Damaged part false alarm detection method, device, equipment and medium

Abstract

The invention relates to the technical field of artificial intelligence and the field of financial science and technology, and discloses a damaged part false alarm detection method, which comprises the following steps: mapping the position of a damaged part to a two-dimensional plane to generate a position point set; clustering the position points by using a clustering model, and determining global collision points; two-dimensional normal distribution with the collision point as the center is constructed, and the probability density of each position point is calculated; calculating a comprehensive probability density based on the maximum probability density; and determining a probability density threshold value by using a false alarm judgment model, comparing the probability density threshold value with the comprehensive probability density, and identifying a false alarm part. According to the method, the damaged parts are accurately modeled by using clustering analysis and probability density calculation, and the accuracy and the automation degree of false alarm identification can be effectively improved by combining with the false alarm judgment model trained based on historical data, the additional claim settlement cost of an insurance company is reduced, and more efficient and intelligent claim settlement management is realized.

Description

Technical Field

[0001] The present invention relates to the fields of artificial intelligence technology and financial technology, and in particular to a method, device, equipment and storage medium for detecting false alarms of damaged parts. Background Art

[0002] In automobile insurance loss assessment scenarios, insurance companies typically settle claims for damaged vehicle parts based on information reported by assessors. However, in practice, there is a certain rate of misreporting of damaged parts by assessors. This misreporting often results in additional claims fees for insurance companies, increasing operating costs and impacting the fairness and accuracy of claims processing.

[0003] Currently, the industry typically uses rule-based approaches based on human experience and knowledge to prevent false positives. These solutions rely primarily on a manually formulated rule base. For example, they determine whether a false positive exists by defining the plausibility of certain component combinations. When a reported component combination fails to conform to the established rules, the system triggers an alarm and displays the relevant violation information. This rule-based approach to false positive identification has improved claims accuracy to a certain extent, but its limitations and shortcomings are also very obvious.

[0004] First, the development and updating of rules is slow, unable to keep pace with changes in vehicle structure and claims models, resulting in the system's inability to flexibly respond to new claims scenarios. Second, these rules rely on one-sided information, which can easily lead to misjudgments or omissions. For example, certain damage situations may be outside the rules and cannot be effectively identified. In addition, the rule-setting process is mainly based on the subjective experience of experts and lacks objective data support, which makes the rules less objective and scientific. Furthermore, the transmission of human experience is prone to gaps, and newcomers find it difficult to master the complete judgment criteria when taking over, further affecting the accuracy of claims.

[0005] More critically, rule maintenance is expensive. The rule base requires regular maintenance and updates, requiring significant manpower and time. Furthermore, this reliance on manual rule creation struggles to maintain effectiveness and comprehensiveness in complex and ever-changing claims scenarios, especially when dealing with large amounts of data.

[0006] In summary, the existing rule-based methods cannot meet the increasingly complex claims needs of the insurance industry. There is an urgent need for an automated, intelligent, and easy-to-maintain false alarm identification solution to improve the efficiency and accuracy of claims, reduce unnecessary economic losses, and achieve more efficient claims management. Summary of the Invention

[0007] The main purpose of the present invention is to provide a method, device, equipment and storage medium for detecting false alarms of damaged parts, aiming to solve the technical problems of the existing technology in detecting false alarm parts with insufficient accuracy and difficulty in adapting to complex damage situations.

[0008] To achieve the above objectives, the present invention provides a method for detecting false alarms of damaged parts, comprising:

[0009] Mapping the positions of multiple detected damaged parts to a two-dimensional plane to generate a corresponding set of position points for each damaged part;

[0010] Use the clustering model to cluster the location points in all location point sets to form one or more clusters, and use the center point of each cluster as the global collision point;

[0011] Construct a two-dimensional normal distribution with each global collision point as the center, and determine the probability density of each position point of each damaged component in all two-dimensional normal distributions;

[0012] The maximum probability density of each location point in all normal distributions is taken as the final probability density of the location point;

[0013] Taking a weighted average of the final probability densities of all locations of each damaged component to obtain the comprehensive probability density of the component;

[0014] Using a false alarm determination model trained based on historical data, a probability density threshold for determining false alarms is determined;

[0015] comparing the combined probability density of each damaged component with the probability density threshold;

[0016] If the comprehensive probability density of a damaged component is less than the probability density threshold, the damaged component is determined to be a false positive component.

[0017] Furthermore, to achieve the above-mentioned object, the present invention provides a device for detecting false alarms of damaged parts, comprising:

[0018] A position mapping module is used to map the positions of multiple damaged parts detected onto a two-dimensional plane and generate a corresponding set of position points for each damaged part;

[0019] A cluster analysis module is used to cluster the location points in the set of all location points using a clustering model to form one or more clusters, and use the center point of each cluster as the global collision point;

[0020] The probability distribution module is used to construct a two-dimensional normal distribution with each global collision point as the center and determine the probability density of each position point of each damaged component in all two-dimensional normal distributions;

[0021] A probability density determination module, configured to take the maximum probability density of each location point in all normal distributions as the final probability density of the location point;

[0022] A weighted average module is used to perform a weighted average of the final probability densities of all locations of each damaged component to obtain a comprehensive probability density of the component;

[0023] A model inference module is used to determine a probability density threshold for determining false alarms using a false alarm determination model trained based on historical data;

[0024] a comparison and determination module, configured to compare the comprehensive probability density of each damaged component with the probability density threshold;

[0025] The false alarm identification module is used to determine that a damaged component is a false alarm component if the comprehensive probability density of the damaged component is less than the probability density threshold.

[0026] Furthermore, to achieve the above-mentioned purpose, the present invention also provides a computer device, which includes a memory, a processor, and a damaged parts false alarm detection program stored in the memory and runnable on the processor, and when the damaged parts false alarm detection program is executed by the processor, the steps of the damaged parts false alarm detection method as described above are implemented.

[0027] Furthermore, to achieve the above-mentioned purpose, the present invention also provides a computer-readable storage medium, on which a damaged parts false alarm detection program is stored. When the damaged parts false alarm detection program is executed by a processor, the steps of the damaged parts false alarm detection method described above are implemented.

