Online loss assessment system based on cancel-after-verification mechanism

Abstract

The invention discloses an online loss assessment system based on a cancel-after-verification mechanism, and the system can automatically extract the features of a damaged part of a vehicle from a loss assessment picture submitted by a user through employing an advanced image processing technology and a natural language processing technology, carries out the semantic understanding of the description of a maintenance condition, and improves the maintenance efficiency. Therefore, the in-depth understanding and accurate evaluation of the vehicle loss condition and the required maintenance work can be realized. Furthermore, deep conjoint analysis is performed on the image features of the vehicle damage area and the semantic features of the maintenance condition to mine the potential association relationship between the image features and the semantic features, and intelligent auditing of the insurance company is realized based on the potential association relationship, so that the auditing accuracy is improved, the fraudulent behavior is effectively prevented, and the safety of the insurance company is improved. Therefore, each claim settlement case can be processed fairly and quickly, so that the vehicle insurance claim settlement service is promoted to develop in a faster and more fair direction.

Description

Technical Field

[0001] The present application relates to the field of intelligent auditing, and more specifically, to an online damage assessment system based on a write-off mechanism. Background Art

[0002] In the field of vehicle insurance claims, the inefficiency and lack of accuracy of the traditional damage assessment process have long plagued insurance companies and car owners. In particular, when it comes to reviewing photos of damaged vehicles and describing repairs, the traditional approach relies heavily on the experience and judgment of insurer personnel, a method that is both inefficient and inaccurate. Manual review can overlook subtle but important details, affecting the final claim decision. More seriously, this experience-based review method has limited ability to detect fraudulent activities such as false claims or exaggeration of the extent of losses, exposing insurance companies to a high risk of financial losses. This problem is further exacerbated during peak periods, when insurance companies handle a large number of cases, making it difficult for manual review to maintain high speed and quality.

[0003] While online damage assessment systems have become increasingly popular with the development of internet technology, streamlining the claims process through digital means, they still face numerous flaws in practical applications. For example, many online platforms lack effective verification mechanisms, failing to ensure the authenticity of damage assessments and making them vulnerable to fraudulent activities. Furthermore, existing online damage assessment systems often lack intelligent review mechanisms for damage assessment photos and repair descriptions, resulting in inefficient review processes and a tendency to overlook potential anomalies.

[0004] Therefore, an optimized online loss assessment system based on the write-off mechanism is expected. Summary of the Invention

[0005] In order to solve the above technical problems, the present application is proposed. The embodiment of the present application provides an online damage assessment system based on a write-off mechanism, which, by adopting advanced image processing technology and natural language processing technology, can automatically extract the features of the damaged parts of the vehicle from the damage assessment photos submitted by the user, and perform semantic understanding of the repair situation description, thereby achieving an in-depth understanding and accurate assessment of the vehicle's loss status and the required repair work. Furthermore, a deep joint analysis is performed on the image features of the damaged area of the vehicle and the semantic features of the repair situation to explore the potential correlation between the two, and based on this, the intelligent review of the insurance company is realized. In this way, the accuracy of the review is improved, the occurrence of fraud is effectively prevented, and it is ensured that each claim case can be handled fairly and quickly, thereby promoting the development of vehicle insurance claims services in a faster and fairer direction.

[0006] According to one aspect of the present application, an online damage assessment system based on a write-off mechanism is provided, which includes:

[0007] A damage identification code presentation module is used to present a damage identification code;

[0008] An authorization confirmation module is used to verify the damage identification code after scanning the damage identification code and pop up an authorization confirmation page for the vehicle disassembly authorization agreement;

[0009] A window prompt module is used to, in response to receiving the consent action, determine whether the party responsible for the accident has insured "commercial insurance"; if the party responsible for the accident has not insured "commercial insurance", a risk prompt window will pop up; if the party responsible for the accident has insured "commercial insurance", a vehicle information registration page will pop up;

[0010] The damage assessment module is used to receive damage assessment photos and repair status descriptions uploaded by users;

[0011] The review module is used to send the damage assessment photos and repair situation description to the insurance company for review to obtain the review results.

[0012] Compared with the existing technology, the online damage assessment system based on the write-off mechanism provided by this application, by adopting advanced image processing technology and natural language processing technology, can automatically extract the features of the damaged parts of the vehicle from the damage assessment photos submitted by the user, and perform semantic understanding of the repair situation description, thereby achieving an in-depth understanding and accurate assessment of the vehicle's loss status and the required repair work. Furthermore, a deep joint analysis is performed on the image features of the damaged area of the vehicle and the semantic features of the repair situation to explore the potential correlation between the two, and based on this, the insurance company's intelligent audit is realized. In this way, the accuracy of the audit is improved, the occurrence of fraud is effectively prevented, and every claim case is ensured to be handled fairly and quickly, thereby promoting the development of vehicle insurance claims services in a faster and more fair direction. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] The above and other purposes, features, and advantages of the present application will become more apparent through a more detailed description of the embodiments of the present application in conjunction with the accompanying drawings. The accompanying drawings are intended to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation of the present application. In the drawings, the same reference numerals generally represent the same components or steps.

[0014] Figure 1 1 is a block diagram of an online damage assessment system based on a write-off mechanism according to an embodiment of the present application;

[0015] Figure 2 Schematic diagram of data flow in an online damage assessment system based on a write-off mechanism according to an embodiment of the present application;

[0016] Figure 3 1 is a block diagram of an audit module in an online damage assessment system based on a write-off mechanism according to an embodiment of the present application;

[0017] Figure 4 This is a block diagram of a vehicle damage-vehicle repair cross-modal joint analysis unit in an online damage assessment system based on a write-off mechanism according to an embodiment of the present application. DETAILED DESCRIPTION

[0018] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described herein.

[0019] As used in this application and the claims, unless the context clearly indicates otherwise, the words "a," "an," "an," and / or "the" are not intended to refer to the singular but may include the plural. Generally speaking, the terms "comprises" and "include" only indicate the inclusion of the steps and elements specifically identified, and these steps and elements do not constitute an exclusive list. A method or apparatus may also include other steps or elements.

