Claims settlement method, device, computer equipment and storage medium based on artificial intelligence
By using artificial intelligence for identity verification and vehicle damage image analysis, auto insurance claims are processed automatically, solving the problems of low efficiency and difficulty in fraud detection in existing claims processes, and achieving a fast and accurate claims process.
Patent Information
- Application Number
- CN202111531893.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2021-12-14
- Publication Date
- 2024-11-15
- Estimated Expiration
- 2041-12-14
AI Technical Summary
Existing auto insurance claims processes are time-consuming, labor-intensive, inefficient, and make it difficult to identify claims fraud.
The system employs an AI-based claims process, which uses identity verification, vehicle damage image prediction, and recognition models to automatically determine the probability of claims and the location of damage, generating vehicle damage assessment results.
It improves claims processing efficiency, reduces waste of human resources, effectively identifies insurance fraud, reduces loss assessment errors, and enhances customer service levels.
Smart Images

Figure CN114219661B_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of artificial intelligence technology, and specifically to an artificial intelligence-based claims settlement method, apparatus, computer equipment and storage medium. Background Art
[0002] With the development of science and technology and economy, people's living standards are improving day by day, and more and more families own cars, which also drives the rapid development of vehicle insurance. Due to the huge number of cars in China's car market, various traffic accidents occur on the road every day. At present, the claims of car accidents are mainly settled by the insurance company sending surveyors to conduct on-site surveys and settle claims. This claim settlement method is not only time-consuming and labor-intensive, but also wastes a lot of human resources, has low claim settlement efficiency, and cannot effectively identify some claim fraud. Summary of the invention
[0003] The main purpose of this application is to provide a claims settlement method, device, computer equipment and storage medium based on artificial intelligence, aiming to solve the technical problems that the existing claims settlement method is not only time-consuming and labor-intensive but also wastes a lot of human resources, has low claims settlement efficiency, and cannot effectively identify some claims fraud.
[0004] This application proposes a claims settlement method based on artificial intelligence, which includes the following steps:
[0005] Determine whether a claim request triggered by a user through a client is received; wherein the claim request carries user information;
[0006] If the claim request is received, the user information, the user's facial movement information and the user's handwriting are obtained, identity verification of the user is performed based on a preset claim user whitelist, standard facial movement information and standard handwriting, and whether the identity verification is passed;
[0007] If the identity verification is passed, the user is reminded to upload the vehicle damage picture based on the preset reminder information;
[0008] Inputting the vehicle damage picture into a preset target prediction model for prediction, and outputting the claim probability corresponding to the vehicle damage picture through the target prediction model;
[0009] Determining whether the claim settlement probability is greater than a preset probability threshold;
[0010] If it is greater than the probability threshold, calling a preset recognition model to perform recognition processing on the vehicle damage picture, and generating target loss part information and target loss degree information corresponding to the vehicle damage picture;
[0011] A vehicle damage assessment result corresponding to the vehicle damage picture is generated based on the target loss location information and the target loss degree information.
[0012] Optionally, the step of obtaining the user information, the user's facial movement information and the user's handwriting, performing identity authentication on the user based on a preset claims user whitelist, standard facial movement information and standard handwriting, and determining whether the identity authentication is passed includes:
[0013] Obtain a preset claims user whitelist, and determine whether there is designated user information identical to the user information in the claims user whitelist;
[0014] If the designated user information exists, displaying preset action input reminder information to remind the user to input corresponding facial action information based on the action input reminder information;
[0015] Collecting the facial action information input by the user, and acquiring standard facial action information corresponding to the designated user information; wherein the facial action information includes a facial action sequence, and the facial action sequence includes a plurality of facial actions executed in sequence;
[0016] Determining whether each facial action included in the facial action information completely matches the facial action at the corresponding position in the standard facial action information;
[0017] If all are matched, display the preset handwriting input reminder information; wherein the handwriting input reminder information includes the standard characters corresponding to the specified user information;
[0018] collecting handwriting input by the user according to the standard characters;
[0019] Extracting first handwriting feature data of the handwriting, and extracting second handwriting feature data of the standard handwriting corresponding to the standard character;
[0020] Determining whether the first handwriting feature data matches the second handwriting feature data;
[0021] If they match, the identity authentication is determined to be successful; otherwise, the identity authentication is determined to be unsuccessful.
[0022] Optionally, before the step of inputting the vehicle damage picture into a preset target prediction model for prediction, the step includes:
[0023] Acquire multiple prediction models generated by pre-training; wherein each of the prediction models is generated by training a preset convolutional neural network model according to a respective different training sample set, each of the training sample sets includes a number of training pictures and labels corresponding to the training pictures, and the number of the prediction models is equal to the number of the training sample sets;
[0024] Get the preset verification sample set;
[0025] Generate a prediction processing time for each prediction model based on the validation sample set;
[0026] Based on the verification sample set, generating evaluation scores for each of the prediction models according to preset calculation rules;
[0027] Acquire a first weight corresponding to the predicted processing time, and acquire a second weight corresponding to the evaluation score;
[0028] Based on the prediction processing time, the evaluation score, the first weight, and the second weight, calling a first preset formula to calculate a model score of each of the prediction models;
[0029] Selecting a prediction model with the highest model score from all the prediction models;
[0030] The prediction model with the highest model score is used as the target prediction model.
[0031] Optionally, the step of generating the evaluation score of each prediction model according to a preset calculation rule based on the verification sample set includes:
[0032] Input the verification sample set into a specified prediction model, and calculate the accuracy of the specified prediction model; wherein the specified prediction model is any one of all the prediction models;
[0033] Inputting the verification sample set into a specified prediction model, and calculating the recall rate of the specified prediction model;
[0034] Get the preset value;
[0035] Based on the precision, the recall and the preset value, a second preset formula is called to calculate the evaluation score of the specified prediction model.
[0036] Optionally, the step of generating a vehicle damage assessment result corresponding to the vehicle damage picture based on the target loss part information and the target loss degree information includes:
[0037] Displaying a preset repair shop list, and receiving a target repair shop selected by the user from the repair shop list;
[0038] Obtaining the prices of auto parts and labor hours of the target repair shop;
[0039] Obtaining vehicle information corresponding to the vehicle damage picture; wherein the vehicle information includes vehicle brand, car series, and vehicle model;
[0040] The corresponding damage amount is calculated based on the target loss part information, the target loss degree information, the vehicle information, the auto parts price and the labor time price;
[0041] The damage assessment amount is used as the vehicle damage assessment result.
[0042] Optionally, before the step of calling a preset recognition model to perform recognition processing on the vehicle damage picture to generate target loss part information and target loss degree information corresponding to the vehicle damage picture, the step includes:
[0043] Collecting a preset number of vehicle damage sample data, and performing a preset process on the vehicle damage sample data to obtain processed vehicle damage sample data;
[0044] Determining loss location information and loss degree information corresponding to the processed vehicle damage sample data;
[0045] The processed vehicle damage sample data is used as the input of a preset classification model, and the loss location information and the loss degree information are used as the output of the classification model for training to obtain an initial classification model;
[0046] Acquire a preset test sample set, and perform an accuracy test on the initial classification model based on the test sample set to obtain a corresponding classification accuracy rate;
[0047] Determining whether the classification accuracy is greater than a preset accuracy threshold;
[0048] If so, the initial classification model is used as the recognition model.
[0049] Optionally, before the step of inputting the vehicle damage picture into a preset target prediction model for prediction, the step includes:
[0050] Generate a blur degree value corresponding to the vehicle damage image based on a preset algorithm;
[0051] Get the preset blur threshold;
[0052] Determining whether the blur level value is less than the blur threshold;
[0053] If it is less than the fuzzy threshold, a processing instruction is generated to input the vehicle damage picture into a preset target prediction model for prediction;
[0054] If it is greater than the blur threshold, a preset shooting reminder message is displayed to remind the user to re-shoot and upload a vehicle damage photo that meets the requirements.