[0028] Beneficial effects: The present invention relates to the fields of artificial intelligence technology and financial technology, and discloses a method for detecting false alarms of damaged parts. The method maps the positions of damaged parts to a two-dimensional plane to generate a set of position points; clusters the position points using a clustering model to determine the global collision point; constructs a two-dimensional normal distribution centered on the collision point to calculate the probability density of each position point; calculates the comprehensive probability density based on the maximum probability density; uses a false alarm determination model to determine the probability density threshold, and compares it with the comprehensive probability density to identify false alarm parts. The present invention uses cluster analysis and probability density calculation to accurately model damaged parts, combined with a false alarm determination model trained based on historical data, which can effectively improve the accuracy and automation of false alarm identification, reduce the insurance company's additional claims costs, and achieve more efficient and intelligent claims management. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] The present invention will be further described below with reference to the accompanying drawings and embodiments, in which:

[0030] Figure 1 A schematic diagram of an application environment of a method for detecting false positives of damaged parts according to an embodiment of the present invention;

[0031] Figure 2 This is a flow chart of an embodiment of a method for detecting false positives of damaged parts according to the present invention;

[0032] Figure 3 This is a schematic diagram of the functional modules of a preferred embodiment of the damaged parts false alarm detection device of the present invention;

[0033] Figure 4 A schematic diagram of the structure of a computer device according to an embodiment of the present invention;

[0034] Figure 5 FIG. 2 is another structural diagram of a computer device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0035] It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0036] The method for detecting false alarms of damaged parts provided by the embodiment of the present invention can be applied in the following situations: Figure 1 In an application environment, the user terminal communicates with the server terminal through a network. The server terminal can map the position of the damaged parts to a two-dimensional plane through the user terminal to generate a set of position points; cluster the position points using a clustering model to determine the global collision point; construct a two-dimensional normal distribution centered on the collision point to calculate the probability density of each position point; calculate the comprehensive probability density based on the maximum probability density; use the false alarm judgment model to determine the probability density threshold, and compare it with the comprehensive probability density to identify the false alarm parts. The present invention uses cluster analysis and probability density calculation to accurately model the damaged parts, and combines it with the false alarm judgment model trained based on historical data, which can effectively improve the accuracy and automation of false alarm identification, reduce the insurance company's additional claims costs, and achieve more efficient and intelligent claims management. Among them, the user terminal can be but is not limited to various personal computers, laptops, smart phones, tablet computers and portable wearable devices. The server terminal can be implemented with an independent server or a server cluster composed of multiple servers. The present invention is described in detail below through specific embodiments.

[0037] See also Figure 2 , Figure 2 This is a flow chart of an embodiment of a method for detecting false positives of damaged parts provided by the present invention. It should be noted that although a logical order is shown in the flow chart, in some cases, the steps shown or described may be performed in a different order than that shown here.

[0038] like Figure 2As shown, the damaged parts false alarm detection method proposed by the present invention includes the following steps:

[0039] S10, mapping the positions of the plurality of detected damaged parts to a two-dimensional plane, and generating a corresponding position point set for each damaged part;

[0040] In this embodiment, damaged parts refer to vehicle parts that are damaged when the vehicle is involved in a collision or accident. Specifically, these parts are damaged due to external impact and need to be repaired or replaced. For example, in a car accident, the front bumper, doors, headlights, hood and other parts may be damaged, and these are all called "damaged parts." If the front side of the car collides with another car, it may cause damage to the front bumper and hood, and these parts are considered damaged parts. In a side collision, the doors and side mirrors may be damaged, and these are also damaged parts.

[0041] The position information of multiple damaged parts detected in three-dimensional space is converted to generate a two-dimensional plane representation. Since the damaged parts of the car are distributed in three-dimensional space, the positions of different parts need to be identified in a unified two-dimensional coordinate system for subsequent analysis and clustering. First, the car is modeled as a three-dimensional rectangular model, and the initial position of each part is represented using a standard three-dimensional coordinate system. Then, through mathematical transformations (such as projection transformations or coordinate mapping functions), the three-dimensional coordinates are converted to two-dimensional coordinates, and the position points of each damaged part are mapped onto a two-dimensional plane, forming a set of two-dimensional plane position points.

[0042] After mapping the damaged parts onto a two-dimensional plane, a corresponding set of location points is generated for each part. This set describes the specific damage range of the part and serves as the basis for subsequent analysis. By obtaining the coordinate range of each damaged part on the two-dimensional plane, a set of discrete location points is generated. This can be achieved by scanning or segmenting the projected area of ​​each part on the plane. Each point is recorded in a data structure (such as a two-dimensional array or coordinate list) and associated with the corresponding part label.

[0043] In one specific implementation, a complete 3D cuboid model of the vehicle body is created. A coordinate mapping method based on the 3D model is used to convert component positions into a 2D representation. First, the coordinate system of the 3D cuboid model of the vehicle body is determined. Then, using the vehicle's geometric characteristics, coordinate mapping rules for each component are defined. A data structure (such as a hash table) is used to store the correspondence between 3D and 2D positions.

[0044] Example: In a car insurance claims system, insurance companies can use this solution to process large amounts of vehicle damage data, mapping the locations of damaged parts across different cases into a unified coordinate system for analysis. This mapping approach can help identify abnormal data or false positives in claims applications, thereby improving data analysis efficiency and reducing insurance companies' claims costs.

[0045] By mapping the locations of damaged parts onto a two-dimensional plane and generating a set of location points, we can provide a structured data foundation for subsequent cluster analysis and probability calculations, enabling more efficient analysis of the extent of component damage. This avoids directly processing complex three-dimensional spatial data, reduces computational complexity, and improves the accuracy and efficiency of false alarm identification.

[0046] S20, clustering the location points in all location point sets using a clustering model to form one or more clusters, and using the center point of each cluster as a global collision point;

[0047] In this embodiment, all damaged component location points mapped onto a two-dimensional plane are grouped. Using a clustering model, closely related points are grouped together to facilitate identification of different damaged areas. Clustering models can be used to classify locations based on their spatial distribution. Commonly used clustering models include K-means and DBSCAN.

[0048] Select an appropriate clustering model based on the distribution characteristics of the points. For example, when using K-means clustering, set the number of clusters, K; when using DBSCAN, set the minimum number of samples and the distance threshold. Input the 2D coordinates of all points into the selected clustering model, which automatically classifies the point collection based on the specified parameters. The model outputs one or more clusters, each representing a concentrated area of ​​damage.