[0020] Although the present application makes various references to certain modules in the system according to embodiments of the present application, any number of different modules can be used and run on the user terminal and / or server. The modules are illustrative only, and different aspects of the system and method can use different modules.

[0021] Flowcharts are used in this application to illustrate the operations performed by the systems according to the embodiments of the present application. It should be understood that the preceding or following operations are not necessarily performed in exact order. Instead, the various steps may be processed in reverse order or simultaneously, as needed. Furthermore, other operations may be added to these processes, or one or more operations may be removed from these processes.

[0022] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described herein.

[0023] Based on this, in the technical solution of the present application, the present application proposes an online damage assessment system based on a write-off mechanism, which includes: presenting a damage assessment identification code; after scanning the damage assessment identification code, writing off the damage assessment identification code and popping up an authorization confirmation page for the vehicle disassembly authorization agreement; in response to receiving the consent action, judging whether the party responsible for the accident has insured "commercial insurance", if the party responsible for the accident has not insured "commercial insurance", a risk warning window pops up; if the party responsible for the accident has insured "commercial insurance", a vehicle information registration page pops up; receiving damage assessment photos and repair situation descriptions uploaded by users; sending the damage assessment photos and repair situation descriptions to the insurance company for review to obtain the review results. Among them, by introducing the write-off mechanism of the damage assessment identification code, the system effectively prevents repeated submission or tampering in the damage assessment process, ensuring the uniqueness and authenticity of the damage assessment data. Secondly, the traditional damage assessment process relies on manual review, which is inefficient and prone to errors. The present application uses a deep learning model to achieve intelligent analysis of damage assessment data, greatly improves the review efficiency, and reduces the subjectivity and errors of human judgment through algorithms. Specifically, the technical concept of the present application is

[0024] Specifically, in the technical solution of the present application, an online damage assessment system based on a write-off mechanism is proposed. Figure 1 This is a block diagram of an online damage assessment system based on a write-off mechanism according to an embodiment of the present application. Figure 2 Schematic diagram of data flow of online damage assessment system based on write-off mechanism according to the embodiment of the present application. Figure 1 and Figure 2 As shown, the online damage assessment system 300 based on the write-off mechanism according to the embodiment of the present application includes: a damage assessment identification code presentation module 310, which is used to present the damage assessment identification code; an authorization confirmation module 320, which is used to verify the damage assessment identification code after scanning the damage assessment identification code and pop up the authorization confirmation page of the vehicle disassembly authorization agreement; a window prompt module 330, which is used to judge whether the party responsible for the accident is insured with "commercial insurance" in response to receiving the consent action. If the party responsible for the accident is not insured with "commercial insurance", a risk prompt window pops up; if the party responsible for the accident is insured with "commercial insurance", a vehicle information registration page pops up; a damage assessment module 340, which is used to receive damage assessment photos and repair situation descriptions uploaded by users; and an audit module 350, which is used to send the damage assessment photos and repair situation descriptions to the insurance company for audit to obtain the audit results.

[0025] In particular, the damage assessment identification code presentation module 310 is used to present the damage assessment identification code. That is, when the user needs to conduct a vehicle damage assessment, he will first obtain a damage assessment identification code specially generated for this case through the front-end interface of the system. This identification code is usually automatically generated by the system according to a certain algorithm and contains a series of information, such as timestamp, case number, etc., to ensure its uniqueness. Users can obtain this damage assessment identification code in a variety of ways, such as receiving SMS, sending emails, or directly displaying it on the application interface. Once the user obtains the damage assessment identification code, the user needs to present the damage assessment identification code to ensure that it can be accurately located on the same claim case. By scanning the damage assessment identification code, the system can quickly identify the corresponding case information and start the next series of processes.

[0026] Specifically, the authorization confirmation module 320 is configured to verify the damage assessment identification code after scanning it and display the authorization confirmation page for the vehicle disassembly authorization agreement. Specifically, once the user presents the damage assessment identification code and it is scanned by the personnel or organization performing the damage assessment, a series of automated operations are initiated. First, the system verifies the scanned damage assessment identification code, ensuring that each code can only be used once, thereby preventing duplicate submission or tampering. This verification mechanism is implemented by updating the identification code's status from "available" to "used." Once an identification code is marked as used, it cannot be used for any other operations, significantly enhancing system security and data authenticity. Subsequently, after the identification code verification is complete, the system automatically displays the authorization confirmation page for the vehicle disassembly authorization agreement. This page is designed to ensure that the vehicle owner or party responsible for the accident is fully aware of the upcoming operations and has formally consented to them. This authorization confirmation page typically details the scope of the upcoming repair work, estimated costs, and relevant liability clauses. The purpose of this is to increase transparency and ensure that car owners or parties responsible for accidents have sufficient understanding of the entire process and can make informed decisions.

[0027] Specifically, the window prompt module 330 is configured to, in response to receiving the consent action, determine whether the responsible party is insured with commercial insurance. If not, a risk warning window will pop up; if so, a vehicle information registration page will pop up. In other words, in an online damage assessment system based on a write-off mechanism, once the user completes the confirmation of the vehicle disassembly authorization agreement, the system will immediately respond to this consent and determine the next steps based on whether the responsible party is insured with commercial insurance. This process is implemented through backend database queries and logical analysis, ensuring accurate identification of the user's insurance status and providing appropriate services accordingly. Once the system receives the user's consent, it will first automatically initiate a backend query of the responsible party's insurance status. This query typically relies on data interaction with the insurance company's core system, obtaining the latest insurance information by calling relevant APIs or directly accessing shared databases. During this process, the system uses the personal information provided by the responsible party (such as license plate number, ID number, etc.) as query criteria to ensure accurate matching of corresponding insurance records. If the query results show that the party responsible for the accident has not purchased "commercial insurance", the system will immediately pop up a risk warning window. This window is designed to remind users of the risks they may currently face. For example, in the absence of commercial insurance, certain types of losses may not be fully compensated. These prompts not only help to raise users' awareness, but also help users understand the financial risks they may face at an early stage, prompting them to take necessary measures to protect their rights and interests. On the contrary, if the query results show that the party responsible for the accident has indeed purchased "commercial insurance", the system will guide users to the vehicle information registration page. This page is designed to collect information required for further processing of claims, including but not limited to the specific model of the vehicle, purchase date, mileage, and detailed description of the damage. In this way, the system can ensure that all necessary information is recorded accurately and prepare for subsequent damage assessment and claims processing.