[0055] The present application also provides an artificial intelligence-based claims settlement device, comprising:
[0056] A first judgment module is used to judge whether a claim request triggered by a user through a client is received; wherein the claim request carries user information;
[0057] a verification module, for obtaining the user information, the user's facial movement information and the user's handwriting upon receiving the claim request, performing identity verification on the user based on a preset claim user whitelist, standard facial movement information and standard handwriting, and determining whether the identity verification is passed;
[0058] A reminder module, used to remind the user to upload a car damage picture based on preset reminder information if the identity authentication is passed;
[0059] A first generating module is used to input the vehicle damage picture into a preset target prediction model for prediction, and output the claim probability corresponding to the vehicle damage picture through the target prediction model;
[0060] A second judgment module is used to judge whether the claim settlement probability is greater than a preset probability threshold;
[0061] The second generation module is used to call a preset recognition model to perform recognition processing on the vehicle damage picture if the probability is greater than the probability threshold, and generate target loss part information and target loss degree information corresponding to the vehicle damage picture;
[0062] The third generating module is used to generate a vehicle damage assessment result corresponding to the vehicle damage picture based on the target loss part information and the target loss degree information.
[0063] The present application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the above method when executing the computer program.
[0064] The present application also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program implements the steps of the above method when executed by a processor.
[0065] The claims settlement method, apparatus, computer device and storage medium based on artificial intelligence provided in this application have the following beneficial effects:
[0066] The claims method, device, computer equipment and storage medium based on artificial intelligence provided in the present application will first authenticate the user after receiving the claims request triggered by the user through the client. If the identity authentication is passed, the user will be reminded to upload the vehicle damage picture based on the reminder information, and then the vehicle damage picture will be input into the preset target prediction model for prediction to obtain the corresponding claim probability. If the claim probability is greater than the preset probability threshold, the preset recognition model will be called to identify the vehicle damage picture, generate the target loss part information and target loss degree information corresponding to the vehicle damage picture, and finally generate and display the vehicle damage assessment result based on the obtained target loss part information and the target loss degree information. This embodiment can effectively identify insurance fraud and avoid the phenomenon of a small number of users defrauding insurance by authenticating the user and calculating the claim probability for the vehicle damage picture. In addition, when it is determined that there is no insurance fraud, it can quickly and efficiently automatically identify the vehicle's damaged parts and degree of loss based on the vehicle damage pictures provided by the user, and then automatically generate the vehicle damage assessment results based on the obtained damage parts and degree of loss to realize automated claims. The claims processing operation is simple, and there is no need for surveyors to participate in the damage assessment. The claims process is fast, which effectively improves the intelligence level of damage assessment and claims efficiency, reduces case damage assessment deviations, and improves customer service levels. BRIEF DESCRIPTION OF THE DRAWINGS
[0067] Figure 1 It is a flowchart of a claims settlement method based on artificial intelligence according to an embodiment of the present application;
[0068] Figure 2 It is a structural schematic diagram of an artificial intelligence-based claims settlement device according to an embodiment of the present application;
[0069] Figure 3 It is a schematic diagram of the structure of a computer device according to an embodiment of the present application.
[0070] The realization of the purpose, functional features and advantages of this application will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION
[0071] It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application. It will be understood by those skilled in the art that, unless otherwise defined, all terms (including technical terms and scientific terms) used herein have the same meaning as the general understanding of those skilled in the art in the field to which the present invention belongs. It should also be understood that terms such as those defined in general dictionaries should be understood to have meanings consistent with the meanings in the context of the prior art, and unless specifically defined as here, will not be interpreted with idealized or overly formal meanings.
[0072] The embodiments of the present application can acquire and process relevant data based on artificial intelligence technology. Among them, artificial intelligence (AI) is the theory, method, technology and application system that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results.
[0073] AI basic technologies generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing technology, operation / interaction systems, mechatronics, etc. AI software technologies mainly include computer vision technology, robotics technology, biometrics technology, speech processing technology, natural language processing technology, and machine learning / deep learning.
[0074] Reference Figure 1 , the claims settlement method based on artificial intelligence in one embodiment of the present application includes:
[0075] S10: Determine whether a claim request triggered by a user through a client is received; wherein the claim request carries user information;
[0076] S20: if the claim request is received, obtaining the user information, the user's facial movement information and the user's handwriting, performing identity authentication on the user based on a preset claim user whitelist, standard facial movement information and standard handwriting, and determining whether the identity authentication is passed;
[0077] S30: If the identity authentication is passed, remind the user to upload the vehicle damage picture based on the preset reminder information;
[0078] S40: inputting the vehicle damage picture into a preset target prediction model for prediction, and outputting a claim probability corresponding to the vehicle damage picture through the target prediction model;
[0079] S50: Determine whether the claim settlement probability is greater than a preset probability threshold;
[0080] S60: If the probability is greater than the probability threshold, calling a preset recognition model to perform recognition processing on the vehicle damage picture, and generating target damage location information and target damage degree information corresponding to the vehicle damage picture;
[0081] S70: Generate a vehicle damage assessment result corresponding to the vehicle damage picture based on the target loss location information and the target loss degree information.
[0082] As described in the above steps S10 to S70, the execution subject of the embodiment of the method is a claim settlement device based on artificial intelligence, which can be specifically a self-service claim settlement system. In practical applications, the above-mentioned claim settlement device based on artificial intelligence can be implemented by a virtual device, such as software code, or by a physical device with relevant execution code written or integrated, and can interact with the user through a keyboard, mouse, remote control, touchpad or voice control device. The claim settlement device based on artificial intelligence in this embodiment can quickly and efficiently realize automated claims settlement according to the vehicle damage pictures provided by the user. The claim settlement operation is simple, and there is no need for surveyors to participate in the claim settlement. The claim settlement process is fast, which effectively improves the intelligence and efficiency of the claim settlement and improves the customer service level. Specifically, first determine whether a claim settlement request triggered by the user through the client is received. Among them, the client enters the self-service claim settlement system by downloading the APP application, and the self-service claim settlement function is turned on. The client sends a claim settlement request through the APP. In addition, the claim settlement request carries user information. The user information is used to identify the identity of the insured. The user information may include name information, etc. If the claim request is received, the user information, the user's facial motion information and the user's handwriting are obtained, the user's identity is authenticated based on the preset claim user whitelist, standard facial motion information and standard handwriting, and it is determined whether the identity authentication is passed. The specific implementation process of authenticating the user based on the user information, calling the preset claim user whitelist, standard facial motion information and standard handwriting will be further described in the subsequent specific embodiments, which will not be repeated here.
[0083] If the identity authentication is passed, the user is reminded to upload a damaged vehicle picture based on the preset reminder information. The reminder information may be information for prompting the user to use the client to take a damaged vehicle picture at a specified angle and upload the damaged vehicle picture, so that the damaged vehicle picture taken by the user includes the damaged position of the vehicle, that is, the picture including the damaged position of the vehicle is used as the damaged vehicle picture. For example, if the rear bumper of the vehicle is damaged, the main view, left view, and right view including the rear bumper of the vehicle can be used as the corresponding damaged vehicle picture. For example, the process of the user taking a damaged vehicle picture through the client can be: first enter the pop-up window to prompt the shooting schematic diagram, and each time you enter, check whether the sound permission is turned on. If it is turned on, the voice prompt operation: please align the box with the damaged part of the vehicle. During the process, it can also be prompted: the distance is too far, the accessories / non-car parts are not detected, please confirm / the distance is too close, please adjust. Then the damaged vehicle picture can be generated by picture shooting or video shooting. During the shooting process, if it is a picture shooting model, it is prompted to select the damaged part first. Display the 3D model, click the button to close it, rotate the model, match the model according to the real-life vehicle model, select the damaged area and confirm to close the model window, guide you to take a picture of the damaged part as the damaged car picture, prompt you to save the picture and return to the upload page. If it is a video shooting mode, guide the user to shoot and record a video of the damaged area, and then use the preset SDK to grab pictures that can be used for damage assessment as damaged car pictures. After the video shooting is completed, the damaged car picture will be returned to the upload page.