[0049] All points are grouped into one or more groups based on spatial proximity, with each group representing a potential area of ​​collision damage. Each class can contain multiple points, which are clustered together to represent a localized area of ​​concentrated damage. Based on the output of the clustering algorithm, each point is assigned to a group, forming one or more classes. A data structure (such as an array or list) is used to store the set of points in each class for subsequent analysis and calculations.

[0050] Within each cluster, the center point of all points is calculated and considered the global collision point. This center point is typically calculated based on the average position or center of mass of all points, representing the center of the damage area and the core point most likely to be impacted. Mathematical methods are used to calculate the average coordinates of all points within each cluster, or the center of mass calculation method is used to determine the location of the center point. This calculated center point is used as the global collision point for subsequent probabilistic analysis.

[0051] By using a clustering model to classify all location points, the collision area of ​​damaged parts can be accurately identified and located, reducing human error and improving the ability to automatically identify damaged areas. By calculating the global collision point, a more accurate central reference point can be provided for subsequent probability analysis, thereby improving the accuracy of false alarm determination.

[0052] S30, constructing a two-dimensional normal distribution with each global collision point as the center, and determining the probability density of each position point of each damaged component in all two-dimensional normal distributions;

[0053] In this embodiment, a two-dimensional normal distribution is established using each global collision point as the center point. This distribution is used to represent the influence range around the collision point, with the probability density being highest at the collision point and gradually decreasing with increasing distance.

[0054] The two-dimensional coordinates of each global impact point are determined and used as the center point (mean) of the normal distribution. Based on the physical characteristics of the vehicle damage and historical collision data, the spread and shape (i.e., standard deviation parameters) of each distribution are defined. The standard deviation determines the width of the distribution, with a larger standard deviation indicating a wider impact range. Using programming tools or mathematical models, a function representing the distribution is created to subsequently calculate the probability density at each location point.

[0055] Calculate the probability density value of each damaged component's location point in the two-dimensional normal distribution of each global collision point. The probability density value indicates the intensity and probability of each location point being affected by the collision.

[0056] The distance from each location point to all global collision points is measured. Closer distances generally indicate higher probability densities, indicating a greater likelihood of impact. Based on the previously defined distribution range (standard deviation), the probability density value for each location point in each normal distribution is calculated. This value quantifies the degree of impact at that location point. During implementation, batch computations are performed using programming languages ​​and related libraries (such as Python's NumPy or SciPy) to ensure computational efficiency and accuracy.

[0057] By constructing a two-dimensional normal distribution centered on the global collision point and calculating the probability density at each location, we can precisely quantify the damage impact range and ensure high accuracy in identifying false positive parts. This quantitative analysis of probability density not only improves automation but also reduces misjudgments during the claims process, minimizing financial losses for insurance companies.

[0058] S40, taking the maximum probability density of each location point in all normal distributions as the final probability density of the location point;

[0059] In this embodiment, to determine the most likely extent to which each location point will be affected by a collision, the probability density values ​​for that location point in the two-dimensional normal distribution constructed from all global collision points are compared and the maximum value is selected. This maximum value reflects the highest likelihood that the location point will be affected near a particular collision point and is thus used as the final probability density for that location point.

[0060] For each location point, calculate its probability density value in the two-dimensional normal distribution corresponding to all global collision points in turn, and store these values ​​in a list or array. Traverse all stored probability density values, find the maximum value, and use this value as the final probability density of the location point. This process can be efficiently implemented using built-in functions in programming languages ​​(such as the max() function in Python). The final probability density value of each location point is stored in a data structure for further processing and analysis in subsequent steps.

[0061] By taking the maximum probability density of each location point in all normal distributions as the final probability density, the highest probability of each location point being affected by a collision can be effectively identified, thereby providing a more accurate and reliable data basis for subsequent false alarm determination.

[0062] S50, performing a weighted average of the final probability densities of all locations of each damaged component to obtain a comprehensive probability density of the component;

[0063] In this embodiment, to determine the overall damage level of each damaged component, the final probability density values ​​for all locations on that component are weighted averaged. Weights can be assigned based on the importance or damage level of the location to highlight the impact of certain key locations on the comprehensive assessment.

[0064] Each location is assigned a weight based on its importance or degree of damage. Weights can be set by analyzing historical data or drawing on expert experience. More critical or vulnerable locations can be assigned higher weights. The final probability density for each location is multiplied by its corresponding weight to obtain a weighted probability density. The weighted probability densities of all locations are summed together to obtain a total. This sum of weighted probability densities is divided by the sum of all weights to obtain the overall probability density for the damaged component.

[0065] The comprehensive probability density quantifies the overall damage level of each damaged component, reflecting the overall likelihood of that component being affected by the collision. A weighted average method provides a more accurate picture of the overall damage level of a component. The component's comprehensive probability density is calculated by dividing the accumulated weighted probability density by the sum of all weights. The comprehensive probability density for each component is stored in a data structure to facilitate false positive detection in subsequent steps.

[0066] By taking a weighted average of the final probability density of all locations for each damaged component, we can provide a comprehensive probability density indicator to more accurately quantify the extent of damage to each component. This rational distribution of weights ensures the impact of key damage points on the overall assessment, optimizes the process of identifying falsely reported components, and improves the intelligence and accuracy of the claims process.

[0067] S60, using a false alarm determination model trained based on historical data, determining a probability density threshold for determining false alarms;

[0068] In this embodiment, a false positive detection model is used to determine whether there are false positives for damaged parts in a claim. This model is trained using historical claims data, which includes information such as the characteristics of damaged parts, probability density values, and claim outcomes. The model learns from this data to identify patterns and characteristics of false positives.

[0069] Collect a large amount of historical data from past claims, including relevant information such as the location distribution of damaged parts in each claim, probability density values, the vehicle's collision speed and angle, and annotated data to identify any false positives. Preprocess the historical data to extract and construct feature data suitable for model learning, such as the comprehensive probability density of location points, part type, and vehicle motion parameters. Feature engineering can include normalization and data cleaning to ensure data quality. Select an appropriate machine learning model (such as a random forest or neural network) for training, using the feature data as input and the claim results or false positives as output labels. Improve the model's prediction accuracy and generalization capabilities through repeated iterations and model optimization.

[0070] The trained false positive detection model outputs a probability density threshold based on the input feature data. This threshold is used to determine whether a damaged component is likely to be a false positive. If the overall probability density of a component falls below this threshold, it is considered a false positive.