[0028] In particular, the damage assessment module 340 is used to receive damage assessment photos and repair situation descriptions uploaded by users. The damage assessment photos include multi-angle views of the damaged parts of the vehicle so that the system can fully understand the specific circumstances of the damage; the repair situation description includes the damage condition of the vehicle and the recommended repair measures. Taking into account that the technical backgrounds of different users may vary, the system may provide some guiding questions or prompts to help users better organize their descriptions, such as asking about the specific type of damage (scratches, dents, etc.), the name of the parts that are expected to be replaced or repaired, and the general expectations for the repair work. Doing so not only helps users organize information more systematically, but also facilitates the background algorithm to understand the repair needs more accurately.

[0029] In particular, the review module 350 is used to send the damage assessment photos and repair situation description to the insurance company for review to obtain the review results. Specifically, in a specific example of this application, if Figure 3 As shown, the audit module 350 includes: an image quality verification unit 351, which is used to perform image quality verification on the damage assessment image to obtain an image quality verification result; an image processing unit 352, which is used to pop up a prompt to re-upload the damage assessment image if the image quality verification result is that the image quality does not meet the requirements; if the image quality verification result is that the image quality meets the requirements, extract the image features of the damaged parts of the vehicle from the damage assessment image to obtain the image feature coding vector of the damaged parts of the vehicle; a maintenance situation semantic understanding unit 353, which is used to perform text semantic understanding on the maintenance situation description to obtain the maintenance situation description semantic understanding coding vector; a vehicle damage-vehicle maintenance cross-modal joint analysis unit 354, which is used to perform vehicle damage-vehicle maintenance cross-modal joint analysis based on core information anchoring on the vehicle damage part image feature coding vector and the maintenance situation description semantic understanding coding vector to obtain the vehicle damage-vehicle maintenance semantic response interaction coding vector; and an audit result generation unit 355, which is used to obtain the audit result based on the vehicle damage-vehicle maintenance semantic response interaction coding vector.

[0030] Specifically, the image quality verification unit 351 is used to perform image quality verification on the damage assessment image to obtain an image quality verification result. It should be understood that the damage assessment image uploaded by the user may have various quality problems, such as image blur, insufficient light, angle deviation and other problems. In the process of vehicle insurance claims, the quality of the damage assessment image directly determines whether the system can accurately extract the features of the damaged parts of the vehicle, thereby affecting the final review results. If the image quality does not meet the requirements, such as image blur, insufficient light, angle deviation or traces of human tampering, the system will not be able to effectively identify the details of the vehicle damage. Therefore, in the technical solution of the present application, the image quality of the damage assessment image is verified. In one example, the image quality of the damage assessment image can be verified by blur detection and light correction to ensure that the uploaded damage assessment image has sufficient clarity and integrity. In this way, the system can effectively avoid misjudgment or missed judgment due to image quality problems, thereby improving the accuracy and efficiency of the review.

[0031] Specifically, the image processing unit 352 is configured to prompt the user to re-upload the damage assessment image if the image quality verification result indicates that the image quality does not meet the requirements. If the image quality verification result indicates that the image quality meets the requirements, the image features of the damaged parts of the vehicle are extracted from the damage assessment image to obtain a vehicle damage image feature encoding vector. In other words, in a specific example, if the image is blurred or poorly lit, the system will immediately prompt the user to re-upload a satisfactory image. For example, the system may prompt "The uploaded image has insufficient lighting. Please retake the image and ensure uniform lighting" or "The image clarity is insufficient. Please retake the image and ensure accurate focus." This instant feedback mechanism not only guides users to upload high-quality images but also effectively avoids subsequent processing errors caused by image quality issues. Conversely, in the technical solution of the present application, if the image quality verification result indicates that the image quality meets the requirements, the damage assessment image is input into a vehicle damage feature extractor based on a DenseNet network to obtain a vehicle damage image feature encoding vector. It should be understood that damage assessment images contain key information about vehicle damage, but this information is often hidden within complex visual data, such as detailed features such as scratches, dents, and cracks. Traditional image processing methods can struggle to effectively extract these detailed features, especially when image quality is high but the damage area is complex. However, DenseNet, as an advanced deep learning model, possesses powerful feature extraction capabilities. It leverages the multi-level information in the image through dense connections, allowing for more accurate capture of the detailed features of vehicle damage.

[0032] Specifically, the maintenance situation semantic understanding unit 353 is configured to perform textual semantic understanding on the maintenance situation description to obtain a maintenance situation description semantic understanding encoding vector. In the technical solution of this application, the maintenance situation description is input into a maintenance situation semantic encoder based on the RoBERTa model to obtain the maintenance situation description semantic understanding encoding vector. It should be understood that in vehicle insurance claims, maintenance situation descriptions may contain a large amount of professional terminology and detailed descriptions, such as part names, damage levels, and repair recommendations. This information needs to be accurately understood and extracted to match the damaged area in the image. Traditional text processing methods struggle to accurately capture the underlying meaning, especially when describing complex or highly specialized maintenance content, which can easily lead to semantic understanding errors. The RoBERTa model, a pre-trained language model based on the Transformer architecture, possesses powerful semantic understanding capabilities and can capture the underlying meaning of text through contextual information, thereby more accurately extracting key information from the maintenance situation description. Using the RoBERTa model to perform textual semantic understanding on the maintenance situation description can more accurately understand the maintenance requirements and avoid misjudgments or omissions due to insufficient semantic understanding.