[0084] Then, the vehicle damage pictures uploaded by the user are received, and the vehicle damage pictures are input into the preset target prediction model for prediction, and the target prediction model is used to output the claim probability corresponding to the vehicle damage pictures. Among them, the preset convolutional neural network model can be trained in advance according to different training samples to generate multiple prediction models, and then the model with higher evaluation scores and faster processing efficiency is selected from all prediction models as the target prediction model, so that the target prediction model can be used to predict the vehicle damage pictures uploaded by the user in the future, which can effectively improve the data accuracy of the claim probability corresponding to the vehicle damage pictures, and improve the analysis and processing efficiency of the vehicle damage pictures, which is conducive to the subsequent effective identification of insurance fraud based on the claim probability generated by the target prediction model. In addition, the process of inputting the vehicle damage picture into the target prediction model for prediction, and outputting the claim probability corresponding to the vehicle damage picture through the target prediction model may include: inputting the vehicle damage picture into the convolution layer in the target prediction model for feature extraction to obtain the first feature data; inputting the first feature data into the pooling layer in the target prediction model for dimensionality reduction processing to obtain the second feature data; inputting the second feature data into the fully connected layer in the target prediction model for mapping to obtain a one-dimensional feature vector; inputting the one-dimensional feature vector into the classifier for classification to obtain the claim probability of the vehicle damage picture. Among them, a classifier SVM, such as a support vector machine, can be used for classification, which is a binary classification model for binary classification of vehicle damage pictures, one category is classified as belonging to claim, and the other category is classified as not belonging to claim. After obtaining the one-dimensional feature vector, the one-dimensional feature vector is input into the classifier, and the classifier maps the one-dimensional feature vector to a value in the range of 0 to 1, which is the claim probability of the vehicle damage picture.
[0085] In addition, each of the prediction models is generated by training a preset convolutional neural network model according to different training sample sets, and each of the training sample sets includes a number of training pictures and labels corresponding to the training pictures. The number of prediction models is equal to the number of training sample sets. Claim pictures can be collected from a database storing auto insurance claim information as training pictures. After collecting the training pictures, it is also necessary to construct training samples for model input, by annotating the training pictures and constructing training samples according to the training pictures and the labels corresponding to the training pictures, for example, (image1, 1), (image2, 0). Since a supervised learning model is used, supervised learning refers to using examples with known correct answers to train the network, and the labels are the desired results. After collecting the training images, you can use Labellmg to annotate the training images one by one. Labellmg is an image annotation tool. Among all the training images, some of the training images belong to claims, and the other part of the images do not meet the claims standards. The training images that belong to claims are marked, and the training images that do not belong to claims are not marked. The training images that belong to claims are marked as 1, and the training images that do not belong to claims are marked as 0, until all the training images are annotated. Then, according to the pre-entered number, the training samples are randomly allocated into multiple corresponding training sample sets. For example, each training sample set selects a certain proportion of sample data from the training sample to generate. It only needs to ensure that each training sample set is a different set, and the training sample set is divided into a training set and a test set again. The training set is used to train the model, and the test set is used for subsequent testing of the accuracy of the trained model. Specifically, the training generation process of the prediction model may include: first, all weights in the convolutional neural network are initialized, that is, set to a small random value close to 0, and then the training image is input into the network and sequentially passes through the convolution layer, pooling layer and fully connected layer, wherein there are multiple convolution layers and pooling layers, and the output value is finally obtained by forward propagation. Then the output value is compared with the target value, and the target value refers to the annotation corresponding to the training image in the training sample set, and the total error between the output value and the target value is calculated. Since the training image passes through the convolution layer, pooling layer and fully connected layer from the input layer to the output layer, and the data is inevitably lost in the process of transmitting between the layers, it also leads to the generation of errors, and the error value caused by each layer is different, so after the total error of the convolutional neural network is calculated, the error needs to be back-propagated to the convolutional neural network to find out how much weight each layer should bear for the total error.Therefore, we perform gradient descent at the end, return the error layer by layer, calculate the error of each layer, and then update the weights, that is, back-propagate the total error, calculate the error of the fully connected layer based on the total error, calculate the error of the pooling layer based on the error of the fully connected layer, calculate the error of the convolution layer based on the error of the pooling layer, and update the weights of each layer based on the error of each layer. All training samples in the training set are input into the convolutional neural network model for training to obtain a trained convolutional neural network model, and then the training samples of the test set are input into the trained convolutional neural network model to test the accuracy of the model. If the accuracy of the trained convolutional neural network model meets the preset requirements, it means that the trained model has been trained and is used as the final prediction model. If the accuracy of the trained convolutional neural network model does not meet the preset requirements, it is necessary to increase training samples and continue training.
[0086] It is then determined whether the claim probability is greater than a preset probability threshold. The value of the probability threshold is not limited and can be set according to actual usage requirements, for example, it can be set to 0.8. After obtaining the claim probability of the vehicle damage picture, the claim probability is compared with the probability threshold. When the claim probability corresponding to the vehicle damage picture is greater than the probability threshold, it means that the vehicle damage picture is very close to the vehicle claim picture that has been claimed, and the claim picture is determined to be claimable. When the claim probability corresponding to the vehicle damage picture is less than the probability threshold, it means that the vehicle damage picture has the risk of insurance fraud or fraud, and the vehicle damage picture is determined not to be claimable. For example, when the claim probability is 0.9, then the claim probability is greater than the probability threshold, and the vehicle damage picture is determined to be claimable. Only when the claim probability is greater than the probability threshold, that is, the vehicle damage picture uploaded by the user is claimable, will further damage assessment be carried out later. Otherwise, it will be determined that the claim case corresponding to the current vehicle damage picture has insurance fraud or fraud, and the claim processing process will be terminated directly, thereby effectively improving the intelligence and accuracy of claim case processing.
[0087] If it is greater than the probability threshold, the preset recognition model is called to perform recognition processing on the vehicle damage picture to generate target loss part information and target loss degree information corresponding to the vehicle damage picture. Finally, the vehicle damage assessment result corresponding to the vehicle damage picture is generated based on the target loss part information and the target loss degree information. Among them, the specific implementation process of generating the vehicle damage assessment result corresponding to the vehicle damage picture based on the target loss part information and the target loss degree information will be further described in the subsequent specific embodiments, which will not be repeated here.
[0088] In this embodiment, after receiving a claim request triggered by a user through a client, the user will first be authenticated. If the identity authentication is passed, the user will be reminded to upload a car damage picture based on the reminder information, and then the car damage picture will be input into a preset target prediction model for prediction to obtain a corresponding claim probability. If the claim probability is greater than a preset probability threshold, the preset recognition model will be called to perform recognition processing on the car damage picture, and target loss part information and target loss degree information corresponding to the car damage picture will be generated. Finally, the vehicle damage assessment result will be generated and displayed based on the obtained target loss part information and the target loss degree information. This embodiment can effectively identify insurance fraud and avoid the phenomenon of a small number of users defrauding insurance by authenticating the user and calculating the claim probability for the car damage picture. In addition, when it is determined that there is no insurance fraud, it can quickly and efficiently automatically identify the vehicle's damaged parts and degree of loss based on the vehicle damage pictures provided by the user, and then automatically generate the vehicle damage assessment results based on the obtained damage parts and degree of loss to realize automated claims. The claims processing operation is simple, and there is no need for surveyors to participate in the damage assessment. The claims process is fast, which effectively improves the intelligence level of damage assessment and claims efficiency, reduces case damage assessment deviations, and improves customer service levels.
[0089] Furthermore, in one embodiment of the present application, the above step S20 includes:
[0090] S200: Obtain a preset claims user whitelist, and determine whether there is designated user information identical to the user information in the claims user whitelist;
[0091] S201: If the designated user information exists, displaying preset motion input reminder information to remind the user to input corresponding facial motion information based on the motion input reminder information;
[0092] S202: collecting the facial action information input by the user, and acquiring standard facial action information corresponding to the designated user information; wherein the facial action information includes a facial action sequence, and the facial action sequence includes a plurality of facial actions executed in sequence;
[0093] S203: Determine whether each facial action included in the facial action information completely matches the facial action at the corresponding position in the standard facial action information;
[0094] S204: If all are matched, displaying preset handwriting input reminder information; wherein the handwriting input reminder information includes the standard character corresponding to the specified user information;
[0095] S205: collecting the handwriting input by the user according to the standard character;
[0096] S206: extracting first handwriting feature data of the handwriting, and extracting second handwriting feature data of the standard handwriting corresponding to the standard character;
[0097] S207: Determine whether the first handwriting feature data matches the second handwriting feature data;
[0098] S208: If they match, it is determined that the identity authentication has passed; otherwise, it is determined that the identity authentication has not passed.
