside of the vehicle body first and then being transmitted to the internal structure to cause damage.

[0059] Execution steps: On the potential association chain, by analyzing the degree of influence of each node on the association chain transmission, the contribution of each network node is determined. For example, in the potential association chain of collision speed → vehicle body impact force → door deformation, the contribution of vehicle body impact force can be set to 0.8, indicating that it has a significant impact on door deformation. Obtain the time difference between adjacent nodes. For example, if the time difference between the collision speed and the vehicle body impact force is 0.2 seconds, and the time difference between the vehicle body impact force and door deformation is 0.3 seconds, and the contribution of door deformation is set to 0.7, calculate the product, 0.2 × 0.8 + 0.3 × 0.7, and accumulate it to obtain the cumulative value. Compare this cumulative value with the preset time sequence verification threshold. If the cumulative value ≥ the time sequence verification threshold, and the occurrence order of the nodes on the chain conforms to the transmission logic of force generation → force transmission → structural damage in accident mechanics, then the time sequence verification of the potential association chain corresponding to collision speed → vehicle body impact force → door deformation is determined to be passed. In the above steps, by quantifying the correlation between the time dimension and node contribution, and combining it with physical laws for verification, the temporal rationality and logical correctness of potential correlation chains are ensured, providing a key basis for screening out truly effective correlation chains and further improving the reliability of vehicle damage correlation analysis.

[0060] Furthermore, the method of this application includes:

[0061] If the cumulative value of the product of the time difference and the contribution does not meet the time series verification threshold, a two-way tracing mechanism is used to deduce the extended potential association chain. The extended potential association chain is then time-tested by superimposing the cumulative value corresponding to the extended potential association chain with the time series compensation value and the time series verification threshold. If it still does not meet the time series verification threshold, the association chain is split and reorganized: the extended potential association chain is split into multiple sub-chains according to the peak of the time difference, and multiple sub-chains that pass the time series verification are retained; for multiple sub-chains that fail the time series verification, a secondary verification is performed in combination with the spatial damage correlation of the accident scene image data.

[0062] Specifically, the two-way tracing mechanism refers to simultaneously tracing the preceding influencing factors of the first node of the chain and extrapolating the derived damage of the last node when the temporal verification of the potential associated chain fails, in order to expand the scope of the associated chain; the temporal compensation value is the value obtained by weighted fusion of the association strength compensation value obtained by two-way tracing and the potential damage expansion coefficient, used to make up for the insufficiency of the original cumulative value; the peak time difference is the maximum value of the time difference between adjacent nodes in the expanded potential associated chain, which is used as the dividing point for splitting the associated chain; the sub-chain is the shorter associated path formed after the associated chain is split; spatial damage correlation refers to the relationship between different damaged areas in the accident scene image in terms of spatial location, such as whether there is a mechanical transmission correlation between the damage of adjacent components.

[0063] Execution steps: When the cumulative value of the product of time difference and contribution does not meet the time series verification threshold, a two-way tracing mechanism is activated: trace the preceding influencing factors of the first node of the chain forward. For example, if the first node of the chain is the impact force on the vehicle body, trace its preceding collision angle to obtain the correlation strength compensation value between the preceding factor and the first node of the chain; extrapolate the derived damage of the last node of the chain backward. For example, if the last node of the chain is the deformation of the car door, extrapolate the derived loosening of the car window bracket to generate a potential damage expansion coefficient. The two are weighted and fused to obtain the time series compensation value, forming an expanded potential correlation chain.

[0064] The accumulated value of the expanded correlation chain is superimposed with the time-series compensation value and compared again with the time-series verification threshold. If it still fails to meet the threshold, the expanded chain is split into multiple sub-chains based on the peak time difference. For example, if the peak time difference is 0.5 seconds, the chain is split with this as the boundary, retaining the sub-chains that pass the time-series verification. For the sub-chains that fail, a secondary verification is performed based on the spatial relationship of the damaged areas in the accident scene images. Specifically, the spatial relationship of the damaged areas includes the spatial correlation between the door dent and the deformation of the adjacent fender. In the above steps, through the progressive processing of expansion, splitting, and secondary verification, potential effective correlations are explored to the maximum extent, avoiding the omission of important correlation chains due to the initial failure of time-series verification. This enhances the completeness and fault tolerance of vehicle damage correlation analysis and provides a more comprehensive guarantee for accurately identifying effective correlation chains.