[0071] The characteristic data of damaged parts in the current claim case is input into the trained false alarm judgment model, and the model will output a probability density threshold based on these characteristic data.

[0072] The threshold of the model output is dynamic and can be adjusted according to different claims scenarios and data characteristics to improve the accuracy of false alarm identification.

[0073] By utilizing a false positive identification model trained on historical data, the system intelligently and dynamically adjusts probability density thresholds based on different claims scenarios and data characteristics, effectively improving the accuracy of false positive part identification. Compared to traditional fixed rule-based approaches, the model continuously optimizes and learns from new data, reducing human misjudgments and missed detections, lowering insurance company claims costs, and improving the efficiency of the overall claims process.

[0074] S70, comparing the comprehensive probability density of each damaged component with the probability density threshold;

[0075] In this embodiment, the comprehensive probability density value of each damaged component is compared with the probability density threshold previously output by the false alarm determination model to determine whether the component is likely to be a false alarm. This step can effectively screen out false alarm components and reduce the risk of erroneous claims.

[0076] First, the comprehensive probability density of each damaged component is extracted. This value is calculated by taking a weighted average of the probability densities of all locations. The probability density threshold used for judgment is obtained from the threshold output by the false alarm determination model. This threshold changes dynamically based on the characteristic data of the current case. The comprehensive probability density value of each component is compared with the threshold. If the comprehensive probability density is less than the threshold, the probability of damage to the component is low and may be a false alarm; otherwise, the damage to the component is considered reasonable. Based on the comparison results, each component is labeled or classified to provide a basis for subsequent processing.

[0077] By comparing the combined probability density of each damaged component with the specified probability density threshold, we can accurately screen out potentially falsely reported components, avoid unnecessary claims expenses, and improve the accuracy and efficiency of claims processing. This dynamic determination mechanism automatically adjusts to the specific characteristics of each case, enhancing the intelligence of claims management.

[0078] S80: If the comprehensive probability density of a damaged component is less than the probability density threshold, the damaged component is determined to be a false positive component.

[0079] In this embodiment, the comprehensive probability density of each damaged component is compared with the threshold provided by the false alarm determination model. If the comprehensive probability density value is lower than the threshold, it indicates that the damage level of the component is low and the probability of damage is insufficient to support a claim, indicating that the component may be a false alarm.

[0080] A simple conditional check is performed for each damaged component. If the combined probability density is less than a pre-defined probability density threshold, the component is marked as a potential false positive. The results are stored in a data structure (such as a Boolean array, a list of tags, or a database table) for further processing or manual review. This step ensures that the false positive status of each component can be recorded and retrieved.

[0081] Once the combined probability density of a part is determined to be below a threshold, the part is definitively identified as a misreported part. A misreported part is a part that is incorrectly reported as damaged in a claim and should not be covered.

[0082] Parts identified as false positives are categorized and may trigger a series of follow-up actions, such as generating an automated report, notifying an auditor for review, or updating the claims system. The system can be configured to further process false positive part information, such as generating an alert, initiating manual review, or outright denying the claim.

[0083] By determining that parts with a comprehensive probability density lower than a threshold are false positive parts, damaged parts that should not be compensated can be efficiently identified, thereby avoiding unnecessary compensation and reducing the financial losses of insurance companies.

[0084] The present invention relates to the fields of artificial intelligence technology and financial technology, and discloses a method for detecting false alarms of damaged parts. The method comprises mapping the positions of damaged parts onto a two-dimensional plane to generate a set of position points; clustering the position points using a clustering model to determine the global collision point; constructing a two-dimensional normal distribution centered on the collision point to calculate the probability density of each position point; calculating the overall probability density based on the maximum probability density; and using a false alarm determination model to determine a probability density threshold and compare it with the overall probability density to identify false alarm parts. The present invention uses cluster analysis and probability density calculation to accurately model damaged parts, combined with a false alarm determination model trained on historical data, to effectively improve the accuracy and automation of false alarm identification, reduce additional claims costs for insurance companies, and achieve more efficient and intelligent claims management.

[0085] In one embodiment, the above S10 includes:

[0086] S101, modeling the target object into a three-dimensional cuboid model;

[0087] S102, dividing the two-dimensional plane on the rectangular parallelepiped model related to the damaged component into a plurality of grid areas;

[0088] S103, determining the position coordinates of each damaged component in the rectangular parallelepiped model;

[0089] S104: Map the position coordinates to corresponding grid areas to generate a set of position points corresponding to each damaged component.

[0090] In this example, the vehicle is modeled in three dimensions, treating it as a cuboid, to more accurately represent the spatial location of damaged components. The three-dimensional model provides a spatial reference for position mapping, ensuring the relative positions of different components in space are accurately recorded.

[0091] Use 3D modeling software or a mathematical model to create a rectangular parallelepiped model of the vehicle. Each plane of the rectangular parallelepiped model corresponds to the six major damaged surfaces of the vehicle: front, rear, left, right, top, and bottom. Input the basic dimensions and shape parameters of the vehicle to construct a standard rectangular parallelepiped model, providing a unified spatial framework.

[0092] To convert 3D spatial positions into 2D positions, each plane of the cuboid model needs to be divided into multiple small mesh regions. Each mesh region describes the location of the damaged component on the 2D plane. The planes associated with the damaged component are divided into equally spaced meshes, with a fixed mesh size chosen to ensure uniform division. These mesh regions are then used to precisely locate the damaged location of each component. The appropriate number and spacing of mesh regions are selected based on the vehicle size and damage accuracy requirements to balance accuracy and computational complexity.

[0093] Obtain the spatial coordinate information of the damaged component, describing its specific location within the 3D cuboid model. Position coordinates are typically obtained from sensor data, manual input, or 3D scanning.

[0094] Use sensors or scanning equipment to obtain the spatial coordinates (e.g., X, Y, and Z coordinates) of damaged parts. If manual input is used, detailed vehicle damage information must be provided to manually annotate the coordinates. Pre-process the acquired coordinates to ensure data accuracy for subsequent mapping operations.