[0033] Specifically, the vehicle damage-vehicle repair cross-modal joint analysis unit 354 is configured to perform a core information-anchored vehicle damage-vehicle repair cross-modal joint analysis on the vehicle damage image feature encoding vector and the repair description semantic understanding encoding vector to obtain a vehicle damage-vehicle repair semantic response interaction encoding vector. It should be understood that in an online vehicle insurance claims assessment system, the vehicle damage image feature encoding vector and the repair description semantic understanding encoding vector represent the visual information of the vehicle damage and the textual description of the repair requirements, respectively. Complex correlations exist between vehicle damage and repair requirements. For example, the severity of vehicle damage may directly affect the formulation of a repair plan, while certain descriptions in the repair description may not align with the image features. Traditional unimodal analysis methods are unable to capture this cross-modal correlation information, resulting in potential bias or omissions in the review results. To address this issue, a core information-anchored vehicle damage-vehicle repair cross-modal joint analysis is performed on the vehicle damage image feature encoding vector and the repair description semantic understanding encoding vector to obtain a vehicle damage-vehicle repair semantic response interaction encoding vector. Through cross-modal joint analysis based on core information anchoring, the system can extract discriminative core feature expressions from high-dimensional feature space, eliminate the interference of redundant and noisy information, and thus more accurately model the relationship between vehicle damage and maintenance needs. Specifically, in this process, by constructing a semantic autocorrelation association matrix of the vehicle damage part image feature encoding vector and the repair description semantic understanding encoding vector, and using the core information anchoring network based on autocorrelation decoupling to refine and purify the internal information of the feature, the system can highlight the parts that are highly relevant to the core semantics and reduce redundant information. Subsequently, through the two-level interaction modeling of feature granularity and feature value granularity, the system can capture the high-order interaction pattern between vehicle damage and maintenance needs at the semantic level, while refining the modeling of local response relationships at the numerical level. This method enables the system to explore the implicit dependency between the two in a complex feature space, providing a solid data foundation for subsequent intelligent review. In this way, not only the damage assessment process is optimized, but also the risk of economic losses faced by insurance companies due to fraud is reduced, and the overall operational efficiency and user experience are improved. Specifically, in a specific example of the present application, such as Figure 4As shown, the vehicle damage-vehicle maintenance cross-modal joint analysis unit 354 includes: a semantic autocorrelation association matrix construction subunit 3541, which is used to construct a vehicle condition semantic autocorrelation association matrix of the vehicle damage part image feature coding vector and the maintenance condition description semantic understanding coding vector to obtain a vehicle damage part image semantic autocorrelation association matrix and a maintenance condition description semantic autocorrelation association matrix; a vehicle damage-vehicle maintenance fine-grained response interaction subunit 3542, which is used to perform fine-grained response interaction based on vehicle damage-maintenance condition anchoring on the vehicle damage part image semantic autocorrelation association matrix and the maintenance condition description semantic autocorrelation association matrix to obtain a vehicle damage-vehicle maintenance feature granularity response interaction coding subunit vector and vehicle damage-vehicle maintenance characteristic value granularity response interaction coding vector; a characteristic disturbance compensation subunit 3543, used to perform characteristic disturbance compensation on the vehicle damage-vehicle maintenance characteristic granularity response interaction coding vector and the vehicle damage-vehicle maintenance characteristic value granularity response interaction coding vector to obtain the compensated vehicle damage-vehicle maintenance characteristic granularity response interaction coding vector and the compensated vehicle damage-vehicle maintenance characteristic value granularity response interaction coding vector; a fusion subunit 3544, used to fuse the compensated vehicle damage-vehicle maintenance characteristic granularity response interaction coding vector and the compensated vehicle damage-vehicle maintenance characteristic value granularity response interaction coding vector to obtain the vehicle damage-vehicle maintenance semantic response interaction coding vector.

[0034] More specifically, the semantic autocorrelation matrix construction subunit 3541 is used to construct a vehicle condition semantic autocorrelation matrix between the vehicle damage image feature encoding vector and the repair description semantic understanding encoding vector, thereby obtaining a vehicle damage image semantic autocorrelation matrix and a repair description semantic autocorrelation matrix. By constructing the semantic autocorrelation matrix, the system can effectively mine implicit interactions between features and their semantic dependency structures in a high-dimensional feature space. This not only improves computational efficiency and reduces the computational burden caused by redundant information, but also enhances the accuracy of feature modeling. In this way, the system can more accurately identify subtle differences between vehicle damage and repair requirements, and even detect potential anomalies. In specific implementations, this step utilizes a dot product operation or other similarity measurement method to explicitly generate an autocorrelation matrix by making the interrelationships between the components of the vehicle damage image feature encoding vector and the repair description semantic understanding encoding vector explicit. This not only preserves the global structure of the feature vectors but also captures their internal semantic patterns. In a specific example, the following semantic autocorrelation association formula is used to construct a vehicle condition semantic autocorrelation association matrix of the vehicle damage part image feature encoding vector and the repair condition description semantic understanding encoding vector to obtain a vehicle damage part image semantic autocorrelation association matrix and a repair condition description semantic autocorrelation association matrix; wherein the semantic autocorrelation association formula is:

[0035]

[0036] Among them, v1 represents the image feature coding vector of the vehicle damage part, v2 represents the semantic understanding coding vector of the maintenance description, and φ(·) are feature mapping functions, such as linear mapping or nonlinear kernel function, M1 represents the semantic autocorrelation matrix of the vehicle damage part image, and M2 represents the semantic autocorrelation matrix of the maintenance situation description.

[0037] More specifically, the vehicle damage-vehicle repair fine-grained response interaction subunit 3542 is configured to perform fine-grained response interaction based on vehicle damage-repair situation anchoring on the semantic autocorrelation matrix of the vehicle damage area image and the semantic autocorrelation matrix of the repair description to obtain a vehicle damage-vehicle repair feature granularity response interaction encoding vector and a vehicle damage-vehicle repair feature value granularity response interaction encoding vector. In an embodiment of the present application, the semantic autocorrelation matrix of the vehicle damage area image and the semantic autocorrelation matrix of the repair description are first input into a vehicle damage-repair situation anchoring network based on autocorrelation decoupling to obtain an anchor encoding vector for the core information of the vehicle damage area image features and an anchor encoding vector for the core information of the repair description semantic features. It should be understood that the semantic autocorrelation matrix of the vehicle damage area image and the semantic autocorrelation matrix of the repair description represent high-dimensional representations of the vehicle damage image features and the semantic features of the repair description, respectively. However, these high-dimensional features often contain a large amount of redundant information and noise. Directly using these features for joint analysis results in high computational complexity and poor modeling results. To address this issue, the system introduces a core information anchoring network based on autocorrelation decoupling. This system extracts and purifies information from the semantic autocorrelation matrix of the vehicle damage image and the semantic autocorrelation matrix of the repair description to obtain the core information anchor encoding vectors of the vehicle damage image features and the core information anchor encoding vectors of the repair description semantic features. During this process, autocorrelation decoupling is used to perform feature distillation on the semantic autocorrelation matrix of the vehicle damage image and the semantic autocorrelation matrix of the repair description, highlighting the portions of the feature vector that are highly correlated with the core semantics. This process is similar to information filtering. By constraining information entropy, the system can reduce the noise of redundant information and retain the most discriminative feature information. This allows the system to more accurately capture the complex relationship between vehicle damage and repair needs and identify potential anomalies. In a specific example, the semantic autocorrelation matrix of the vehicle damage part image and the semantic autocorrelation matrix of the repair condition description are respectively input into the vehicle damage-repair condition anchoring network based on autocorrelation decoupling to obtain the core information anchor coding vector of the vehicle damage part image feature and the core information anchor coding vector of the repair condition description semantic feature; wherein, the core information anchoring formula is:

[0038]

[0039]

[0040] Among them, decouple(·) represents feature decoupling, x 11 ,x 12 ,x 1i ,x1n The first, second, i-th and n-th row vectors of the semantic autocorrelation matrix of the vehicle damage part image are respectively represented as the semantic autocorrelation vector of the vehicle damage part image, W 1i and b 1i are the semantic weight matrix and semantic bias vector of the vehicle damage part image respectively, is the semantic modulation vector of the vehicle damage part image, e 1i is the semantic core information anchoring factor of the vehicle damage part image, Sigmoid(·) represents the Sigmoid function, a 1i is the anchor weight of the semantic core information of the vehicle damage part image, n represents the number of row vectors of the semantic autocorrelation matrix of the vehicle damage part image, c1 represents the anchor coding vector of the feature core information of the vehicle damage part image, x 21 ,x 22 ,x 2i ,x 2n The first, second, i-th and n-th row vectors of the maintenance description semantic autocorrelation matrix are respectively represented as the maintenance description semantic autocorrelation vectors, W 2i and b 2i are the maintenance description semantic weight matrix and maintenance description semantic bias vector respectively, The semantic modulation vector for maintenance description, e 2i The anchoring factor of the semantic core information of the maintenance situation description is a 2i is the anchor weight of the semantic core information of the maintenance situation description, m represents the number of row vectors in the semantic autocorrelation matrix of the maintenance situation description, and c2 represents the anchor coding vector of the semantic feature core information of the maintenance situation description.

[0041] Next, the vehicle damage part image feature core information anchor coding vector and the repair description semantic feature core information anchor coding vector are input into the feature granularity response interactive encoder to obtain the vehicle damage-vehicle repair feature granularity response interactive coding vector; and the vehicle damage part image feature core information anchor coding vector and the repair description semantic feature core information anchor coding vector are input into the feature value granularity response interactive encoder to obtain the vehicle damage-vehicle repair feature value granularity response interactive coding vector. Considering that traditional manual review is difficult to accurately capture the implicit association between vehicle damage images and repair instructions, traditional single-dimensional analysis (such as only considering image features or text descriptions) is difficult to fully and accurately reflect the true condition of vehicle damage and its corresponding repair needs. By combining the interactive coding of feature granularity and feature value granularity, the potential information in the data can be more deeply mined, providing more accurate damage assessment results. In the technical solution of this application, the core information anchor code vector of the vehicle damage image features and the core information anchor code vector of the semantic features of the repair description are first input into a feature granularity response interactive encoder to obtain a vehicle damage-vehicle repair feature granularity response interactive code vector. During this process, the system does not simply compare surface information, but instead deconstructs the inherent logic of image features and text descriptions at a micro level by establishing a multi-level semantic mapping relationship. Next, the core information anchor code vector of the vehicle damage image features and the core information anchor code vector of the semantic features of the repair description are input into a feature value granularity response interactive encoder to obtain a vehicle damage-vehicle repair feature value granularity response interactive code vector. In one example, the numerical features of damage edge detection extracted by the image convolutional neural network can be numerically coupled with the numerical features of labor costs and parts prices in the repair description through numerical operations such as element-by-element multiplication and weighted summation. For example, the area feature value of the door dent area will produce a multiplicative effect with the labor time coefficient of the sheet metal repair, while the paint area value will form an exponential relationship with the paint usage parameter. This underlying numerical interaction not only strengthens the quantitative correlation between damage severity and repair cost, but more importantly, by introducing a nonlinear decision boundary through the activation function, it can capture unusual numerical patterns, such as "minor scratches but a full vehicle repaint claim." The synergistic effect of these two granularities enables the system to possess both the abstract ability to understand semantic logic and the precision to process numerical details. For example, in a case where a minor dent in the engine hood requires assembly replacement, the feature granularity interaction identifies the semantic mismatch between the image damage severity and the textual repair plan, while the feature value interaction, through the cost calculation model, identifies deviations from the reasonable range of repair costs. This three-dimensional analysis mechanism significantly enhances the system's ability to identify fraudulent activities such as exaggerating losses and fabricating repair items. It also reduces the manual cross-verification process, which previously took hours, to milliseconds, achieving a qualitative improvement in claims review efficiency while ensuring accurate damage assessments.In a specific example, the vehicle damage part image feature core information anchor coding vector and the repair condition description semantic feature core information anchor coding vector are respectively input into the feature granularity response interaction encoder and the feature value granularity response interaction encoder using the following fine-grained response interaction formula to obtain the vehicle damage-vehicle repair feature granularity response interaction coding vector and the vehicle damage-vehicle repair feature value granularity response interaction coding vector; wherein, the fine-grained response interaction formula is:.

[0042]

[0043] in, For positional addition, W VT and b VT are the feature interaction weight matrix and feature interaction bias vector respectively, tanh(·) is the tanh function, E granular Represents the vehicle damage-vehicle repair feature granularity response interaction coding vector, E value Represents the vehicle damage-vehicle repair feature value granularity response interaction coding vector.