[0065] Furthermore, by using a two-way tracing mechanism to extrapolate and obtain the extended potential association chain, the method of this application includes:

[0066] By tracing back the preceding influencing factors of the first node of the chain, the correlation strength compensation value with the first node of the chain is determined; by extrapolating the derived damage of the last node of the chain, a potential damage propagation coefficient is generated; and the correlation strength compensation value and the potential damage propagation coefficient are weighted and fused to obtain the time-series compensation value.

[0067] Specifically, the head node refers to the starting network node in a potential associated chain, usually an influencing factor; the precursor influencing factor refers to other factors that affect the head node before it; the association strength complement is a supplementary value that measures the degree of association between the precursor influencing factor and the head node, used to improve the quantification of association strength; the tail node is the last network node in a potential associated chain, often a certain type of vehicle damage; derived damage is other damage further caused by the damage at the tail node; the potential damage expansion coefficient is a coefficient used to quantify the degree of association between derived damage and the tail node; weighted fusion is a comprehensive calculation after assigning appropriate weights to the association strength complement and the potential damage expansion coefficient to obtain the time-series compensation value.

[0068] Execution steps: For the head node of a potential related chain, trace back its preceding influencing factors. For example, if the head node is the side impact force of the vehicle body, its preceding influencing factors may include the mass of the colliding vehicle and the collision angle. By calculating the correlation strength between these preceding influencing factors and the side impact force of the vehicle body, determine the correlation strength complement. Assume that the correlation strength complement between the collision angle and the side impact force of the vehicle body is 0.3.

[0069] For damage derived from the tail node of the chain, if the tail node is a deformed door, its derived damage may include damage to the door hinge or window malfunction. A potential damage expansion coefficient is generated based on the correlation between the derived damage and the tail node; for example, the potential damage expansion coefficient for door hinge damage is 0.4. Appropriate weights are assigned to the correlation strength compensation value and the potential damage expansion coefficient, such as 0.5 and 0.5 respectively. A weighted sum is then used to obtain the time-series compensation value: 0.3 × 0.5 + 0.4 × 0.5 = 0.35. In the above steps, by supplementing the correlation information of precursor factors and derived damage, the generated time-series compensation value provides a basis for re-verification of the expanded potential correlation chain, effectively improving the ability to capture complex correlation chains and ensuring the comprehensiveness of vehicle damage correlation analysis.

[0070] In summary, the beneficial effects of the embodiments of this application are:

[0071] This application provides a vehicle damage correlation analysis method and system based on multi-source data. It utilizes multi-source vehicle accident data acquisition, including accident scene image data, vehicle sensor data, environmental data, and vehicle attribute data, and establishes a multi-dimensional feature matrix. Based on this matrix, it performs correlation analysis, using forward propagation to determine the correlation strength between multiple influencing factors and various vehicle damage types, identifying dominant damage factors and effective correlation chains, and generating a vehicle damage impact ranking. The dominant damage factors are used as core nodes for hierarchical division, and effective correlation chains are used as directional connection paths for network nodes. A vehicle damage correlation analysis network is constructed using multiple influencing factors and associated vehicle damage types as network nodes, with correlation strength values ​​as edge weights. Based on this network and the vehicle damage impact ranking, a vehicle damage diagnostic report is generated. This application achieves the technical effect of using forward propagation to determine correlation strength values, identify dominant damage factors, effective correlation chains, and generate an impact ranking, accurately quantifying the correlation between influencing factors and vehicle damage types, constructing a vehicle damage correlation analysis network, and generating a diagnostic report based on the impact ranking. This effectively improves the comprehensiveness, accuracy, and reliability of vehicle damage diagnosis.