[0095] The three-dimensional position coordinates are converted and mapped into a two-dimensional grid area to generate a set of two-dimensional position points for the damaged component. Each position point represents the specific damaged location of the component on the two-dimensional plane. The three-dimensional coordinates are converted into two-dimensional coordinates through mathematical transformations (such as projection transformation or coordinate mapping) and matched to the relevant planes of the rectangular model. Based on the converted two-dimensional coordinates, each damaged component is mapped to the corresponding grid area, and all position points are stored in a data structure (such as an array or list).

[0096] This embodiment precisely maps the locations of multiple detected damaged parts onto a two-dimensional plane and generates a corresponding set of location points, providing an accurate data foundation for subsequent damage analysis and false positive detection. This mapping method simplifies complex calculations of three-dimensional spatial positions, improves the accuracy of false positive component identification, and provides an efficient solution for insurance claims management systems.

[0097] In one embodiment, the above S20 includes:

[0098] S201, determining a clustering model according to distribution characteristics of the location points in the set of all location points;

[0099] S202, inputting the location points into the clustering model, and dividing the location points into one or more clusters using the clustering model;

[0100] S203 , determining the center point of each class according to the average position of all position points in each class, and using the center point as the global collision point of the corresponding class.

[0101] In this embodiment, before clustering, it is necessary to select an appropriate clustering model. The specific model selected depends on the distribution characteristics of the location points. Common clustering models include K-means and DBSCAN. K-means is suitable for situations where the point distribution is relatively uniform and the number of clusters is preset, while DBSCAN is suitable for situations where the point distribution is dense and noisy.

[0102] First, analyze the distribution of all location points to determine the data density, distribution pattern, and possible number of clusters. Based on the analysis, select an appropriate clustering algorithm. For example, if the location points are clearly clustered and the number of clusters is unknown, choose DBSCAN; if the number of clusters can be reasonably estimated, use K-means.

[0103] All location points are passed as input to the selected clustering model. The clustering model then uses its algorithm to classify the location points into clusters, with each cluster representing a single lesion area. The 2D coordinates of all location points are input to the clustering model, which then performs the clustering operation. The model then classifies the location points into one or more clusters based on a distance metric and parameter settings. For K-means, the number of clusters, K, must be specified; for DBSCAN, the minimum number of samples and the neighborhood distance must be set.

[0104] The cluster's center is determined by calculating the average position of all points within each cluster. This center is defined as the global collision point, representing the center of the primary collision zone corresponding to that cluster. For each cluster, the X and Y coordinates of all points are averaged to determine the coordinates of the center point. This calculated center point is marked as the global collision point and provides the basis for subsequent damage analysis and probability calculations.

[0105] This embodiment clusters all location point sets by using a clustering model and uses the center point of each cluster as the global collision point, which can effectively identify and locate the core position of the damaged area and provide an accurate collision point reference.

[0106] In one embodiment, the above S30 includes:

[0107] S301, setting the position coordinates of each global collision point to the mean parameter of the corresponding two-dimensional normal distribution;

[0108] S302, setting the standard deviation parameter of each two-dimensional normal distribution;

[0109] S303, determining the distance between each position point and all global collision points;

[0110] S304 : determining the probability density of each position point under each two-dimensional normal distribution based on the distance between each position point and each global collision point and the standard deviation parameter of each two-dimensional normal distribution.

[0111] In this embodiment, the coordinates of each global collision point represent the center of a two-dimensional normal distribution, known as the mean parameter. This parameter determines how the probability density of the distribution reaches its highest point at the collision point and gradually decreases with distance. For each global collision point, its two-dimensional coordinates (X, Y) are used as the mean parameter of the normal distribution. The mean parameter is used to construct the center point of the distribution, thereby reflecting the starting point of the collision impact.

[0112] The standard deviation parameter determines the spread and shape of the two-dimensional normal distribution. A larger standard deviation indicates a wider distribution of impact, while a smaller standard deviation indicates a concentrated impact near the impact point. Based on the physical characteristics of vehicle damage or empirical data, set an appropriate standard deviation parameter for each global impact point. The standard deviation can be adjusted based on the impact intensity, component material, or vehicle type to ensure that the distribution accurately reflects the actual damage impact.

[0113] Calculate the distance between each damaged component's location and all global collision points. Distance is a key factor influencing probability density. Closer distances indicate higher probability density, while greater distances indicate lower probability density. Use the two-dimensional Euclidean distance formula or other appropriate distance metric to calculate the distance from each location point to all global collision points. Distance calculations can be implemented using vectorized operations for improved efficiency.

[0114] Using the distance from each location point to the global collision point and the standard deviation parameter, the probability density of each location point under each normal distribution is calculated. The probability density indicates the likelihood that a location point will be affected by the collision point, with higher values ​​indicating a greater likelihood of impact. Using a standard probability density function, the probability density of each location point is determined based on its distance from the global collision point and the standard deviation parameter. Mathematical libraries (such as Python's SciPy) can be used to efficiently calculate probability densities in programming.

[0115] This embodiment achieves accurate quantification of damage impact by constructing a two-dimensional normal distribution with the global collision point as the center and calculating the probability density of each position point.

[0116] In one embodiment, before the above S60, the method further includes:

[0117] S601, collecting historical claims data, and extracting feature data related to the determination of false alarms of damaged parts from the historical claims data;

[0118] S602, determining a basic machine learning model based on the feature data;

[0119] S603: Use the feature data to train the basic machine model to generate a false alarm determination model for determining a probability density threshold.

[0120] In this embodiment, data from past claims cases is collected for subsequent false positive detection model training. Historical claims data may include details about each case, such as the location of damaged parts, damage type, vehicle speed, collision angle, and claim outcome. Relevant historical data is extracted from the insurance company's claims database to ensure coverage of diverse claim scenarios and damage types. The collected data is stored in a structured database and organized and cleaned to remove duplicate or invalid data to improve data quality.

[0121] Feature data refers to important indicators that determine whether a damaged part is a false positive. Examples include the location distribution of damaged parts, vehicle speed, collision angle, and probability density values. Extracting feature data helps improve model accuracy and effectiveness. Analyze historical claims data to extract key features relevant to false positive determinations. Features can be numerical, categorical, or combined, and may require normalization or standardization. Clean and transform feature data to ensure consistent data formatting and address missing or outliers to optimize model training.