[0099] As described in the above steps S200 to S208, the steps of obtaining the user information, the user's facial action information and the user's handwriting, performing identity authentication on the user based on the preset claim user whitelist, standard facial action information and standard handwriting, and judging whether the identity authentication is passed may specifically include: first obtaining the preset claim user whitelist, and judging whether there is designated user information identical to the user information in the claim user whitelist. Wherein, the preset claim user whitelist is a list data pre-created to store legal user information. If the designated user information exists, the preset action input reminder information is displayed to remind the user to input the corresponding facial action information based on the action input reminder information. Wherein, in order to ensure that the user inputs the facial action information, the action input reminder information may be provided and displayed in advance to prompt the user to input the corresponding facial action information, such as blinking, sticking out the tongue, pinching the nose, covering the ears, etc., so that the facial action information is subsequently compared with the pre-stored standard facial action information corresponding to the designated user information to obtain the identity comparison result corresponding to the user. Then, the facial action information input by the user is collected, and the standard facial action information corresponding to the designated user information is obtained. The facial action information includes a facial action sequence, and the facial action sequence includes multiple facial actions performed in sequence. The facial action information is an image sequence including the user's facial actions, and the user's facial action information can be obtained through a camera. The image sequence includes multiple images, and each image in the image sequence includes a facial action. Taking the acquisition of the facial action sequence of the target user as an example, the device can collect multiple images through a camera, and each image includes a facial action performed by the user. Specifically, a facial action sequence refers to an image sequence including multiple facial actions. For example, a facial action sequence includes four facial actions of blinking, sticking out the tongue, pinching the nose, and covering the ears performed in sequence. The facial action sequence of the user can be obtained in the following manner: a facial action of the user is collected through a camera; a continuing collection instruction or a stopping collection instruction is detected; wherein the continuing collection instruction is used to indicate the continuing collection of facial actions, and the stopping collection instruction is used to indicate the stopping of the collection of facial actions; if the continuing collection instruction is obtained, the step of obtaining a facial action of the user through the camera is started again; if the stopping collection instruction is obtained, the facial action sequence of the user is generated according to each facial action of the user that has been obtained. In addition, the continuing collection instruction and the stopping collection instruction can be triggered by the corresponding selection control. In the time dimension, a facial action sequence is formed by executing multiple facial actions in sequence. In addition, the standard facial action information is the facial action information entered in advance by the corresponding legal user. Then, it is determined whether each facial action contained in the facial action information matches the facial action at the corresponding position in the standard facial action information. Wherein.The process of comparing whether each group of facial movements matches may include: identifying the user's single facial movement from the image containing the user's single facial movement; if the user's single facial movement is the same as the single facial movement at the corresponding position in the standard facial movement information, then the group of facial movements is determined to be matched, otherwise the group of facial movements is determined to be unmatched. If all are matched, the preset handwriting input reminder information is displayed. Wherein, the handwriting input reminder information includes the standard word corresponding to the specified user information. Wherein, the content of the handwriting input reminder information is not specifically limited, and can be set according to actual needs, for example, it can be "Please write the standard word in handwriting, thank you". Subsequently, the handwriting input by the user according to the standard word is collected. Wherein, the user can also be reminded to input handwriting at a preset position, and the preset position can be a position corresponding to the touch screen. The user can write handwriting corresponding to the standard word on the handwriting input device of the device, such as the touch screen of the device, so that the handwriting information corresponding to the standard word can be collected. After obtaining the handwriting information, the first handwriting feature data of the handwriting is extracted, and the second handwriting feature data of the standard handwriting corresponding to the standard word is extracted. Among them, the extraction process and matching process of the handwriting feature data of the handwriting can refer to the existing technology, and will not be elaborated here. Finally, it is determined whether the first handwriting feature data matches the second handwriting feature data. If they match, it is determined that the identity authentication is passed, otherwise it is determined that the identity authentication is not passed. In this embodiment, by adopting a multi-dimensional matching method to authenticate the user, the user will be considered to have passed the identity authentication only when the user information, facial motion recognition and handwriting recognition are all verified at the same time, thereby improving the security of the user's identity authentication process. In addition, since both the facial motion information and the handwriting information can be collected by the camera, it will not affect the convenience of information collection, and will not increase the hardware cost of the device. In addition, by introducing facial motion information combined with handwriting information for identity authentication, even if the user's other identity information is obtained by others, as long as the facial motion information and handwriting information set by the user are not leaked to others, others still cannot forge the user to pass the identity authentication, which effectively improves the security of identity authentication. In addition, only when the user passes the identity authentication, the received claim request will be further processed later, avoiding responding to suspected insurance fraud that has not passed the identity authentication, and ensuring the security and standardization of the claim request during the processing.
[0100] Furthermore, in an embodiment of the present application, before the above step S40, the following steps are included:
[0101] S400: Acquire a plurality of prediction models generated by pre-training; wherein each of the prediction models is generated by training a preset convolutional neural network model according to a respective different training sample set, each of the training sample sets includes a plurality of training images and labels corresponding to the training images, and the number of the prediction models is equal to the number of the training sample sets;
[0102] S401: Obtain a preset verification sample set;
[0103] S402: generating a prediction processing time of each prediction model based on the verification sample set;
[0104] S403: Based on the verification sample set, generating evaluation scores for each prediction model according to preset calculation rules;
[0105] S404: Acquire a first weight corresponding to the predicted processing time, and acquire a second weight corresponding to the evaluation score;
[0106] S405: Based on the prediction processing time, the evaluation score, the first weight, and the second weight, a first preset formula is called to calculate a model score of each prediction model;
[0107] S406: Filtering out the prediction model with the highest model score from all the prediction models;
[0108] S407: The prediction model with the highest model score is used as the target prediction model.
[0109] As described in the above steps S400 to S407, before executing the step of inputting the vehicle damage picture into the preset target prediction model for prediction, a process of determining the target prediction model may also be included. Specifically, first obtain a plurality of prediction models generated by pre-training. Among them, each of the prediction models is generated by training a preset convolutional neural network model according to different training sample sets, and each of the training sample sets includes a number of training pictures and labels corresponding to the training pictures, and the number of the prediction models is equal to the number of the training sample sets. Then obtain a preset verification sample set. Among them, the verification sample set can be generated based on the training samples, such as randomly obtaining data with a preset numerical ratio from the training samples as the verification sample set, and the preset numerical ratio can be set according to actual needs, for example, it can be set to 20%. In addition, the verification sample set includes a plurality of verification pictures, and labels corresponding to each of the verification pictures. And the prediction processing time of each prediction model is generated based on the verification sample set. Wherein, the implementation process of generating the predicted processing time may include: when the specified prediction model obtains each of the verification pictures, respectively counting the processing time of the predicted probability output by the specified prediction model corresponding to each of the verification pictures. Wherein, the specified prediction model is any one of all the prediction models, and the verification sample set includes multiple verification pictures; removing the first processing time with the largest value and the second processing time with the smallest value from all the processing times to obtain the corresponding third processing time; calculating the sum of all the third processing times and obtaining the number of all the third processing times; calculating the ratio between the sum and the number; using the ratio as the predicted processing time of the specified prediction model. And based on the verification sample set, generating the evaluation score of each of the prediction models according to the preset calculation rules. Wherein, the specific implementation process of calculating and generating the evaluation score will be further described in the subsequent specific embodiments, which will not be repeated here. Then, obtaining the first weight corresponding to the predicted processing time, and obtaining the second weight corresponding to the evaluation score. Wherein, the values of the first weight and the second weight are not specifically limited, and can be set according to actual needs. Subsequently, based on the prediction processing time, the evaluation score, the first weight and the second weight, the first preset formula is called to calculate the model score of each prediction model. The first preset formula is specifically: S=Y*a+P*b, S is the model score, Y is the prediction processing time, a is the first weight, P is the evaluation score, and b is the second weight. Finally, the prediction model with the highest model score is selected from all the prediction models, and the prediction model with the highest model score is used as the target prediction model.In this embodiment, the prediction processing time and evaluation score of each prediction model are generated by using a preset verification sample set, and then the target prediction model can be determined based on the obtained evaluation score and prediction processing time. Since the obtained target prediction model has a higher evaluation score and faster processing efficiency, the target prediction model can be used to predict the vehicle damage pictures uploaded by the user in the future, which can effectively improve the data accuracy of the claim probability corresponding to the vehicle damage pictures, and improve the analysis and processing efficiency of the vehicle damage pictures, which is conducive to the subsequent effective identification of insurance fraud based on the claim probability generated by the targeted prediction model.
