WO2021012891A1 - 车辆定损方法、装置、设备和存储介质 - Google Patents

车辆定损方法、装置、设备和存储介质 Download PDF

Info

Publication number
WO2021012891A1
WO2021012891A1 PCT/CN2020/098767 CN2020098767W WO2021012891A1 WO 2021012891 A1 WO2021012891 A1 WO 2021012891A1 CN 2020098767 W CN2020098767 W CN 2020098767W WO 2021012891 A1 WO2021012891 A1 WO 2021012891A1
Authority
WO
WIPO (PCT)
Prior art keywords
vehicle accident
pspnet
refers
client
vehicle
Prior art date
Application number
PCT/CN2020/098767
Other languages
English (en)
French (fr)
Inventor
林立辉
刘金萍
尹钏
王鸿
林峰
Original Assignee
平安科技(深圳)有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 平安科技(深圳)有限公司 filed Critical 平安科技(深圳)有限公司
Publication of WO2021012891A1 publication Critical patent/WO2021012891A1/zh

Links

Images

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00Pattern recognition
    • G06F18/20Analysing
    • G06F18/21Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00Pattern recognition
    • G06F18/20Analysing
    • G06F18/22Matching criteria, e.g. proximity measures
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06NCOMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00Computing arrangements based on biological models
    • G06N3/02Neural networks
    • G06N3/04Architecture, e.g. interconnection topology
    • G06N3/045Combinations of networks
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06NCOMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00Computing arrangements based on biological models
    • G06N3/02Neural networks
    • G06N3/08Learning methods
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06QINFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q40/00Finance; Insurance; Tax strategies; Processing of corporate or income taxes
    • G06Q40/08Insurance
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06QINFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q50/00Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
    • G06Q50/40Business processes related to the transportation industry
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06VIMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00Scenes; Scene-specific elements
    • G06V20/50Context or environment of the image
    • G06V20/56Context or environment of the image exterior to a vehicle by using sensors mounted on the vehicle
    • G06V20/58Recognition of moving objects or obstacles, e.g. vehicles or pedestrians; Recognition of traffic objects, e.g. traffic signs, traffic lights or roads
    • G06V20/584Recognition of moving objects or obstacles, e.g. vehicles or pedestrians; Recognition of traffic objects, e.g. traffic signs, traffic lights or roads of vehicle lights or traffic lights
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06VIMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V2201/00Indexing scheme relating to image or video recognition or understanding
    • G06V2201/08Detecting or categorising vehicles

Definitions

  • This application relates to the field of image detection, and in particular to a method, device, equipment and storage medium for vehicle damage assessment.
  • a new claim settlement service model which applies AI image recognition loss assessment technology to auto insurance accident claim settlement scenarios, and provides a convenient client terminal so that users can quickly and conveniently complete the entire claims settlement process.
  • this application provides a vehicle damage assessment method, including:
  • Each vehicle accident case corresponds to at least one vehicle accident image.
  • the vehicle accident image set and the amount of compensation for the multiple vehicle accident cases are input into the semantic segmentation network PSPnet to train the PSPnet under multiple sets of hyperparameters; the pooling layer and the convolutional layer in the PSPnet are trained to pass Calculation.
  • the function value of the loss function refers to the vehicle accident in the PSPnet test set
  • the regularization coefficient C is calculated by the following formula:
  • C 0 is the initial value of the loss function
  • n' refers to the number of images in the vehicle accident image collection
  • refers to the ratio of the regular term to the C 0 term
  • the regular term refers to The sum of squares.
  • an instruction message is sent to the client.
  • the car damage photos sent by the client After receiving the car damage photos sent by the client, the car damage photos sent by the client are input to the target PSPnet, a loss assessment strategy is generated, and the loss assessment strategy is sent to the client.
  • the present application provides a vehicle damage assessment device, which has the function of implementing the method corresponding to the vehicle damage assessment platform provided in the first aspect.
  • the function can be realized by hardware, or by hardware executing corresponding software.