[0122] Select an appropriate basic machine learning model, such as a random forest, support vector machine, decision tree, or neural network, for training and learning rules for false positive detection. Model selection should be based on the nature and complexity of the feature data. Evaluate different machine learning algorithms and select the most appropriate model based on the distribution and dimensionality of the feature data. For example, a random forest model is suitable for processing feature data with high dimensionality and nonlinear relationships, while a neural network is suitable for learning complex patterns. Set model hyperparameters, such as the number of trees, learning rate, and activation function, to improve model performance.

[0123] The extracted feature data is fed into a basic machine learning model for training to generate a model capable of automatically identifying false positives based on current claims data. The model output is a probability density threshold for comparison. The feature data is divided into a training set and a validation set, and the model is trained using supervised learning methods. The accuracy and generalization of the model are improved by optimizing the loss function (such as mean squared error or cross-entropy loss) and adjusting model parameters. The model performance is evaluated using the validation set, and the model hyperparameters are adjusted based on the evaluation results to ensure that the model can effectively identify false positives. The resulting model will determine the probability density threshold for false positives during the actual claims process.

[0124] This embodiment uses a false alarm determination model trained on historical data to automatically learn false alarm determination rules and dynamically output probability density thresholds based on specific claims data. This reduces the subjectivity and inaccuracy of human judgment, improves the efficiency and accuracy of false alarm identification, and effectively reduces insurance companies' claims costs.

[0125] In one embodiment, the above S60 includes:

[0126] S605, obtaining current feature data of the target object, wherein the feature data includes all position point distributions, movement speeds, and collision angles of damaged parts;

[0127] S606, inputting all position point distributions, movement speeds, and collision angles of the damaged parts into the false alarm determination model;

[0128] S607: Determine a probability density threshold currently used for determining false alarms through the false alarm determination model.

[0129] In this embodiment, the claim analysis process requires the acquisition of characteristic data related to the current claim scenario. This data includes the location distribution of damaged parts (describing the specific location and area of ​​damage), vehicle speed (affecting the intensity and extent of the collision), and collision angle (determining the directionality of the damage).

[0130] Sensors or accident analysis software are used to obtain the location distribution of damaged components on a two-dimensional plane. This can be achieved using laser scanning, 3D imaging, or damage record data. The vehicle's speed and angle of movement at the time of the collision are obtained from the accident recorder, the vehicle's black box, or other sensors. This data can generally accurately describe the physical characteristics of the collision.

[0131] The acquired feature data is passed as input to a pre-trained false alarm detection model. The model uses this data to analyze and determine whether a false alarm is likely. All feature data is consolidated into an input format acceptable to the model. For example, the location distribution can be represented as a matrix, and the speed and angle can be combined as numerical features into a single input vector. This consolidated input vector is then fed into the false alarm detection model, triggering the model to perform inference. Data preprocessing methods, such as normalization or standardization, can be used to improve the model's analysis accuracy.

[0132] The false positive detection model calculates the input feature data and outputs a probability density threshold. This threshold is used to determine whether a part in the current claim scenario is likely to be a false positive. The higher the threshold, the stricter the model's false positive detection criteria.

[0133] The false alarm determination model infers the input feature data and outputs a dynamically adjusted probability density threshold. The model utilizes a variety of algorithms, such as neural networks or decision trees, to make accurate decisions. The output probability density threshold is saved for use in subsequent steps when determining false alarms for damaged parts. The threshold can be recorded in the claims management system for data tracking and analysis.

[0134] This embodiment utilizes a false alarm determination model trained on historical data to adjust and output the probability density threshold for false alarm determination in real time based on the characteristic data of the current claims scenario. This improves the accuracy and flexibility of false alarm identification, reduces the occurrence of false alarm claims, and lowers claims costs for insurance companies. Furthermore, the dynamic threshold determination method can adapt to changing accident scenarios, enhancing the automation and intelligence of the claims process.

[0135] In one embodiment, the above S50 includes:

[0136] S501, assigning a weight to each location point of each damaged component according to the importance or damage degree of the location point;

[0137] S502, generating a weighted probability density for each location point based on the final probability density and corresponding weight of each location point;

[0138] S503, accumulating the weighted probability densities of all the location points to obtain the sum of the weighted probability densities;

[0139] S504: Determine the comprehensive probability density of each component based on the sum of the weighted probability densities and the sum of the weights of all position points.

[0140] In this embodiment, each damaged component has multiple locations on a two-dimensional plane. These locations may have different degrees of damage or different importance in determining the overall damage of the component. Therefore, a weight is assigned to each location to reflect its relative importance in the overall assessment.

[0141] Weights can be assigned based on various criteria, such as proximity to the global collision point, importance of the component's overall impact based on historical damage data, or expert experience. This can be achieved using simple linear weight assignment or automated machine learning models.

[0142] To obtain the weighted probability density of each location point, you need to multiply the final probability density of the location point by its corresponding weight. The weighted probability density reflects the contribution of the location point to the overall evaluation.

[0143] For each location point, calculate the weighted probability density = final probability density × weight. In this way, the impact of different location points on the overall probability density of the component can be weighted and adjusted.

[0144] The weighted probability densities of all locations are accumulated to obtain a total weighted probability density value. This value is used to calculate the combined probability density. Each location is traversed and its weighted probability density is accumulated to obtain the sum of the weighted probability densities. This accumulation operation can be implemented using a loop structure or matrix operations to improve computational efficiency.

[0145] The comprehensive probability density quantifies the overall damage probability of each component. It is calculated by dividing the sum of the weighted probability densities by the sum of the weights of all locations. This balances the contribution of each location to the overall judgment and provides a more accurate damage assessment. The formula used is:

[0146] The final overall probability density is calculated by dividing the sum of the weighted probability densities by the sum of the weights. This value is used for subsequent false positive determinations or claims analysis.

[0147] This embodiment more accurately quantifies the overall damage probability of each component by taking a weighted average of the final probability density of all locations for each damaged component. The weighting method is flexible and can be adjusted to suit different damage scenarios. The weighted average calculation improves the accuracy of false positive detection, thereby optimizing the claims process, reducing the risk of false positives, and lowering insurance company claims costs.