[0110] Furthermore, in an embodiment of the present application, the above step S403 includes:
[0111] S4030: Input the verification sample set into a specified prediction model, and calculate the accuracy of the specified prediction model; wherein the specified prediction model is any one of all the prediction models;
[0112] S4031: Input the verification sample set into a specified prediction model, and calculate the recall rate of the specified prediction model;
[0113] S4032: Obtaining a preset value;
[0114] S4033: Based on the precision, the recall and the preset value, a second preset formula is called to calculate an evaluation score of the specified prediction model.
[0115] As described in steps S4030 to S4033 above, the step of generating the evaluation scores of each prediction model according to the preset calculation rules based on the verification sample set may specifically include: first, inputting the verification sample set into the specified prediction model, and calculating the precision of the specified prediction model. Wherein, the specified prediction model is any one of all the prediction models. And inputting the verification sample set into the specified prediction model, and calculating the recall of the specified prediction model. Wherein, the precision and recall of the generated prediction model can be calculated by the confusion matrix. The confusion matrix is used to measure the accuracy of a classifier. For the binary classification problem, its samples are divided into four situations: true positive examples (TruePositive, TP), false positive examples (FalsePositive, FP), true negative examples (TrueNegative, TN), and false negative examples (FalseNegative, FN) according to the combination of the true category and the predicted category of the classifier. Based on the confusion matrix, the precision and recall can be calculated according to the following formula: precision P = TP / (TP+FP), indicating that both the real situation and the predicted result are true / the predicted result is true; recall R = TP / (TP+FN), indicating that both the real situation and the predicted result are true / the real situation is true. Then obtain a preset value. Wherein, the preset value may refer to the ratio between the first preset weight of the precision and the second preset weight of the recall. Wherein, the values of the first preset weight and the second preset weight are not specifically limited and can be set according to actual needs. For example, the first preset weight is 1, the second preset weight is 0.5, or the first preset weight is 0.5, the second preset weight is 1, and so on. Finally, based on the precision, the recall and the preset value, the second preset formula is called to calculate the evaluation score of the specified prediction model. Wherein, the evaluation score of the specified prediction model refers to the ratio of the precision, the recall, and the preset weights of the precision and the recall of the specified prediction model. The evaluation score is calculated and generated, and is used to characterize the data comprehensive processing capability of the specified prediction model. Specifically, the precision, recall and preset values may be substituted into a second preset formula to generate a corresponding evaluation score. The second preset formula is specifically: in, For evaluation scores, is the preset value, P is the precision, and R is the recall. For example, if the precision is 0.95, the recall is 0.8, the first weight of the precision is 1, and the second weight of the recall is 0.5, then the preset value is 1 / 0.5=2. In this embodiment, by generating the precision and recall rate of the specified prediction model based on the verification sample set, the second preset formula related to the recall rate and the recall rate can be used to quickly generate the evaluation score of the specified prediction model, which is conducive to the subsequent rapid and accurate determination of the target prediction model from all prediction models based on the evaluation score of the specified prediction model. Since the evaluation score corresponds to the data comprehensive processing capability of the model, the target prediction model finally determined has a high data comprehensive processing capability and processing efficiency, so that the target prediction model is used to predict the vehicle damage pictures uploaded by the user in the future, which can effectively improve the data accuracy of the claim probability corresponding to the vehicle damage pictures, and improve the analysis and processing efficiency of the vehicle damage pictures, which is conducive to the subsequent effective identification of insurance fraud based on the claim probability generated by the target prediction model.
[0116] Furthermore, in an embodiment of the present application, the above step S70 includes:
[0117] S700: Displaying a preset repair shop list, and receiving a target repair shop selected by the user from the repair shop list;
[0118] S701: Obtaining the auto parts prices and labor hour prices of the target repair shop;
[0119] S702: Acquire vehicle information corresponding to the vehicle damage picture; wherein the vehicle information includes vehicle brand, vehicle series, and vehicle model;
[0120] S703: Calculate a corresponding damage amount based on the target loss part information, the target loss degree information, the vehicle information, the auto parts price and the labor time price;
[0121] S704: Using the assessed damage amount as the vehicle damage assessment result.
[0122] As described in the above steps S700 to S704, the step of generating the vehicle damage assessment result corresponding to the vehicle damage picture based on the target loss part information and the target loss degree information may specifically include: first displaying a preset list of repair shops, and receiving the target repair shop selected by the user from the list of repair shops. The repair shop list is a pre-created list of multiple repair shops. Then the auto parts price and labor time price of the target repair shop are obtained. Then the vehicle information corresponding to the vehicle damage picture is obtained. The vehicle information includes the vehicle brand, car series, and vehicle model. In addition, the vehicle information can be obtained by user input, or can be obtained by querying the user information of the user. Subsequently, the corresponding damage assessment amount is calculated based on the target loss part information, the target loss degree information, the vehicle information, the auto parts price, and the labor time price. When calculating the damage assessment amount, the damage assessment scheme can be searched from the preset vehicle claim damage assessment database according to the vehicle information, the loss part, the loss degree, and the target repair shop, and the damage assessment price is calculated. The damage assessment scheme includes loss parts, loss degree, parts price, labor time, etc. Specifically, first, it is possible to determine whether the lost parts are regular parts according to the information such as the vehicle brand, car series, model, lost parts, and degree of loss; if it is a regular part, then the selected work time items are searched in the work time library and the selected parts items are searched in the parts library according to the model, loss part and degree of loss; then, the client is notified to prompt the user to select a repair shop from the repair shop list to obtain the corresponding target repair shop; the parts price and work time price of the selected target repair shop are subsequently called to calculate the loss assessment amount. Finally, the loss assessment amount is used as the vehicle loss assessment result. In this embodiment, after obtaining the target loss part information and target loss degree information corresponding to the vehicle damage picture uploaded by the user, the target repair shop will be further obtained according to the user's selection operation on the repair shop list, so as to obtain the auto parts price and work time price of the target repair shop, and obtain the vehicle information corresponding to the vehicle damage picture, and then the loss assessment amount corresponding to the vehicle damage picture can be calculated according to all the above-mentioned data, that is, the vehicle loss assessment result, thereby effectively realizing the automation of vehicle claims, improving the intelligence and efficiency of claims, reducing the risk of human operation, reducing the deviation of case claims, and improving the level of customer service.
[0123] Furthermore, in an embodiment of the present application, before the above step S60, the following steps are included:
[0124] S600: collecting a preset number of vehicle damage sample data, and performing a preset process on the vehicle damage sample data to obtain processed vehicle damage sample data;
[0125] S601: Determine the loss location information and loss degree information corresponding to the processed vehicle damage sample data;
[0126] S602: Using the processed vehicle damage sample data as input of a preset classification model, and using the damage location information and the damage degree information as outputs of the classification model to train and obtain an initial classification model;
[0127] S603: Acquire a preset test sample set, and perform an accuracy test on the initial classification model based on the test sample set to obtain a corresponding classification accuracy rate;
[0128] S604: Determine whether the classification accuracy is greater than a preset accuracy threshold;
[0129] S605: If yes, use the initial classification model as the recognition model.