[0072] Example 2, based on the same inventive concept as the vehicle damage correlation analysis method based on multi-source data in the previous examples, such as... Figure 2 As shown in the embodiment of this application, a vehicle damage correlation analysis system based on multi-source data is provided, wherein the system includes:

[0073] Multi-dimensional feature matrix building module M100: Acquires multi-source vehicle accident data, including accident scene image data, vehicle sensor data, environmental data, and vehicle attribute data, and builds a multi-dimensional feature matrix.

[0074] The correlation analysis module M200 performs correlation analysis based on the multi-dimensional feature matrix, uses forward propagation to determine the correlation strength values ​​between multiple influencing factors and various vehicle damage types, identifies the dominant factors of vehicle damage and effective correlation chains, and generates a ranking of vehicle damage impact.

[0075] Hierarchical partitioning module M300: The vehicle damage dominant factors are used as core nodes for hierarchical partitioning, the effective correlation chains are used as directional connection paths for network nodes, the multiple influencing factors and the associated vehicle damage types are used as network nodes, and the correlation strength value is used as the edge weight to construct a vehicle damage correlation analysis network.

[0076] Vehicle damage diagnosis report generation module M400: Based on the vehicle damage correlation analysis network and combined with the vehicle damage impact ranking, it generates a vehicle damage diagnosis report.

[0077] Furthermore, the multi-dimensional feature matrix establishment module M100 is used to perform the following method:

[0078] The damage area features are extracted by semantic segmentation of the accident scene image data in the multi-source data of the vehicle accident; the temporal features are extracted from the vehicle sensor data in the multi-source data of the vehicle accident to determine the damage time features; and a multi-dimensional feature matrix is ​​obtained based on the damage area features and the damage time features.

[0079] Furthermore, the multi-dimensional feature matrix establishment module M100 is also used to perform the following method:

[0080] By leveraging the cross-features of environmental data and vehicle attribute data from the multi-source vehicle accident data, a coupling factor between road surface friction coefficient and vehicle weight, an adaptation index between visibility and speed limit, and a dynamic response coefficient between vehicle wheelbase and steering angle are generated. The damage area features, damage time features, and cross-features are standardized, and the multi-dimensional feature matrix is ​​obtained according to the matrix structure of accident sample × feature dimension.

[0081] Furthermore, the correlation analysis module M200 is also used to perform the following methods:

[0082] The forward propagation process employs a multilayer perceptron, with the input layer being the multidimensional feature matrix and the hidden layer undergoing nonlinear transformation using the ReLU activation function. When analyzing the correlation strength between multiple influencing factors and various vehicle damage types, a weighted fusion of Pearson correlation coefficient and mutual information entropy is introduced.

[0083] Furthermore, the correlation analysis module M200 is also used to perform the following methods:

[0084] Hierarchical clustering is used to group the vehicle damage-dominant factors, and a correlation strength filtering threshold is used to screen potential correlation chains. The potential correlation chains are then subjected to time-series verification to obtain the effective correlation chains.

[0085] Furthermore, the correlation analysis module M200 is also used to perform the following methods:

[0086] Based on the dominant factors of vehicle damage, the inter-class distance is determined, and multiple factor components and multiple factor groups are obtained by dividing the data according to the inter-class distance. Path analysis is performed based on the multiple factor components and multiple factor groups. The multilayer perceptron is traversed by depth-first search. When the association strength values ​​of M consecutive network nodes in the vehicle damage propagation process all meet the association strength filtering threshold, they are marked as potential association chains, where M is greater than or equal to 2.

[0087] Furthermore, the correlation analysis module M200 is also used to perform the following methods:

[0088] On the potential associated chain, the contribution of the on-chain network nodes is obtained; based on the contribution of the on-chain network nodes, the cumulative value of the product of the time difference and the contribution of adjacent on-chain network nodes in the potential associated chain is obtained; when the cumulative value of the product of the time difference and the contribution meets the timing verification threshold, and the occurrence order of the on-chain nodes conforms to the accident mechanics transmission logic, the timing verification is determined to be passed.