  • the hardware or software includes one or more modules corresponding to the above functions, and the modules may be software and/or hardware.
  • the vehicle damage assessment device includes:
  • the acquisition module acquires a collection of vehicle accident images to be processed and the amount of compensation for multiple vehicle accident cases.
  • Each vehicle accident case corresponds to at least one vehicle accident image.
  • the processing module inputs the set of vehicle accident images and the compensation amount of the multiple vehicle accident cases into the semantic segmentation network PSPnet to train the PSPnet under multiple sets of hyperparameters; train the pooling layer and volume in the PSPnet Layer by layer
  • the function value of the loss function refers to the vehicle accident in the PSPnet test set
  • the regularization coefficient C is calculated by the following formula:
  • C 0 is the initial value of the loss function
  • n' refers to the number of images in the vehicle accident image collection
  • refers to the ratio of the regular term to the C 0 term
  • the regular term refers to The sum of squares.
  • an instruction message is sent to the client.
  • a vehicle damage assessment device which includes at least one connected processor, a memory, and an input/output unit, wherein the memory is used to store program codes, and the processor is used to call the memory
  • each of the vehicle accident cases corresponds to at least one vehicle accident image
  • the vehicle accident image set and the amount of compensation for the multiple vehicle accident cases are input into the semantic segmentation network PSPnet to train the PSPnet under multiple sets of hyperparameters; the pooling layer and the convolutional layer in the PSPnet are trained to pass Calculation; where, Represents the weight obtained by training the kth neuron in the nth layer in the multi-layer perceptron of the PSPnet convolutional layer according to the output of the n-1th layer in the multi-layer perceptron of the PSPnet convolutional layer, Represents the corresponding offset, Represents the output of the j-th vehicle accident image of the i-th vehicle accident case in the nth layer of the PSPnet after being input to the PSPnet, i, j, and k are any positive integers, and n is a natural number; when n is 0 , Refers to the vehicle accident image; when n is the last layer of the PSPnet, Refers to the amount of compensation for the said vehicle accident case;
  • the function value of the loss function refers to the vehicle accident in the PSPnet test set
  • the regularization coefficient C is calculated by the following formula:
  • C 0 is the initial value of the loss function
  • n' refers to the number of images in the vehicle accident image collection
  • refers to the ratio of the regular term to the C 0 term
  • the regular term refers to Sum of squares
  • Another aspect of the present application provides a computer storage medium, the computer readable storage medium stores a computer program, the computer program includes program instructions, and when the program instructions are executed by a processor, they are used to implement the following steps:
  • each of the vehicle accident cases corresponds to at least one vehicle accident image
  • the vehicle accident image set and the amount of compensation for the multiple vehicle accident cases are input into the semantic segmentation network PSPnet to train the PSPnet under multiple sets of hyperparameters; the pooling layer and the convolutional layer in the PSPnet are trained to pass Calculation; where, Represents the weight obtained by training the kth neuron in the nth layer in the multi-layer perceptron of the PSPnet convolutional layer according to the output of the n-1th layer in the multi-layer perceptron of the PSPnet convolutional layer, Represents the corresponding offset, Represents the output of the j-th vehicle accident image of the i-th vehicle accident case in the nth layer of the PSPnet after being input to the PSPnet, i, j, and k are any positive integers, and n is a natural number; when n is 0 , Refers to the vehicle accident image; when n is the last layer of the PSPnet, Refers to the amount of compensation for the said vehicle accident case;
  • the function value of the loss function refers to the vehicle accident in the PSPnet test set
  • the regularization coefficient C is calculated by the following formula:
  • C 0 is the initial value of the loss function
  • n' refers to the number of images in the vehicle accident image collection
  • refers to the ratio of the regular term to the C 0 term
  • the regular term refers to Sum of squares
  • this application proposes a new claim service model, which uses AI image recognition and loss assessment technology in auto insurance accident claim scenarios, and provides convenient clients such as car owners and small programs, aiming to provide customers with a simple
  • the entire process of claim settlement can be completed by reporting the case and registering, taking pictures and uploading the damaged part, confirming the payment plan and entering the payment account.