[0130] As described in the above steps S600 to S605, before executing the step of calling the preset recognition model to perform recognition processing on the vehicle damage picture and generating the target loss part information and target loss degree information corresponding to the vehicle damage picture, a recognition model training generation process may also be included. Specifically, firstly, a preset number of vehicle damage sample data are collected, and the vehicle damage sample data are subjected to preset processing to obtain the processed vehicle damage sample data. Among them, the preset processing process may include: analog-to-digital conversion, binarization, image smoothing, transformation, enhancement, restoration, filtering, etc. Then, the loss part information and loss degree information corresponding to the processed vehicle damage sample data are determined. Among them, the loss part information and loss degree information can be generated by professionals after analyzing and identifying the processed vehicle damage sample data. After that, the processed vehicle damage sample data is used as the input of the preset classification model, and the loss part information and the loss degree information are used as the output of the classification model for training to obtain an initial classification model. Among them, the classification model can adopt a classification model established according to any classification method in the prior art, such as SVM, Bayesian algorithm, nearest neighbor classification algorithm, maximum expectation algorithm, etc. Subsequently, a preset test sample set is obtained, and the accuracy test of the initial classification model is performed based on the test sample set to obtain the corresponding classification accuracy rate. Finally, it is determined whether the classification accuracy rate is greater than a preset accuracy rate threshold. If so, the initial classification model is used as the recognition model. Among them, after the classification model is trained for a period of time to obtain an initial classification model, the accuracy of the initial classification model is tested using the test sample set, that is, several test samples in the test sample set are input into the initial classification model, and the initial classification model predicts the classification label result of each test sample, and compares the classification label result of each test sample with the correct label (loss location information and loss degree information) marked for the test sample. If they are consistent, the initial classification model accurately classifies the test sample. If they are inconsistent, the initial classification model classifies the test sample incorrectly, thereby statistically obtaining the prediction accuracy of the initial classification model (prediction accuracy = number of test samples with accurate predictions / total number of test samples). If the obtained prediction accuracy meets the set accuracy requirement, that is, it is greater than the preset accuracy threshold, the initial classification model converges, thereby ending the training of the initial classification model, and the initial classification model that has finished training is used as the recognition model, and is used to recognize and process the vehicle damage pictures uploaded by the user.Furthermore, if the prediction accuracy is less than or equal to the accuracy threshold, it means that the training of the trained initial classification model has not reached the preset standard. It may be that the number of samples of vehicle damage sample data used for training is too small or the number of samples in the test sample set is too small. Therefore, in this case, the preset number of vehicle damage sample data is increased (that is, the number of samples is increased, for example, a fixed number is increased each time or a random number is increased each time), and then the above-mentioned training steps and testing steps are re-executed on this basis. The cycle is executed until the model accuracy of the trained initial classification model is greater than the preset accuracy threshold, and the model training is terminated. In the present embodiment, the preset classification model is trained and tested based on vehicle damage training samples including a preset number of vehicle damage sample data and loss location information and loss degree information corresponding to the vehicle damage sample data, so that a recognition model that meets the actual use needs can be intelligently and quickly generated, which is beneficial for subsequent recognition and processing of vehicle damage pictures uploaded by users based on the recognition model, and then accurate and rapid generation of target loss location information and target loss degree information corresponding to the vehicle damage pictures, which is beneficial for subsequent automatic and rapid generation of corresponding vehicle damage assessment results based on the target loss location information and the target loss degree information, and displaying the vehicle damage assessment results to users, thereby realizing the automation of vehicle claims, improving the intelligence level and efficiency of claims, reducing the risk of human operation, reducing case claims deviations, and improving customer service levels.
[0131] Furthermore, in an embodiment of the present application, before the above step S40, the following steps are included:
[0132] S410: generating a blur level value corresponding to the vehicle damage image based on a preset algorithm;
[0133] S411: Obtaining a preset fuzzy threshold;
[0134] S412: Determine whether the blur level value is less than the blur threshold;
[0135] S413: If the value is less than the fuzzy threshold, a processing instruction is generated to input the vehicle damage image into a preset target prediction model for prediction;
[0136] S414: If the blur threshold is greater than the blur threshold, a preset shooting reminder message is displayed to remind the user to reshoot and upload a vehicle damage photo that meets the requirements.
[0137] As described in steps S410 to S414 above, before the step of inputting the vehicle damage picture into the preset target prediction model for prediction, a quality detection process for the vehicle damage picture may also be included. Specifically, first, a blur degree value corresponding to the vehicle damage picture is generated based on a preset algorithm. Among them, the blur degree value is used to indicate the blur degree of the vehicle damage picture. When the blur degree value is larger, the blur degree of the corresponding vehicle damage picture is also larger, and vice versa, the blur degree of the corresponding vehicle damage picture is also smaller. In addition, the preset algorithm is specifically a Laplace operator, and the blur degree value of the vehicle damage picture can be calculated by the Laplace operator. Then a preset blur threshold is obtained. Among them, the value of the blur threshold is not specifically limited and can be set according to actual needs. If the blur degree value corresponding to the generated vehicle damage picture is greater than the blur threshold, it indicates that the vehicle damage picture does not meet the requirements and the user needs to re-shoot and upload it. Subsequently, it is determined whether the blur degree value is less than the blur threshold. If it is less than the blur threshold, a processing instruction is generated to input the vehicle damage picture into the preset target prediction model for prediction. If it is greater than the blur threshold, a preset shooting reminder message is displayed to remind the user to reshoot and upload a car damage photo that meets the requirements. In this embodiment, after the user takes the car damage photo and uploads it, the car damage picture will be intelligently blurred to determine whether it meets the standard processing requirements, that is, the blur degree value corresponding to the car damage picture is calculated and compared with the preset blur threshold. Only when the blur degree value is less than the blur threshold, a processing instruction will be generated to input the car damage picture into the preset target prediction model to execute the subsequent claim processing for the car damage picture, ensuring the standardization and accuracy of the claim process, avoiding the waste of work caused by processing car damage pictures that do not meet the processing requirements, and improving the intelligence of the claim processing.
[0138] The claims settlement method based on artificial intelligence in the embodiment of the present application can also be applied to the field of blockchain, such as storing the above-mentioned vehicle damage assessment results and other data on the blockchain. By using blockchain to store and manage the above-mentioned vehicle damage assessment results, the security and non-tamperability of the above-mentioned vehicle damage assessment results can be effectively guaranteed.
[0139] The above-mentioned blockchain is a new application model of computer technologies such as distributed data storage, peer-to-peer transmission, consensus mechanism, encryption algorithm, etc. Blockchain is essentially a decentralized database, a string of data blocks generated by cryptographic methods. Each data block contains a batch of network transaction information, which is used to verify the validity of its information (anti-counterfeiting) and generate the next block. Blockchain can include the underlying blockchain platform, platform product service layer, and application service layer.
[0140] The underlying blockchain platform can include processing modules such as user management, basic services, smart contracts, and operation monitoring. Among them, the user management module is responsible for the identity information management of all blockchain participants, including maintaining the generation of public and private keys (account management), key management, and the maintenance of the corresponding relationship between the user's real identity and the blockchain address (authority management), etc., and under authorization, supervises and audits the transactions of certain real identities and provides risk control rule configuration (risk control audit). The basic service module is deployed on all blockchain node devices to verify the validity of business requests and record the valid requests to the storage after consensus is reached. For a new business request, the basic service first performs interface adaptation analysis and authentication processing (interface adaptation), and then encrypts the business information through the consensus algorithm (consensus management), and transmits it to the shared ledger in a complete and consistent manner after encryption (network communication), and records and stores it. The smart contract module is responsible for the registration and issuance of contracts, contract triggering, and contract execution. Developers can define the contract logic in a certain programming language and publish it to the blockchain (contract registration). According to the logic of the contract terms, call the key or other events to trigger the execution and complete the contract logic. At the same time, it also provides the function of contract upgrade and cancellation. The operation monitoring module is mainly responsible for the deployment, configuration modification, contract setting, cloud adaptation and visual output of the real-time status of product operation during the product release process, such as alarms, monitoring network conditions, monitoring node equipment health status, etc.
[0141] Reference Figure 2 In one embodiment of the present application, there is also provided a claims settlement device based on artificial intelligence, including:
[0142] The first judgment module 1 is used to judge whether a claim request triggered by a user through a client is received; wherein the claim request carries user information;
[0143] Verification module 2 is used for obtaining the user information, the user's facial movement information and the user's handwriting upon receiving the claim request, performing identity verification on the user based on a preset claim user whitelist, standard facial movement information and standard handwriting, and determining whether the identity verification is passed;
[0144] Reminding module 3, for reminding the user to upload a car damage picture based on preset reminder information if the identity verification is passed;
[0145] The first generating module 4 is used to input the vehicle damage picture into a preset target prediction model for prediction, and output the claim probability corresponding to the vehicle damage picture through the target prediction model;
[0146] The second judgment module 5 is used to judge whether the claim settlement probability is greater than a preset probability threshold;
[0147] The second generating module 6 is used to call a preset recognition model to perform recognition processing on the vehicle damage picture if the probability is greater than the probability threshold, and generate target loss part information and target loss degree information corresponding to the vehicle damage picture;
[0148] The third generating module 7 is used to generate a vehicle damage assessment result corresponding to the vehicle damage picture based on the target loss part information and the target loss degree information.
[0149] In this embodiment, the operations performed by the above modules or units respectively correspond one-to-one to the steps of the artificial intelligence-based claims settlement method in the aforementioned implementation method, and will not be repeated here.