[0044] More specifically, the feature perturbation compensation subunit 3543 is used to perform feature perturbation compensation on the vehicle damage-vehicle repair feature granularity response interaction coding vector and the vehicle damage-vehicle repair feature value granularity response interaction coding vector to obtain a compensated vehicle damage-vehicle repair feature granularity response interaction coding vector and a compensated vehicle damage-vehicle repair feature value granularity response interaction coding vector. It should be understood that the interaction modeling of feature granularity focuses on the spatial topological relationship of the overall feature distribution, while the interaction of feature value granularity focuses on the nonlinear diffusion process of each dimension. The difference in the representation of the two can easily cause dynamic perturbations in the feature manifold interface during fusion. This perturbation manifests as localized distortions in the spatial distribution of the coding vectors after the fusion of multi-granularity information, similar to the stress concentration phenomenon at the joint of different materials, which may disrupt the continuity of key semantics in cross-modal joint analysis. For example, when the image features of a vehicle's front fender scratch interact with the semantic features of the "partial painting" repair instructions, the feature granularity encoding may focus on the matching degree between the damage area and the painting process, while the feature value granularity encoding quantifies the numerical correlation between the paint loss coefficient and the labor cost. If the two are directly superimposed, the interface distortion of the feature space distribution may occur. In order to solve this problem, before fusing the vehicle damage-vehicle repair feature granularity response interaction coding vector and the vehicle damage-vehicle repair feature value granularity response interaction coding vector, it is first necessary to perform feature disturbance compensation on them. In the technical solution of the present application, the vehicle damage-vehicle repair feature granularity response interaction coding vector and the vehicle damage-vehicle repair feature value granularity response interaction coding vector are subjected to feature disturbance compensation to obtain the compensated vehicle damage-vehicle repair feature granularity response interaction coding vector and the compensated vehicle damage-vehicle repair feature value granularity response interaction coding vector. Specifically, the vehicle damage-vehicle repair feature granularity response interaction coding vector and the vehicle damage-vehicle repair feature value granularity response interaction coding vector are subjected to feature- and eigenvalue-based scalability constraint adjustments to adjust the growth exponents of the features and eigenvalues during the diffusion process. Based on this, disturbance compensation is performed on the vehicle damage-vehicle repair feature granularity response interaction coding vector and the vehicle damage-vehicle repair feature value granularity response interaction coding vector to obtain the compensated vehicle damage-vehicle repair feature granularity response interaction coding vector and the compensated vehicle damage-vehicle repair feature value granularity response interaction coding vector. During this process, the system first uses the vehicle damage-vehicle repair feature value granularity response interaction coding vector as a dynamic constraint benchmark and then, through gradient diffusion analysis, deconstructs the interaction strength at the numerical level layer by layer. For example, in the case of engine hood deformation, the system tracks the gradient change between the damage area value and the repair cost coefficient along the eigenvalue dimension to construct a growth index curve. This curve acts as an elastic threshold for the numerical interaction relationship: when the repair cost corresponding to the deformed area exceeds the preset growth exponent, the system automatically identifies it as an abnormal signal.Subsequently, the vehicle damage-vehicle maintenance feature granularity response interaction coding vector is placed under the constraint framework for distribution calibration. In this way, the manifold interface stability of the vehicle damage-vehicle maintenance feature response interaction coding vector can be improved, thereby ensuring that the final vehicle damage-vehicle maintenance semantic response interaction coding vector contains rich detail information while maintaining a high degree of consistency and stability. In a specific example, the vehicle damage-vehicle maintenance feature granularity response interaction coding vector and the vehicle damage-vehicle maintenance feature value granularity response interaction coding vector are subjected to feature perturbation compensation using the following compensation formula to obtain the compensated vehicle damage-vehicle maintenance feature granularity response interaction coding vector and the compensated vehicle damage-vehicle maintenance feature value granularity response interaction coding vector; wherein, the compensation formula is:.

[0045]

[0046] Among them, v a and v b E granular and E value The corresponding eigenvalue, v a ′ and v′ b E granular ′ and E value The corresponding eigenvalue of ′, cos (·) represents the cosine function.

[0047] More specifically, the fusion subunit 3544 is used to fuse the compensated vehicle damage-vehicle maintenance feature granularity response interactive coding vector and the compensated vehicle damage-vehicle maintenance feature value granularity response interactive coding vector to obtain the vehicle damage-vehicle maintenance semantic response interactive coding vector. It should be understood that traditional single-dimensional analysis (such as only considering image features or text descriptions) is difficult to fully and accurately reflect the true condition of vehicle damage and its corresponding maintenance needs. By combining the interactive coding of feature granularity and feature value granularity, the potential information in the data can be mined more deeply to provide more accurate damage assessment results. In a specific example, a cascade method is used to fuse the compensated vehicle damage-vehicle maintenance feature granularity response interactive coding vector and the compensated vehicle damage-vehicle maintenance feature value granularity response interactive coding vector; specifically, the cascade formula is;

[0048] V f =concat[E granular ;E value ]

[0049] Among them, concat[·;·] represents feature fusion, V f Represents the vehicle damage-vehicle repair semantic response interaction encoding vector.

[0050] Specifically, the audit result generation unit 355 is configured to generate the audit result based on the vehicle damage-vehicle repair semantic response interaction code vector. In the technical solution of the present application, the vehicle damage-vehicle repair semantic response interaction code vector is input into a classifier-based intelligent audit module to generate the audit result, which indicates whether there are any anomalies. Traditional claims audits rely primarily on manual judgment, which can easily overlook details and has limited ability to identify fraudulent activities such as false claims or exaggeration of the extent of losses. Here, the classifier is a machine learning model trained on a large number of real-world cases that can automatically determine whether there are anomalies based on the input data. For example, if there is a logical inconsistency between the damage assessment photo and the repair instructions in a case, or if the data does not conform to common sense, the classifier may flag the case as suspicious, prompting further manual review. By introducing a classifier-based intelligent audit module, machine learning algorithms can be used to analyze complex code vectors, thereby more accurately assessing the actual damage condition of the vehicle and its corresponding repair needs. Furthermore, this approach can effectively identify potential anomalies, such as image manipulation and mismatched repair instructions, greatly improving the reliability of the audit.