[0148] In one embodiment, a device for detecting false alarms of damaged parts is provided, which corresponds one-to-one to the method for detecting false alarms of damaged parts in the above embodiment. Figure 3 , Figure 3 This is a schematic diagram of the functional modules of a preferred embodiment of the damaged parts false alarm detection device of the present invention. These modules include a position mapping module 10, a cluster analysis module 20, a probability distribution module 30, a probability density determination module 40, a weighted average module 50, a model inference module 60, a comparison and determination module 70, and a false alarm identification module 80. Each functional module is described in detail below:

[0149] A position mapping module 10 is used to map the positions of the multiple damaged parts detected to a two-dimensional plane, and generate a corresponding position point set for each damaged part;

[0150] A cluster analysis module 20 is configured to cluster all the location points in the location point set using a clustering model to form one or more clusters, and use the center point of each cluster as a global collision point;

[0151] The probability distribution module 30 is used to construct a two-dimensional normal distribution with each global collision point as the center, and determine the probability density of each position point of each damaged component in all two-dimensional normal distributions;

[0152] A probability density determination module 40 is configured to take the maximum probability density of each location point among all normal distributions as the final probability density of the location point;

[0153] A weighted averaging module 50 is used to perform a weighted average of the final probability densities of all locations of each damaged component to obtain a comprehensive probability density of the component;

[0154] A model reasoning module 60 is configured to determine a probability density threshold for determining false alarms using a false alarm determination model trained based on historical data;

[0155] a comparison and determination module 70, configured to compare the comprehensive probability density of each damaged component with the probability density threshold;

[0156] The false alarm identification module 80 is configured to determine that a damaged component is a false alarm component if the comprehensive probability density of the damaged component is less than the probability density threshold.

[0157] In one embodiment, the location mapping module 10 is specifically configured to:

[0158] Model the target object as a three-dimensional cuboid model;

[0159] Dividing a two-dimensional plane on the cuboid model related to the damaged component into a plurality of grid areas;

[0160] Determine the position coordinates of each damaged component in the rectangular parallelepiped model;

[0161] The position coordinates are mapped to corresponding grid areas to generate a set of position points corresponding to each damaged component.

[0162] In one embodiment, the cluster analysis module 20 is specifically configured to:

[0163] Determine a clustering model based on the distribution characteristics of the location points in the set of all location points;

[0164] Inputting the location points into the clustering model, and dividing the location points into one or more clusters using the clustering model;

[0165] The center point of each class is determined according to the average position of all position points in each class, and the center point is used as the global collision point of the corresponding class.

[0166] In one embodiment, the probability distribution module 30 is specifically configured to:

[0167] Set the position coordinates of each global collision point to the mean parameter of the corresponding two-dimensional normal distribution;

[0168] Set the standard deviation parameter for each two-dimensional normal distribution;

[0169] Determine the distance between each position point and all global collision points;

[0170] Based on the distance between each position point and each global collision point and the standard deviation parameter of each two-dimensional normal distribution, the probability density of each position point under each two-dimensional normal distribution is determined.

[0171] In one embodiment, the model inference module 60 is specifically configured to:

[0172] Collecting historical claims data, and extracting feature data related to the determination of false alarms of damaged parts from the historical claims data;

[0173] Determining a basic machine learning model based on the feature data;

[0174] The basic machine model is trained using the feature data to generate a false alarm determination model for determining a probability density threshold.

[0175] In one embodiment, the model inference module 60 is specifically configured to:

[0176] Acquiring current feature data of the target object, including the distribution of all locations of damaged parts, movement speed, and collision angle;

[0177] Inputting all position point distributions, movement speeds, and collision angles of the damaged parts into the false alarm determination model;

[0178] The probability density threshold currently used for determining false alarms is determined by the false alarm determination model.

[0179] In one embodiment, the weighted averaging module 50 is specifically configured to:

[0180] Assign a weight to each location point of each damaged component according to the importance or damage degree of the location point;

[0181] Generate a weighted probability density for each location point based on the final probability density and corresponding weight of each location point;

[0182] Accumulate the weighted probability densities of all location points to obtain the sum of the weighted probability densities;

[0183] Based on the sum of the weighted probability densities and the sum of the weights of all location points, the comprehensive probability density of each component is determined.

[0184] In one embodiment, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as follows: Figure 4 As shown. The computer device includes a processor, memory, network interface, and database connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes non-volatile and / or volatile storage media and internal memory. The non-volatile storage medium stores an operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external user terminal via a network connection. When the computer program is executed by the processor, it implements the functions or steps on the server side of a method for detecting false alarms of damaged parts.

[0185] In one embodiment, a computer device is provided. The computer device may be a user terminal, and its internal structure diagram may be as follows: Figure 5 As shown. The computer device includes a processor, memory, network interface, display screen and input device connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external server via a network connection. When the computer program is executed by the processor, it realizes the functions or steps on the user side of a method for detecting false alarms of damaged parts.

[0186] In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the following steps are performed:

[0187] Mapping the positions of multiple detected damaged parts to a two-dimensional plane to generate a corresponding set of position points for each damaged part;

[0188] Use the clustering model to cluster the location points in all location point sets to form one or more clusters, and use the center point of each cluster as the global collision point;

[0189] Construct a two-dimensional normal distribution with each global collision point as the center, and determine the probability density of each position point of each damaged component in all two-dimensional normal distributions;

[0190] The maximum probability density of each location point in all normal distributions is taken as the final probability density of the location point;

[0191] Taking a weighted average of the final probability densities of all locations of each damaged component to obtain the comprehensive probability density of the component;

[0192] Using a false alarm determination model trained based on historical data, a probability density threshold for determining false alarms is determined;

[0193] comparing the combined probability density of each damaged component with the probability density threshold;

[0194] If the comprehensive probability density of a damaged component is less than the probability density threshold, the damaged component is determined to be a false positive component.

[0195] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the following steps are implemented:

[0196] Mapping the positions of multiple detected damaged parts to a two-dimensional plane to generate a corresponding set of position points for each damaged part;

[0197] Use the clustering model to cluster the location points in all location point sets to form one or more clusters, and use the center point of each cluster as the global collision point;

[0198] Construct a two-dimensional normal distribution with each global collision point as the center, and determine the probability density of each position point of each damaged component in all two-dimensional normal distributions;

[0199] The maximum probability density of each location point in all normal distributions is taken as the final probability density of the location point;

[0200] Taking a weighted average of the final probability densities of all locations of each damaged component to obtain the comprehensive probability density of the component;

[0201] Using a false alarm determination model trained based on historical data, a probability density threshold for determining false alarms is determined;

[0202] comparing the combined probability density of each damaged component with the probability density threshold;

[0203] If the comprehensive probability density of a damaged component is less than the probability density threshold, the damaged component is determined to be a false positive component.