[0150] Furthermore, in one embodiment of the present application, the verification module 2 includes:
[0151] A first acquisition unit is used to acquire a preset claims user whitelist and determine whether there is designated user information identical to the user information in the claims user whitelist;
[0152] A first reminder unit, configured to display preset action input reminder information if the specified user information exists, so as to remind the user to input corresponding facial action information based on the action input reminder information;
[0153] A first collecting unit, configured to collect the facial action information input by the user, and obtain standard facial action information corresponding to the designated user information; wherein the facial action information includes a facial action sequence, and the facial action sequence includes a plurality of facial actions executed in sequence;
[0154] a first judging unit, configured to judge whether each facial action included in the facial action information completely matches the facial action at a corresponding position in the standard facial action information;
[0155] A second reminder unit, used for displaying preset handwriting input reminder information if all matches are found; wherein the handwriting input reminder information includes a standard character corresponding to the specified user information;
[0156] A second collecting unit, used for collecting handwriting input by the user according to the standard character;
[0157] An extraction unit, used for extracting first handwriting feature data of the handwriting, and extracting second handwriting feature data of the standard handwriting corresponding to the standard character;
[0158] A second judging unit, used for judging whether the first handwriting feature data matches the second handwriting feature data;
[0159] The determination unit is used to determine that the identity authentication is passed if there is a match, and otherwise determine that the identity authentication is not passed.
[0160] In this embodiment, the operations performed by the above modules or units respectively correspond one-to-one to the steps of the artificial intelligence-based claims settlement method in the aforementioned implementation method, and will not be repeated here.
[0161] Furthermore, in one embodiment of the present application, the above-mentioned claims settlement device based on artificial intelligence includes:
[0162] A first acquisition module is used to acquire a plurality of prediction models generated by pre-training; wherein each of the prediction models is generated by training a preset convolutional neural network model according to a respective different training sample set, each of the training sample sets includes a plurality of training images and labels corresponding to the training images, and the number of the prediction models is equal to the number of the training sample sets;
[0163] A second acquisition module is used to acquire a preset verification sample set;
[0164] A fourth generating module, used for generating a prediction processing time of each prediction model based on the verification sample set;
[0165] A fifth generating module, used to generate an evaluation score of each of the prediction models based on the verification sample set and according to a preset calculation rule;
[0166] A third acquisition module, used to acquire a first weight corresponding to the predicted processing time, and to acquire a second weight corresponding to the evaluation score;
[0167] A calculation module, configured to calculate a model score of each of the prediction models by calling a first preset formula based on the prediction processing time, the evaluation score, the first weight, and the second weight;
[0168] A screening module, used to screen out the prediction model with the highest model score from all the prediction models;
[0169] The first determination module is used to select the prediction model with the highest model score as the target prediction model.
[0170] In this embodiment, the operations performed by the above modules or units respectively correspond one-to-one to the steps of the artificial intelligence-based claims settlement method in the aforementioned implementation method, and will not be repeated here.
[0171] Furthermore, in one embodiment of the present application, the above-mentioned calculation module includes:
[0172] A first calculation unit, used for inputting the verification sample set into a specified prediction model, and calculating the accuracy of the specified prediction model; wherein the specified prediction model is any one of all the prediction models;
[0173] A second calculation unit is used to input the verification sample set into a specified prediction model to calculate the recall rate of the specified prediction model;
[0174] A second acquisition unit, used to acquire a preset value;
[0175] The third calculation unit is used to call the second preset formula to calculate the evaluation score of the specified prediction model based on the precision, the recall and the preset value.
[0176] In this embodiment, the operations performed by the above modules or units respectively correspond one-to-one to the steps of the artificial intelligence-based claims settlement method in the aforementioned implementation method, and will not be repeated here.
[0177] Furthermore, in one embodiment of the present application, the third generation module 7 includes:
[0178] A receiving unit, used for displaying a preset repair shop list and receiving a target repair shop selected by the user from the repair shop list;
[0179] A third acquisition unit is used to acquire the auto parts price and labor time price of the target repair shop;
[0180] A fourth acquisition unit is used to acquire vehicle information corresponding to the vehicle damage picture; wherein the vehicle information includes vehicle brand, vehicle series, and vehicle model;
[0181] A fourth calculation unit, configured to calculate a corresponding damage assessment amount based on the target loss part information, the target loss degree information, the vehicle information, the auto parts price and the labor time price;
[0182] A determination unit is used to use the damage assessment amount as the vehicle damage assessment result.
[0183] In this embodiment, the operations performed by the above modules or units respectively correspond one-to-one to the steps of the artificial intelligence-based claims settlement method in the aforementioned implementation method, and will not be repeated here.
[0184] Furthermore, in one embodiment of the present application, the above-mentioned claims settlement device based on artificial intelligence includes:
[0185] A processing module, used to collect a preset number of vehicle damage sample data, and perform preset processing on the vehicle damage sample data to obtain processed vehicle damage sample data;
[0186] A second determination module is used to determine the loss location information and loss degree information corresponding to the processed vehicle damage sample data;
[0187] A training module, used to use the processed vehicle damage sample data as input of a preset classification model, and use the loss location information and the loss degree information as outputs of the classification model to train and obtain an initial classification model;
[0188] A testing module, used to obtain a preset test sample set, and perform an accuracy test on the initial classification model based on the test sample set to obtain a corresponding classification accuracy rate;
[0189] A third judgment module is used to judge whether the classification accuracy is greater than a preset accuracy threshold;
[0190] The third determining module is configured to, if yes, use the initial classification model as the recognition model.
[0191] In this embodiment, the operations performed by the above modules or units respectively correspond one-to-one to the steps of the artificial intelligence-based claims settlement method in the aforementioned implementation method, and will not be repeated here.
[0192] Furthermore, in one embodiment of the present application, the above-mentioned claims settlement device based on artificial intelligence includes:
[0193] A sixth generating module, configured to generate a blur degree value corresponding to the vehicle damage image based on a preset algorithm;
[0194] A fourth acquisition module, used to acquire a preset fuzzy threshold;
[0195] A fourth judgment module, used to judge whether the blur degree value is less than the blur threshold;
[0196] A seventh generating module, for generating a processing instruction for inputting the vehicle damage picture into a preset target prediction model for prediction if the value is less than the fuzzy threshold;
[0197] The display module is used to display preset shooting reminder information if the blur threshold is greater than the blur threshold, so as to remind the user to re-shoot and upload the vehicle damage photos that meet the requirements.
[0198] In this embodiment, the operations performed by the above modules or units respectively correspond one-to-one to the steps of the artificial intelligence-based claims settlement method in the aforementioned implementation method, and will not be repeated here.
[0199] Reference Figure 3 In an embodiment of the present application, a computer device is also provided. The computer device may be a server, and its internal structure may be as follows: Figure 3As shown. The computer device includes a processor, a memory, a network interface, a display screen, an input device and a database connected through a system bus. Among them, the processor designed for the computer device is used to provide computing and control capabilities. The memory of the computer device includes a storage medium and an internal memory. The storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the storage medium. The database of the computer device is used to store user information, reminder information, vehicle damage pictures, claim probability, target loss part information, target loss degree information and vehicle damage assessment results. The network interface of the computer device is used to communicate with an external terminal through a network connection. The display screen of the computer device is an indispensable graphic output device in the computer, which is used to convert digital signals into optical signals so that text and graphics are displayed on the screen of the display screen. The input device of the computer device is the main device for information exchange between the computer and the user or other devices, and is used to transmit data, instructions and certain sign information to the computer. When the computer program is executed by the processor, a claims method based on artificial intelligence is implemented.
[0200] The processor executes the steps of the above-mentioned claims settlement method based on artificial intelligence:
[0201] Determine whether a claim request triggered by a user through a client is received; wherein the claim request carries user information; if the claim request is received, obtain the user information, the user's facial movement information and the user's handwriting, authenticate the user based on a preset claim user whitelist, standard facial movement information and standard handwriting, and determine whether the identity authentication is passed; if the identity authentication is passed, remind the user to upload a vehicle damage picture based on a preset reminder information; input the vehicle damage picture into a preset target prediction model for prediction, and output a claim probability corresponding to the vehicle damage picture through the target prediction model; determine whether the claim probability is greater than a preset probability threshold; if it is greater than the probability threshold, call a preset recognition model to perform recognition processing on the vehicle damage picture, and generate target loss part information and target loss degree information corresponding to the vehicle damage picture; generate a vehicle damage assessment result corresponding to the vehicle damage picture based on the target loss part information and the target loss degree information.