[0089] Furthermore, the correlation analysis module M200 is also used to perform the following methods:

[0090] If the cumulative value of the product of the time difference and the contribution does not meet the time series verification threshold, a two-way tracing mechanism is used to deduce the extended potential association chain. The extended potential association chain is then time-tested by superimposing the cumulative value corresponding to the extended potential association chain with the time series compensation value and the time series verification threshold. If it still does not meet the time series verification threshold, the association chain is split and reorganized: the extended potential association chain is split into multiple sub-chains according to the peak of the time difference, and multiple sub-chains that pass the time series verification are retained; for multiple sub-chains that fail the time series verification, a secondary verification is performed in combination with the spatial damage correlation of the accident scene image data.

[0091] Furthermore, the correlation analysis module M200 is also used to perform the following methods:

[0092] By tracing back the preceding influencing factors of the first node of the chain, the correlation strength compensation value with the first node of the chain is determined; by extrapolating the derived damage of the last node of the chain, a potential damage propagation coefficient is generated; and the correlation strength compensation value and the potential damage propagation coefficient are weighted and fused to obtain the time-series compensation value.

[0093] In summary, any step can be stored as a computer instruction or program in an unrestricted computer memory and can be called and recognized by an unrestricted computer processor; no further restrictions are imposed here.

[0094] Furthermore, the above technical solutions only embody the preferred technical solutions of the embodiments of this application. Any changes that those skilled in the art may make to certain parts of these solutions embody the novel principles of the embodiments of this application. Obviously, those skilled in the art can make various modifications and variations to this application without departing from the scope of this application.

Claims

1. A vehicle damage correlation analysis method based on multi-source data, characterized in that, The method includes: Acquire multi-source vehicle accident data, including accident scene image data, vehicle sensor data, environmental data, and vehicle attribute data, and establish a multi-dimensional feature matrix; Based on the multi-dimensional feature matrix, correlation analysis is performed, and forward propagation is used to determine the correlation strength values ​​between multiple influencing factors and various vehicle damage types. The dominant factors of vehicle damage and effective correlation chains are identified, and a ranking of vehicle damage impact is generated. The vehicle damage dominant factors are used as core nodes for hierarchical division, the effective correlation chains are used as directional connection paths for network nodes, the multiple influencing factors and the associated vehicle damage types are used as network nodes, and the correlation strength value is used as the edge weight to construct a vehicle damage correlation analysis network. Based on the vehicle damage correlation analysis network and the vehicle damage impact ranking, a vehicle damage diagnosis report is generated. The method for establishing a multi-dimensional feature matrix includes: The damage region features are extracted from the accident scene image data in the multi-source data of the vehicle accident. Temporal features are extracted from the vehicle sensor data in the multi-source data of the vehicle accident to determine the damage time characteristics; Based on the damage area features and damage time features, a multi-dimensional feature matrix is ​​obtained; Based on the multi-dimensional feature matrix, correlation analysis is performed, and forward propagation is used to determine the correlation strength values ​​between multiple influencing factors and various vehicle damage types. The method includes: The forward propagation process employs a multilayer perceptron, with the input layer being the multidimensional feature matrix and the hidden layer undergoing nonlinear transformation using the ReLU activation function. When analyzing the correlation strength between multiple influencing factors and various types of vehicle damage, a weighted fusion of Pearson correlation coefficient and mutual information entropy is introduced; The method for identifying the dominant factors and effective correlation chains of vehicle damage includes: Based on the dominant factors of vehicle damage, hierarchical clustering is used to group the data, and the potential association chains are mined by filtering the association strength threshold. The potential association chains are subjected to time-series verification to obtain the effective association chains.