[0051] In particular, in the technical solution of the present application, the vehicle damage part image feature coding vector and the maintenance condition description semantic understanding coding vector respectively represent the vehicle damage part image semantic coding features and the maintenance condition description semantic coding features. When performing feature cross-modal interaction analysis based on core information attention, the modal difference between the vehicle damage part image feature coding vector and the maintenance condition description semantic understanding coding vector will lead to insufficient long-distance coding representation of cross-modal dynamic interaction, thereby reducing the expression effect of the vehicle damage-vehicle maintenance semantic response interaction coding vector, and affecting the accuracy of the audit results obtained by its input into the classifier-based intelligent audit module.

[0052] In a preferred embodiment, the vehicle damage-vehicle repair semantic response interaction encoding vector is input into a classifier-based intelligent audit module to obtain the audit result, including the following steps:

[0053] Based on the median of the vehicle damage-vehicle repair semantic response interaction coding vector, the vehicle damage-vehicle repair semantic response interaction coding vector is dynamically prioritized to obtain a vehicle damage-vehicle repair semantic response interaction dynamic priority queue feature vector. Wherein, V represents the vehicle damage-vehicle repair semantic response interaction encoding vector, V mid represents the median of the vehicle damage-vehicle repair semantic response interaction encoding vector, represents the subtraction by position, V′ represents the dynamic priority queue feature vector of the vehicle damage-vehicle repair semantic response interaction;

[0054] The vehicle damage-vehicle repair semantic response interaction heterogeneous node collaboration matrix of the vehicle damage-vehicle repair semantic response interaction dynamic priority queue feature vector is calculated and expressed as:

[0055] M i,j =w1v′ i +w2v′ j

[0056] v′ i ,v′ j ∈V′

[0057] Among them, v′ i and v′ j They represent the eigenvalues of the i-th position and the j-th position of the dynamic priority queue feature vector of the vehicle damage-vehicle maintenance semantic response interaction, w1 represents the first weight hyperparameter, w2 represents the second weight hyperparameter, M i,j The eigenvalue of the (i, j) position of the vehicle damage-vehicle repair semantic response interaction heterogeneous node collaboration matrix;

[0058] The conjugate component of the vehicle damage-vehicle repair semantic response interaction encoding vector is mapped to the vehicle damage-vehicle repair semantic response interaction heterogeneous node cooperation matrix to obtain the first vehicle damage-vehicle repair semantic response interaction multi-scale modeling unit vector Where T represents the transpose of the vector, represents matrix multiplication, ⊙ represents positional point multiplication, Sigmoid represents activation function, M represents vehicle damage-vehicle repair semantic response interaction heterogeneous node coordination matrix, and V1 represents the first vehicle damage-vehicle repair semantic response interaction multi-scale modeling unit vector;

[0059] The vehicle damage-vehicle repair semantic response interaction encoding vector is mapped to the vehicle damage-vehicle repair semantic response interaction heterogeneous node collaboration matrix to obtain a second vehicle damage-vehicle repair semantic response interaction multi-scale modeling unit vector Among them, V2 represents the second vehicle damage-vehicle repair semantic response interaction multi-scale modeling unit vector;

[0060] The first vehicle damage-vehicle repair semantic response interactive multi-scale modeling unit vector and the second vehicle damage-vehicle repair semantic response interactive multi-scale modeling unit vector are dynamically evolved to obtain an optimized vehicle damage-vehicle repair semantic response interactive encoding vector V o =(ω1⊙V1⊙V)⊕(ω2⊙V2⊙V), where represents vector addition, ω1 represents the third weight hyperparameter, ω2 represents the fourth weight hyperparameter, V o It represents the optimized vehicle damage-vehicle repair semantic response interaction coding vector. It should be noted that w1=0.6, w1=0.4, ω1=0.7, ω2=0.3 are preset based on experience and can be adjusted according to actual conditions. This embodiment does not specifically limit this.

[0061] Finally, the optimized vehicle damage-vehicle repair semantic response interaction encoding vector is input into a classifier-based intelligent audit module to obtain the audit result.

[0062] In this way, by integrating and analyzing the dynamic priority queue features of the vehicle damage-vehicle repair semantic response interaction encoding vector within a distributed heterogeneous node cluster, and addressing the misalignment of cross-domain feature interactions caused by the long-range dispersion of local density thresholds, a multi-scale modeling module with internal state fusion is employed to deconstruct the complex interaction mechanism of the distributed heterogeneous node cluster. Furthermore, by constructing a multi-scale feature map with cross-order transitions, the vehicle damage-vehicle repair semantic response interaction encoding vector is restructured. Ultimately, an optimized iteration of its deep structural representation is completed within a dynamic state evolution framework, significantly improving the feature generalization capability of the vehicle damage-vehicle repair semantic response interaction encoding vector. This improves the accuracy of the audit results obtained by its input into the classifier-based intelligent audit module.

[0063] As described above, the online damage assessment system 300 based on the write-off mechanism according to the embodiment of the present application can be implemented in various wireless terminals, such as a server having an online damage assessment algorithm based on the write-off mechanism. In one possible implementation, the online damage assessment system 300 based on the write-off mechanism according to the embodiment of the present application can be integrated into the wireless terminal as a software module and / or hardware module. For example, the online damage assessment system 300 based on the write-off mechanism can be a software module in the operating system of the wireless terminal, or it can be an application developed for the wireless terminal; of course, the online damage assessment system 300 based on the write-off mechanism can also be one of the many hardware modules of the wireless terminal.

[0064] Alternatively, in another example, the online damage assessment system 300 based on the write-off mechanism and the wireless terminal may also be separate devices, and the online damage assessment system 300 based on the write-off mechanism may be connected to the wireless terminal through a wired and / or wireless network and transmit interactive information in accordance with an agreed data format.

[0065] While various embodiments of the present disclosure have been described above, the above descriptions are illustrative, non-exhaustive, and not intended to be limiting of the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is selected to best explain the principles of the embodiments, their practical applications, or improvements to existing technologies, or to enable others skilled in the art to understand the embodiments disclosed herein.

[0066] The personal information and other data involved in this application have been obtained with full consent and authorization, and the collection, use and processing of relevant information comply with the relevant laws, regulations and standards of relevant countries and regions.

