Automatic loss assessment agent combining vision and language model

Abstract

The invention discloses an automatic loss assessment agent combining visual and language models, and belongs to technical innovation of artificial intelligence in the field of insurance claim settlement. According to the intelligent agent, automatic processing of vehicle insurance loss assessment is realized by constructing a vehicle loss picture recognition pre-training model and an intelligent agent workflow system. The system comprises a multi-modal data input interface (supporting synchronous uploading of pictures and tables) and an intelligent classification processing module (text information is recognized by adopting a large language model; the image is processed through a loop body), the pre-training model calling module is used for circularly judging the vehicle damage image and obtaining feedback, the accessory database docking module is used for retrieving related cost items, and the intelligent report generation module is used for generating a damage assessment list in combination with report information, a loss part and accessory data. The technical effects of the method are obvious: the processing speed is high, the loss assessment accuracy is stable (not influenced by human factors), and the industrial pain points of non-uniform traditional loss assessment standards, low efficiency, manual data transfer and the like are effectively solved. The system is especially suitable for a vehicle insurance claim settlement scene, and can greatly improve the loss assessment efficiency and accuracy.

Description

Technical Field

[0001] This invention relates to an innovative application of artificial intelligence in the insurance field, specifically involving an intelligent vehicle damage assessment system and method based on multimodal data fusion, which is particularly suitable for automated damage assessment in auto insurance claims scenarios. Background Technology

[0002] The current auto insurance claims settlement industry faces three major pain points: 1. Inconsistent damage assessment standards result in a 10-20% difference in the assessment of the same damage by different claims adjusters. 2. Traditional damage assessment relies on manual inspection and assessment, which is inefficient. 3. Some steps in the insurance company's system processes require manual data transfer. Summary of the Invention

[0003] Technical solutions and implementation methods: The core of this invention lies in constructing a pre-trained model for vehicle damage image recognition and an intelligent agent (workflow) for vehicle insurance damage assessment, including: 1. Pre-trained model for vehicle damage image recognition. See the instruction manual for details. Figure 1 . 2. Overview of the intelligent agent (workflow) for auto insurance damage assessment. See the instruction manual for details. Figure 2 . 3. Intelligent agent dialog box, used for inputting multimodal data (supports simultaneous uploading of images / tables). See the instruction manual for details. Figure 3 . 4. Classify and process multimodal data (text information is recognized using a large language model, and image files are input into the loop). See the instruction manual for details. Figure 4 . 5. Iterate through the pre-trained model to judge the vehicle damage images and receive feedback results. See the instruction manual for details. Figure 5 . 6. Connect the top-ranked accessory item in the model feedback results to the accessory database to retrieve the relevant cost items and receive the feedback results. See the instruction manual for details. Figure 6 . 7. Based on the reported information, the damaged parts, and the relevant parts information, a damage assessment report is generated using a language model. See the instruction manual for details. Figure 7 . Technical effects:

[0004] 1. Improved efficiency: Processing time is significantly reduced across all functional modules. For example, when integrating with a pre-trained image recognition model, processing a single image takes only 0.45 seconds, and querying a single accessory takes only 0.3 seconds. 2. Stable accuracy: Does not accept the impact of differences in claims adjusters' experience. Attached Figure Description

[0005] Figure 1 The information interface of the vehicle damage image recognition model trained in the Baidu Smart Cloud Platform. Figure 2 Overview of the intelligent agent (workflow) for auto insurance damage assessment. Figure 3 The intelligent agent dialog box is used to upload multimodal data such as police report information and vehicle damage photos (supports simultaneous uploading of images / tables). Figure 4 The input text information is recognized using a large language model, and the vehicle damage image file is input into the loop body for recognition and judgment processing using a pre-trained model. Figure 5 The program iteratively calls the pre-trained model to assess vehicle damage images and receives feedback results. Figure 6 Connect the top-ranked accessory item in the model feedback results to the accessory database to retrieve the relevant cost items and receive the feedback results. Figure 7 Based on the reported information, the damaged parts, and the relevant parts information, a damage assessment report is generated using a language model.

Claims

1. A vehicle damage assessment system based on a cloud-based intelligent model, characterized in that, include: The data input interface is used to receive vehicle damage photos and structured accident report information forms uploaded by users; Workflow coordination engine, configured as follows: a) Parse the accident description text data in the accident report information form, including the accident report number, time of the accident, driver, and location; b) Call the pre-trained Baidu Smart Cloud vehicle damage recognition model API, submit vehicle damage images and obtain the recognition results (damaged parts and repair solutions (replacement or repair)) from the model. c) Based on the recognition results fed back by the model, query the parts cost database to obtain the amount, standard labor cost and paint cost of the corresponding parts; The report generation module integrates the reported information, vehicle damage image recognition results, and parts cost data to generate a damage assessment report that includes repair plans and cost details.

2. The system according to claim 1, characterized in that, The workflow coordination engine includes: Parallel processing unit, simultaneously performs table parsing, image recognition API calls, and parts database queries; The data association module maps image recognition results to accessory data.

3. The system according to claim 1, characterized in that, The report generation module implements an automatic cost totaling function, calculating the total amount of parts costs, labor costs, and paint costs.

4. A vehicle intelligent damage assessment method, characterized in that, Including the following steps: S1. Receive vehicle damage photos and electronic accident report forms through the smart agent interface; S2. Workflow engine executes in parallel: a) Parse the table to obtain vehicle information and accident description; b) Use the cloud-based vehicle damage recognition model to obtain the recognition results of the image; c) Query the parts database to obtain repair cost data; S3. Establish a correlation matrix between accident description, damaged parts, and spare parts data; S4. Automatically generate a damage assessment report that includes a repair plan, a parts list, and a total cost.

5. A computer-readable storage medium storing a computer program that performs the method of any one of claims 4. Its characteristics are: 1) A unified data processing architecture integrating three streams (image recognition + table parsing + parts query) 2) Parallel processing and intelligent association mechanism of workflow engine.

Images