2. The vehicle damage correlation analysis method based on multi-source data as described in claim 1, characterized in that, Based on the damage region features and damage time features, a multi-dimensional feature matrix is ​​obtained. The method includes: By leveraging the cross-features of environmental data and vehicle attribute data in the multi-source vehicle accident data, a coupling factor between road surface friction coefficient and vehicle weight, an adaptation index between visibility and speed limit, and a dynamic response coefficient between vehicle wheelbase and steering angle are generated. The damage area features, damage time features, and intersection features are standardized, and the multi-dimensional feature matrix is ​​obtained by following the matrix structure of accident sample × feature dimension.

3. The vehicle damage correlation analysis method based on multi-source data as described in claim 1, characterized in that, Based on the dominant factors of vehicle damage, hierarchical clustering is used to group them, and combined with correlation strength filtering thresholds, potential correlation chains are mined. The method includes: Based on the dominant factors of vehicle damage, the inter-class distance is determined, and multiple factor components and multiple factor group classes are obtained by dividing the data according to the inter-class distance; Path analysis is performed based on the multiple factor components and multiple factor group classes. The multilayer perceptron is traversed by depth-first search. When the correlation strength values ​​of M consecutive network nodes in the vehicle damage transmission process all meet the correlation strength filtering threshold, they are marked as potential correlation chains, where M is greater than or equal to 2.

4. The vehicle damage correlation analysis method based on multi-source data as described in claim 3, characterized in that, The method for performing time-series verification on the potential association chains includes: On the potential association chain, obtain the contribution of the network nodes on the chain; Based on the contribution of the on-chain network nodes, obtain the cumulative value of the product of the time difference and the contribution of adjacent on-chain network nodes in the potential associated chains; When the cumulative value of the product of the time difference and the contribution meets the timing verification threshold, and the occurrence order of the nodes on the chain conforms to the accident mechanics transmission logic, the timing verification is determined to be passed.

5. The vehicle damage correlation analysis method based on multi-source data as described in claim 4, characterized in that, The method includes: If the cumulative value of the product of the time difference and the contribution does not meet the time-series verification threshold, a two-way tracing mechanism is used to deduce the extended potential association chain. The extended potential association chain is time-series verified by superimposing the cumulative value corresponding to the extended potential association chain with the time-series compensation value and the time-series verification threshold. If it still does not meet the time-series verification threshold, the association chain is split and reorganized: the extended potential association chain is split into multiple sub-chains according to the peak time difference, and multiple sub-chains that pass the time-series verification are retained; for multiple sub-chains that fail the time-series verification, a secondary verification is performed in combination with the spatial damage correlation of the accident scene image data.

6. The vehicle damage correlation analysis method based on multi-source data as described in claim 5, characterized in that, The extended potential association chain is obtained by using a two-way tracing mechanism. The method includes: By tracing back the preceding influencing factors of the first node in the chain, the correlation strength complement value with the first node in the chain is determined; By extrapolating the derived damage to the tail node, a potential damage propagation coefficient is generated. The temporal compensation value is obtained by weighting and fusing the correlation strength compensation value with the potential damage propagation coefficient.

7. A vehicle damage correlation analysis system based on multi-source data, characterized in that, The system is used to implement the vehicle damage correlation analysis method based on multi-source data as described in any one of claims 1-6, wherein the system comprises: Multi-dimensional feature matrix establishment module: Acquires multi-source vehicle accident data, including accident scene image data, vehicle sensor data, environmental data, and vehicle attribute data, and establishes a multi-dimensional feature matrix; Association Analysis Module: Based on the multi-dimensional feature matrix, association analysis is performed. Forward propagation is used to determine the association strength values ​​between multiple influencing factors and various vehicle damage types, identify the dominant factors of vehicle damage and effective association chains, and generate a ranking of vehicle damage impact. Hierarchical partitioning module: The vehicle damage dominant factors are used as core nodes for hierarchical partitioning, the effective correlation chains are used as directional connection paths of network nodes, the multiple influencing factors and the associated vehicle damage types are used as network nodes, and the correlation strength value is used as the edge weight to construct a vehicle damage correlation analysis network. Vehicle damage diagnosis report generation module: Based on the vehicle damage correlation analysis network and the vehicle damage impact ranking, generate a vehicle damage diagnosis report.

Images