  • we provide direct service The customer can contact the claim adjuster for help at any time during the self-assisted compensation process.
  • AI self-service reporting rules screen and filter risk cases; increase the control of the upper limit of the amount of case compensation, compulsory conversion of excess cases, and the intervention of claims adjusters.
  • FIG. 1 is a schematic flowchart of a method for determining vehicle damage in an embodiment of this application
  • FIG. 2 is a schematic diagram of the structure of a device for determining vehicle damage in an embodiment of the application
  • Fig. 3 is a schematic structural diagram of a computer device in an embodiment of the application.
  • the following is an example of a method for determining vehicle damage provided in this application, and the method includes:
  • Each vehicle accident case corresponds to at least one vehicle accident image.
  • the vehicle accident image collection includes multiple labeled vehicle accident images.
  • the marked area of the vehicle accident image includes vehicle characteristic information such as the front, rear, headlights, taillights, and rearview mirrors.
  • the vehicle accident image collection is divided into a test set and a training set.
  • the vehicle accident refers to a traffic accident that occurs on the road by a motor vehicle, does not involve casualties, and only causes property damage worth less than 5,000 yuan, such as front and rear bumpers, lights, hoods, doors and windows, rearview mirrors An accident in which the vehicle can continue to drive when external components are damaged.
  • the hyperparameters include at least one of the size of the convolution kernel, the learning rate ⁇ , the batch parameter, the number of neural network layers, the activation function, the optimizer, the batch size, and the number of epochs to be trained.
  • the hyperparameters are adjusted in a customized manner. By adjusting the hyperparameters, multiple sets of hyperparameters are obtained, and the test value and theoretical value under each set of hyperparameters are calculated respectively. The test value and theoretical value can calculate the error of the model, and the neural network under different parameters can be calculated by the error. The model is evaluated to obtain the neural network model parameters with the smallest error.
  • the function value of the loss function refers to the difference between the actual payout amount after the vehicle accident image in the PSPnet test set is input to the convolutional neural network and the expected payout amount after the vehicle accident image in the test set is input to the convolutional neural network
  • the sum of squares; the regularization coefficient C is calculated by the following formula:
  • C 0 is the initial value of the loss function
  • n' refers to the number of images in the vehicle accident image collection
  • refers to the ratio of the regular term to the C 0 term
  • the regular term refers to Sum of squares
  • the instruction message is used to order the client to send the damaged photo.
  • the client refers to a program that provides customer self-assisted compensation services.
  • the insured’s license plate is identified through the mobile phone number or WeChat ID carried by the client, and the location of the insurance is located through GPS, the associated vehicle information and associated insurance policy information are obtained according to the mobile phone number or WeChat ID, and based on the acquired vehicle information and associated insurance policy information
  • the report number is automatically generated to complete the compensation process by itself.
  • the claim settlement process refers to the processes of reporting and registering, taking pictures and uploading the damaged part, confirming the claim and contacting the special payment plan, and entering the payment account.
  • the method further includes:
  • the data cleaning includes at least one of data standardization, feature extraction, and elimination of duplicate values.
  • the elimination of duplicate values refers to eliminating the vehicle accident images with similarity higher than a threshold by calculating the similarity between the vehicle accident images in the vehicle accident image collection.
  • the data standardization refers to scaling the data in the data set according to a certain ratio, and mapping the data in the data set to the same plane.
  • the feature extraction refers to the use of a computer to extract image information and determine whether each image point belongs to an image feature.
  • the result of feature extraction is to divide the points on the image into different subsets, which often belong to isolated points, continuous curves or continuous regions. It is the process of selecting some of the most effective features from the original features to reduce the dimensionality of the data set, and is to improve the performance of the learning algorithm.