[0202] Those skilled in the art will understand that Figure 3 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the device or computer equipment to which the solution of the present application is applied. An embodiment of the present application also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, an artificial intelligence-based claims method is implemented, specifically:
[0203] Determine whether a claim request triggered by a user through a client is received; wherein the claim request carries user information; if the claim request is received, obtain the user information, the user's facial movement information and the user's handwriting, authenticate the user based on a preset claim user whitelist, standard facial movement information and standard handwriting, and determine whether the identity authentication is passed; if the identity authentication is passed, remind the user to upload a vehicle damage picture based on a preset reminder information; input the vehicle damage picture into a preset target prediction model for prediction, and output a claim probability corresponding to the vehicle damage picture through the target prediction model; determine whether the claim probability is greater than a preset probability threshold; if it is greater than the probability threshold, call a preset recognition model to perform recognition processing on the vehicle damage picture, and generate target loss part information and target loss degree information corresponding to the vehicle damage picture; generate a vehicle damage assessment result corresponding to the vehicle damage picture based on the target loss part information and the target loss degree information.
[0204] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a 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 provided in the present application and used in the embodiments 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. As an illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double-speed data rate SDRAM (SSRSDRAM), 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).
[0205] The above description is only a preferred embodiment of the present application, and does not limit the patent scope of the present application. Any equivalent structure or equivalent process transformation made using the contents of the present application specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present application.
Claims
1. A claims settlement method based on artificial intelligence, characterized in that: include: Determine whether a claim request triggered by a user through a client is received; wherein the claim request carries user information; If the claim request is received, the user information, the user's facial movement information and the user's handwriting are obtained, identity verification of the user is performed based on a preset claim user whitelist, standard facial movement information and standard handwriting, and whether the identity verification is passed; If the identity verification is passed, the user is reminded to upload the vehicle damage picture based on the preset reminder information; Inputting the vehicle damage picture into a preset target prediction model for prediction, and outputting the claim probability corresponding to the vehicle damage picture through the target prediction model; Determining whether the claim settlement probability is greater than a preset probability threshold; If it is greater than the probability threshold, calling a preset recognition model to perform recognition processing on the vehicle damage picture, and generating target loss part information and target loss degree information corresponding to the vehicle damage picture; Generate a vehicle damage assessment result corresponding to the vehicle damage picture based on the target damage location information and the target damage degree information; Before the step of inputting the vehicle damage picture into a preset target prediction model for prediction, the method includes: Acquire multiple prediction models generated by pre-training; wherein each of the prediction models is generated by training a preset convolutional neural network model according to a respective different training sample set, each of the training sample sets includes a number of training pictures and labels corresponding to the training pictures, and the number of the prediction models is equal to the number of the training sample sets; Get the preset verification sample set; Generate a prediction processing time for each prediction model based on the validation sample set; Based on the verification sample set, generating evaluation scores for each of the prediction models according to preset calculation rules; Acquire a first weight corresponding to the predicted processing time, and acquire a second weight corresponding to the evaluation score; Based on the prediction processing time, the evaluation score, the first weight, and the second weight, calling a first preset formula to calculate a model score of each of the prediction models; Selecting a prediction model with the highest model score from all the prediction models; The prediction model with the highest model score is used as the target prediction model; The step of generating a vehicle damage assessment result corresponding to the vehicle damage picture based on the target damage location information and the target damage degree information comprises: Displaying a preset repair shop list, and receiving a target repair shop selected by the user from the repair shop list; Obtaining the prices of auto parts and labor hours of the target repair shop; Obtaining vehicle information corresponding to the vehicle damage picture; wherein the vehicle information includes vehicle brand, car series, and vehicle model; The corresponding damage amount is calculated based on the target loss part information, the target loss degree information, the vehicle information, the auto parts price and the labor time price; The damage assessment amount is used as the vehicle damage assessment result.
2. The claims settlement method based on artificial intelligence according to claim 1, characterized in that: The steps of obtaining the user information, the user's facial movement information and the user's handwriting, performing identity authentication on the user based on a preset claims user whitelist, standard facial movement information and standard handwriting, and determining whether the identity authentication is passed include: Obtain a preset claims user whitelist, and determine whether there is designated user information identical to the user information in the claims user whitelist; If the designated user information exists, displaying preset action input reminder information to remind the user to input corresponding facial action information based on the action input reminder information; Collecting the facial action information input by the user, and acquiring standard facial action information corresponding to the designated user information; wherein the facial action information includes a facial action sequence, and the facial action sequence includes a plurality of facial actions executed in sequence; Determining whether each facial action included in the facial action information completely matches the facial action at the corresponding position in the standard facial action information; If all are matched, display the preset handwriting input reminder information; wherein the handwriting input reminder information includes the standard characters corresponding to the specified user information; collecting handwriting input by the user according to the standard characters; Extracting first handwriting feature data of the handwriting, and extracting second handwriting feature data of the standard handwriting corresponding to the standard character; Determining whether the first handwriting feature data matches the second handwriting feature data; If they match, the identity authentication is determined to be successful; otherwise, the identity authentication is determined to be unsuccessful.
3. The claims settlement method based on artificial intelligence according to claim 1, characterized in that: The step of generating the evaluation score of each prediction model based on the verification sample set according to a preset calculation rule includes: Input the verification sample set into a specified prediction model, and calculate the accuracy of the specified prediction model; wherein the specified prediction model is any one of all the prediction models; Inputting the verification sample set into a specified prediction model, and calculating the recall rate of the specified prediction model; Get the preset value; Based on the precision, the recall and the preset value, a second preset formula is called to calculate the evaluation score of the specified prediction model.
4. The claims settlement method based on artificial intelligence according to claim 1, characterized in that: Before the step of calling a preset recognition model to perform recognition processing on the vehicle damage picture and generating target loss part information and target loss degree information corresponding to the vehicle damage picture, the method includes: Collecting a preset number of vehicle damage sample data, and performing a preset process on the vehicle damage sample data to obtain processed vehicle damage sample data; Determining loss location information and loss degree information corresponding to the processed vehicle damage sample data; The processed vehicle damage sample data is used as the input of a preset classification model, and the loss location information and the loss degree information are used as the output of the classification model for training to obtain an initial classification model; Acquire a preset test sample set, and perform an accuracy test on the initial classification model based on the test sample set to obtain a corresponding classification accuracy rate; Determining whether the classification accuracy is greater than a preset accuracy threshold; If so, the initial classification model is used as the recognition model.
5. The claims settlement method based on artificial intelligence according to claim 1, characterized in that: Before the step of inputting the vehicle damage picture into a preset target prediction model for prediction, the method includes: Generate a blur degree value corresponding to the vehicle damage image based on a preset algorithm; Get the preset blur threshold; Determining whether the blur level value is less than the blur threshold; If it is less than the fuzzy threshold, a processing instruction is generated to input the vehicle damage picture into a preset target prediction model for prediction; If it is greater than the blur threshold, a preset shooting reminder message is displayed to remind the user to re-shoot and upload a vehicle damage photo that meets the requirements.
6. An artificial intelligence-based claims settlement device, used to execute the method described in any one of claims 1 to 5, characterized in that: include: A first judgment module is used to judge whether a claim request triggered by a user through a client is received; wherein the claim request carries user information; a verification module, for obtaining the user information, the user's facial movement information and the user's handwriting upon receiving the claim request, performing identity verification on the user based on a preset claim user whitelist, standard facial movement information and standard handwriting, and determining whether the identity verification is passed; A reminder module, used to remind the user to upload a car damage picture based on preset reminder information if the identity authentication is passed; A first generating module is used to input the vehicle damage picture into a preset target prediction model for prediction, and output the claim probability corresponding to the vehicle damage picture through the target prediction model; A second judgment module is used to judge whether the claim settlement probability is greater than a preset probability threshold; The second generation module is used to call a preset recognition model to perform recognition processing on the vehicle damage picture if the probability is greater than the probability threshold, and generate target loss part information and target loss degree information corresponding to the vehicle damage picture; The third generating module is used to generate a vehicle damage assessment result corresponding to the vehicle damage picture based on the target loss part information and the target loss degree information.
7. A computer device comprising a memory and a processor, wherein a computer program is stored in the memory, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 5 are implemented.
8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 5 are implemented.
Citation Information
Patent Citations
Car insurance intelligent loss assessment method and system
CN106600422A
Method and device for evaluating vehicle loss recognition model
CN110009509A