[0204] It should be noted that the above functions or steps that can be implemented by the computer-readable storage medium or computer device can be found in the relevant descriptions of the server side and the user side in the aforementioned method embodiment. To avoid repetition, they will not be described one by one here.

[0205] Those skilled in the art will appreciate that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0206] Those skilled in the art will clearly understand that for the sake of convenience and brevity of description, only the division of the above-mentioned functional units and modules is used as an example. In actual applications, the above-mentioned functions can be distributed and completed by different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0207] It should be noted that if any software tools or components other than those of the Company appear in the embodiments of this application, they are merely for illustration and do not represent actual use. The above embodiments are intended only to illustrate the technical solutions of the present invention, not to limit them. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the above embodiments, or replace some of the technical features therein with equivalents. These modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included in the scope of protection of the present invention.

Claims

1. A method for detecting false alarms of damaged parts, characterized in that: The following steps are involved: Mapping the positions of multiple damaged parts detected to a two-dimensional plane, and generating a corresponding position point set for each damaged part; Clustering the location points in all location point sets using a clustering model to form one or more clusters, and using the center point of each cluster as the global collision point; Construct a two-dimensional normal distribution with each global collision point as the center, and determine the probability density of each position point of each damaged component in all two-dimensional normal distributions; Taking the maximum probability density of each location point in all normal distributions as the final probability density of the location point; Taking a weighted average of the final probability densities of all the position points of each damaged component to obtain a comprehensive probability density of the component; Using a false alarm determination model trained based on historical data, a probability density threshold for determining false alarms is determined; comparing the combined probability density of each damaged component with the probability density threshold; If the comprehensive probability density of a damaged component is less than the probability density threshold, the damaged component is determined to be a false positive component.

2. The method for detecting false alarms of damaged parts according to claim 1, characterized in that: The positions of multiple damaged parts detected are mapped to a two-dimensional plane, and a corresponding set of position points is generated for each damaged part, including: Model the target object as a three-dimensional cuboid model; Dividing a two-dimensional plane on the cuboid model related to the damaged component into a plurality of grid areas; Determine the position coordinates of each damaged component in the rectangular parallelepiped model; The position coordinates are mapped to corresponding grid areas to generate a set of position points corresponding to each damaged component.

3. The method for detecting false alarms of damaged parts according to claim 1, characterized in that: Use the clustering model to cluster the location points in all location point sets to form one or more classes, and use the center point of each class as the global collision point, including: Determine a clustering model according to the distribution characteristics of the location points in the set of all location points; Inputting the location points into the clustering model, and dividing the location points into one or more classes by the clustering model; According to the average position of all position points in each class, the center point of each class is determined, and the center point is used as the global collision point of the corresponding class.

4. The method for detecting false alarms of damaged parts according to claim 1, characterized in that: A two-dimensional normal distribution is constructed with each global collision point as the center, and the probability density of each position point of each damaged component in all two-dimensional normal distributions is determined, including: Set the position coordinates of each global collision point to the mean parameter of the corresponding two-dimensional normal distribution; Set the standard deviation parameter for each 2D normal distribution; Determine the distance between each position point and all global collision points; Based on the distance between each position point and each global collision point and the standard deviation parameter of each two-dimensional normal distribution, the probability density of each position point under each two-dimensional normal distribution is determined.

5. The method for detecting false alarms of damaged parts according to claim 1, characterized in that: Before determining the probability density threshold for determining false alarms using the false alarm determination model trained based on historical data, the following steps are also included: Collecting historical claims data, and extracting characteristic data related to the determination of false alarms of damaged parts from the historical claims data; Determine a basic machine learning model based on the feature data; The basic machine model is trained using the feature data to generate a false alarm determination model for determining a probability density threshold.

6. The method for detecting false positives of damaged parts as claimed in claim 1, characterized in that: Use the false alarm determination model trained based on historical data to determine the probability density threshold for determining false alarms, including: Acquire current feature data of the target object, wherein the feature data includes all position point distributions, moving speeds, and collision angles of damaged parts; Inputting all position point distributions, moving speeds and collision angles of the damaged parts into the false alarm determination model; The probability density threshold currently used for determining false alarms is determined by the false alarm determination model.

7. The method for detecting false alarms of damaged parts according to claim 1, characterized in that: The final probability density of all position points of each damaged component is weighted averaged to obtain the comprehensive probability density of the component, including: Assign a weight to each location point of each damaged component according to the importance or degree of damage of the location point; Based on the final probability density and corresponding weight of each location point, a weighted probability density of each location point is generated; Accumulate the weighted probability densities of all location points to obtain the sum of the weighted probability densities; Based on the sum of the weighted probability densities and the sum of the weights of all location points, the comprehensive probability density of each component is determined.

8. A damaged parts false alarm detection device, characterized in that: The damaged parts false alarm detection device comprises: A position mapping module, used to map the positions of multiple damaged parts detected to a two-dimensional plane, and generate a corresponding position point set for each damaged part; A clustering analysis module, used for clustering the location points in all location point sets using a clustering model to form one or more classes, and taking the center point of each class as a global collision point; The probability distribution module is used to construct a two-dimensional normal distribution with each global collision point as the center, and determine the probability density of each position point of each damaged component in all two-dimensional normal distributions; A probability density determination module, used to take the maximum probability density of each location point in all normal distributions as the final probability density of the location point; A weighted average module, used to perform a weighted average on the final probability density of all position points of each damaged component to obtain a comprehensive probability density of the component; A model reasoning module, used to determine a probability density threshold for determining a false alarm using a false alarm determination model trained based on historical data; A comparison and determination module, used for comparing the comprehensive probability density of each damaged component with the probability density threshold; The false alarm identification module is used to determine that a damaged component is a false alarm component if the comprehensive probability density of the damaged component is less than the probability density threshold.

9. A computer device, characterized in that: The computer device includes a memory, a processor, and a damaged parts false alarm detection program stored in the memory and executable on the processor. When the damaged parts false alarm detection program is executed by the processor, the steps of the damaged parts false alarm detection method as described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium, characterized in that: The storage medium stores a damaged parts false alarm detection program, which, when executed by the processor, implements the steps of the damaged parts false alarm detection method according to any one of claims 1 to 7.

Images