  • the main methods include at least: Principal Component Analysis (PCA), Linear Discriminant Analysis (Latent Dirichlet Allocation, LDA), and Singular Value Decomposition (SVD).
  • the method further includes:
  • N is the total number of convolutional surfaces
  • j is a natural number
  • i is the number of adjacent surfaces
  • k, ⁇ , and ⁇ are adjustable parameters.
  • Taking the 96-layer convolutional neural network as an example take the sum of the features of the car damage images of the 8, 9, 10, 11, and 12 layers of the convolutional neural network, and calculate the features of the car damage images in the 10th layer.
  • the method before inputting the vehicle accident image collection and the compensation amounts of the multiple vehicle accident cases into the semantic segmentation network PSPnet, the method further includes:
  • the weights and biases of the neuron are carried out by the following method: If the neuron y j and n vehicle accident images are input x 1 , x 2 ,..., x n, the corresponding connection weights are w 1j , w 2j ,..., w nj , the weight is initialized, and the initialization method means that the weight w nj satisfies the following normal distribution:
  • This method can speed up the convergence speed of the neural network.
  • the method further includes:
  • the data expansion refers to the operation of rotation, flip transformation, translation transformation, scale transformation, color change, and scaling transformation on the vehicle accident image collection.
  • Data expansion is achieved through the following mathematical formula:
  • x and y refer to the pixel coordinates of the vehicle accident image
  • refers to the angle of rotation
  • a refers to the unit distance of translation along the x axis of the vehicle accident image pixel
  • b refers to the unit distance of translation along the y axis of the vehicle accident image.
  • the data in the vehicle accident image collection can be increased.
  • the image data in the vehicle accident image collection will increase, so as to increase the amount of data in the vehicle accident image collection and improve the accuracy of model prediction.
  • the sending an instruction message to the client, and after the client receives the instruction message, the method further includes:
  • the client terminal prompts the user to take a car damage photo, and the step of taking the car damage photo further includes:
  • the frame of each shooting requirement is displayed, and when the corresponding frame is triggered, the client is triggered to start the camera application.
  • the user After receiving the instruction message, the user is prompted to take a photo of the car damage.
  • the client After taking a photo, the client preprocesses the photo.
  • the preprocessing refers to extracting vehicle characteristic information such as the front, rear, headlights, taillights, and rearview mirrors, and compressing the photos.
  • the client judges whether they meet the shooting preset rules, and if they do not meet the preset rules, the client prompts the user to take another shot. If the rule is preset, the damaged photos of the user will be taken and uploaded to the server.
  • the preset rule is used to filter car damage photos that meet the resolution higher than the threshold, the image size meets the threshold, and the image format is jpg or png.
  • the construction of the damage assessment strategy means that the vehicle damage assessment system determines the damage assessment information of the accident vehicle according to the acquired image information of the damaged parts of the vehicle, and determines the damage assessment information of the vehicle and the target PSPnet.
  • the accounting rules calculate the fixed loss amount.
  • the sending and sending the loss determination strategy to the client the method further includes:
  • the client will send a reminder that this case cannot be compensated by the self-assistant; for example, the case with the maximum compensation amount of 5000 cannot be paid by the client Self-service processing, when the customer declares, the case will be directly transferred to the claims adjuster for processing.
  • the straight person service refers to the customer self-assisted compensation process, after the client receives the user's request for help signal, the client sends the user request to the server, and the server establishes communication between the clients after receiving the request. In this way, the claim adjuster can answer questions online, and then solve the customer's questions.
  • the self-service claim settlement process adds AI self-service report rules.
  • the AI self-service report rule refers to sending user information to the server when receiving a user’s loss determination request on the client.
  • the server determines whether the user information meets the report rules, and if the rules are met, the client is ordered Send photos, if not satisfied, the client is not allowed to complete the self-assisted compensation.
  • a schematic structural diagram of a vehicle damage assessment device 20 is applicable to vehicle damage assessment.
  • the device for determining vehicle damage in the embodiment of the present application can implement steps corresponding to the method for determining vehicle damage performed in the embodiment corresponding to FIG. 1.