Claims

1. An online damage assessment system based on a write-off mechanism, characterized in that: include: A damage identification code presentation module is used to present a damage identification code; An authorization confirmation module is used to verify the damage identification code after scanning the damage identification code and pop up an authorization confirmation page for the vehicle disassembly authorization agreement; A window prompt module is used to determine whether the party responsible for the accident has purchased "commercial insurance" in response to receiving the consent action, and if the party responsible for the accident has not purchased "commercial insurance", a risk prompt window is popped up; If the party responsible for the accident has purchased "commercial insurance", the vehicle information registration page will pop up; The damage assessment module is used to receive damage assessment photos and repair status descriptions uploaded by users; The review module is used to send the damage assessment photos and repair situation description to the insurance company for review to obtain the review results.

2. The online damage assessment system based on the write-off mechanism according to claim 1 is characterized in that: The audit module includes: An image quality verification unit, configured to perform image quality verification on the damage assessment image to obtain an image quality verification result; an image processing unit configured to, if the image quality check result indicates that the image quality does not meet the requirements, pop up a prompt to re-upload the damage assessment image; and, if the image quality check result indicates that the image quality meets the requirements, extract image features of the damaged part of the vehicle from the damage assessment image to obtain an image feature coding vector of the damaged part of the vehicle; A maintenance situation semantic understanding unit, configured to perform text semantic understanding on the maintenance situation description to obtain a maintenance situation description semantic understanding encoding vector; a vehicle damage-vehicle maintenance cross-modal joint analysis unit, configured to perform a vehicle damage-vehicle maintenance cross-modal joint analysis based on core information anchoring on the vehicle damage part image feature coding vector and the maintenance description semantic understanding coding vector to obtain a vehicle damage-vehicle maintenance semantic response interaction coding vector; An audit result generating unit is used to obtain the audit result based on the vehicle damage-vehicle repair semantic response interaction coding vector.

3. The online damage assessment system based on the write-off mechanism according to claim 2 is characterized in that: The image processing unit is configured to: If the image quality verification result shows that the image quality meets the requirements, the damage assessment image is input into a vehicle damaged part feature extractor based on a DenseNet network to obtain a vehicle damaged part image feature coding vector.

4. The online damage assessment system based on the write-off mechanism according to claim 3 is characterized in that: The maintenance situation semantic understanding unit is used to: The maintenance situation description is input into a maintenance situation semantic encoder based on the RoBERTa model to obtain a semantic understanding encoding vector of the maintenance situation description.

5. The online damage assessment system based on the write-off mechanism according to claim 4 is characterized in that: The vehicle damage-vehicle repair cross-modal joint analysis unit includes: A semantic autocorrelation association matrix construction subunit is used to construct a vehicle condition semantic autocorrelation association matrix of the vehicle damage part image feature coding vector and the maintenance condition description semantic understanding coding vector to obtain a vehicle damage part image semantic autocorrelation association matrix and a maintenance condition description semantic autocorrelation association matrix; A vehicle damage-vehicle repair fine-grained response interaction subunit is configured to perform fine-grained response interaction based on vehicle damage-repair situation anchoring on the semantic autocorrelation association matrix of the vehicle damage part image and the semantic autocorrelation association matrix of the repair situation description to obtain a vehicle damage-vehicle repair feature granularity response interaction coding vector and a vehicle damage-vehicle repair feature value granularity response interaction coding vector; a characteristic disturbance compensation subunit, configured to perform characteristic disturbance compensation on the vehicle damage-vehicle maintenance characteristic granularity response interaction coding vector and the vehicle damage-vehicle maintenance characteristic value granularity response interaction coding vector to obtain a compensated vehicle damage-vehicle maintenance characteristic granularity response interaction coding vector and a compensated vehicle damage-vehicle maintenance characteristic value granularity response interaction coding vector; A fusion subunit is used to fuse the compensated vehicle damage-vehicle maintenance feature granularity response interaction coding vector and the compensated vehicle damage-vehicle maintenance feature value granularity response interaction coding vector to obtain the vehicle damage-vehicle maintenance semantic response interaction coding vector.

6. The online damage assessment system based on the write-off mechanism according to claim 5 is characterized in that: The vehicle damage-vehicle repair fine-grained response interaction sub-unit is used to: Inputting the semantic autocorrelation matrix of the vehicle damage part image and the semantic autocorrelation matrix of the maintenance condition description into a vehicle damage-repair condition anchoring network based on autocorrelation decoupling to obtain an anchor coding vector of the core information of the vehicle damage part image feature and an anchor coding vector of the core information of the semantic feature of the maintenance condition description; Inputting the vehicle damage part image feature core information anchor coding vector and the repair condition description semantic feature core information anchor coding vector into a feature granularity response interactive encoder to obtain the vehicle damage-vehicle repair feature granularity response interactive coding vector; The vehicle damage part image feature core information anchor coding vector and the repair condition description semantic feature core information anchor coding vector are input into the feature value granularity response interactive encoder to obtain the vehicle damage-vehicle repair feature value granularity response interactive coding vector.

7. The online damage assessment system based on the write-off mechanism according to claim 6 is characterized in that: The characteristic disturbance compensation subunit is used to: The vehicle damage-vehicle maintenance feature granularity response interaction coding vector and the vehicle damage-vehicle maintenance feature value granularity response interaction coding vector are respectively subjected to scalability constraint adjustment based on features and feature values to adjust the growth exponents of features and feature values in the diffusion process, and based on this, the vehicle damage-vehicle maintenance feature granularity response interaction coding vector and the vehicle damage-vehicle maintenance feature value granularity response interaction coding vector are subjected to disturbance compensation to obtain the compensated vehicle damage-vehicle maintenance feature granularity response interaction coding vector and the compensated vehicle damage-vehicle maintenance feature value granularity response interaction coding vector.

8. The online damage assessment system based on the write-off mechanism according to claim 7 is characterized in that: The audit result generating unit is used to: The vehicle damage-vehicle repair semantic response interaction coding vector is input into a classifier-based intelligent audit module to obtain the audit result, and the audit result is used to indicate whether there is an abnormality.

Images