  • the functions implemented by the vehicle damage assessment device 20 can be implemented by hardware, or implemented by hardware executing corresponding software.
  • the hardware or software includes one or more modules corresponding to the above functions, and the modules may be software and/or hardware.
  • the vehicle damage assessment device may include an input/output module 201 and a processing module 202.
  • the input output module 201 can be used to control the input, output, and acquisition operations of the input output module 201.
  • the input and output module 201 may be used to obtain a collection of vehicle accident images to be processed and the amount of compensation for multiple vehicle accident cases.
  • Each vehicle accident case corresponds to at least one vehicle accident image.
  • the processing module 202 can be used for
  • the set of vehicle accident images and the compensation amounts of the multiple vehicle accident cases are input into the semantic segmentation network PSPnet to train the PSPnet under multiple sets of hyperparameters.
  • the error of the PSPnet under different hyperparameters is calculated through the loss function and the regularization coefficient, and the PSPnet under a set of hyperparameters with the smallest error is taken as the target PSPnet.
  • the function value of the loss function refers to the difference between the actual payout amount after the vehicle accident image in the PSPnet test set is input to the convolutional neural network and the expected payout amount after the vehicle accident image in the test set is input to the convolutional neural network
  • the sum of squares; the regularization coefficient C is calculated by the following formula:
  • C 0 is the initial value of the loss function
  • n' refers to the number of images in the vehicle accident image collection
  • refers to the ratio of the regular term to the C 0 term
  • the regular term refers to The sum of squares.
  • the processing module 202 is further configured to:
  • the data cleaning includes at least one of data standardization, feature extraction, and elimination of duplicate values.
  • the elimination of duplicate values refers to eliminating the vehicle accident images with similarity higher than a threshold by calculating the similarity between the vehicle accident images in the vehicle accident image collection.
  • the data standardization is achieved by the following mathematical formula:
  • N is the total number of convolutional surfaces
  • j is a natural number
  • i is the number of adjacent surfaces
  • k, ⁇ , and ⁇ are adjustable parameters.
  • the processing module 202 is further configured to:
  • the weights and biases of the neuron are carried out by the following method: If the neuron y j and n vehicle accident images are input x 1 , x 2 ,..., x n, the corresponding connection weights are w 1j , w 2j ,..., w nj , the weight is initialized, and the initialization method means that the weight w nj satisfies the following normal distribution:
  • the processing module 202 is further configured to:
  • the data expansion refers to the operation of rotation, flip transformation, translation transformation, scale transformation, color change, and scaling transformation on the vehicle accident image collection.
  • Data expansion is achieved through the following mathematical formula:
  • x and y refer to the pixel coordinates of the vehicle accident image
  • refers to the angle of rotation
  • a refers to the unit distance of translation along the x axis of the vehicle accident image pixel
  • b refers to the unit distance of translation along the y axis of the vehicle accident image.
  • the processing module 202 is further configured to:
  • the client terminal prompts the user to take a car damage photo, and the step of taking the car damage photo further includes:
  • the frame of each shooting requirement is displayed, and when the corresponding frame is triggered, the client is triggered to start the camera application.
  • the user After receiving the instruction message, the user is prompted to take a photo of the car damage.
  • the client After taking a photo, the client preprocesses the photo.
  • the preprocessing refers to extracting vehicle characteristic information such as the front, rear, headlights, taillights, and rearview mirrors, and compressing the photos.
  • the client judges whether they meet the shooting preset rules, and if they do not meet the preset rules, the client prompts the user to take another shot. If the rule is preset, the damaged photos of the user will be taken and uploaded to the server.
  • the preset rule is used to filter car damage photos that meet the resolution higher than the threshold, the image size meets the threshold, and the image format is jpg or png.
  • the construction of the damage assessment strategy means that the vehicle damage assessment system determines the damage assessment information of the accident vehicle according to the acquired image information of the damaged parts of the vehicle, and determines the damage assessment information of the vehicle and the target PSPnet.
  • the accounting rules calculate the fixed loss amount.
  • the processing module 202 is further configured to:
  • the straight person service refers to the customer self-assisted compensation process, after the client receives the user's request for help signal, the client sends the user request to the server, and the server establishes communication between the clients after receiving the request.
  • the self-service claim settlement process adds AI self-service report rules.
  • the AI self-service report rule refers to sending user information to the server when receiving a user’s loss determination request on the client.
  • the server determines whether the user information meets the report rules, and if the rules are met, the client is ordered Send photos, if not satisfied, the client is not allowed to complete the self-assisted compensation.
  • the above describes the creation device in the embodiment of the present application from the perspective of modular functional entities.
  • a computer device from the perspective of hardware, as shown in FIG. 3, which includes: a processor, a memory, an input and output unit (or Is a transceiver, not identified in FIG. 3) and a computer program stored in the memory and running on the processor.
  • the computer program may be a program corresponding to the method for determining vehicle damage in the embodiment corresponding to FIG. 1.
  • the processor executes the computer program to realize the execution by the vehicle damage assessment device 20 in the embodiment corresponding to FIG. 2 The steps in the method of vehicle damage assessment.
  • the processor executes the computer program
  • the function of each module in the vehicle damage assessment device 20 of the embodiment corresponding to FIG. 2 is realized.
  • the computer program may be a program corresponding to the method for determining vehicle damage in the embodiment corresponding to FIG. 1.
  • the so-called processor can be a central processing unit (Central Processing Unit, CPU), other general-purpose processors, digital signal processors (Digital Signal Processor, DSP), application specific integrated circuits (Application Specific Integrated Circuit, ASIC), ready-made Field-Programmable Gate Array (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc.
  • the general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.
  • the processor is the control center of the computer device, and various interfaces and lines are used to connect various parts of the entire computer device.
  • the memory may be used to store the computer program and/or module, and the processor implements the computer by running or executing the computer program and/or module stored in the memory, and calling data stored in the memory.
  • the memory may mainly include a storage program area and a storage data area, where the storage program area may store an operating system, an application program required by at least one function (such as a sound playback function, an image playback function, etc.); the storage data area may store Data created based on the use of mobile phones (such as audio data, video data, etc.), etc.
  • the memory may include high-speed random access memory, and may also include non-volatile memory, such as hard disks, memory, plug-in hard disks, smart media cards (SMC), and secure digital (SD) cards , Flash Card, at least one magnetic disk storage device, flash memory device, or other volatile solid-state storage device.
  • non-volatile memory such as hard disks, memory, plug-in hard disks, smart media cards (SMC), and secure digital (SD) cards , Flash Card, at least one magnetic disk storage device, flash memory device, or other volatile solid-state storage device.
  • the input and output units may also be replaced by receivers and transmitters, and they may be the same or different physical entities. When they are the same physical entity, they can be collectively referred to as input and output units.
  • the input and output can be a transceiver.
  • the memory may be integrated in the processor, or may be provided separately from the processor.
  • the method of the above embodiments can be implemented by means of software plus the necessary general hardware platform. Of course, it can also be implemented by hardware, but in many cases the former is better. ⁇ Based on this understanding, the technical solution of this application essentially or the part that contributes to the existing technology can be embodied in the form of a software product.
  • the computer software product is stored in a storage medium (such as ROM/RAM), including Several instructions are used to make a terminal (which may be a mobile phone, a computer, a server, or a network device, etc.) execute the method described in each embodiment of the present application.
  • the computer-readable storage medium may be non-volatile or volatile.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Business, Economics & Management (AREA)
  • Evolutionary Computation (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • General Engineering & Computer Science (AREA)
  • Health & Medical Sciences (AREA)
  • General Health & Medical Sciences (AREA)
  • Software Systems (AREA)
  • Evolutionary Biology (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Biophysics (AREA)
  • Mathematical Physics (AREA)
  • Biomedical Technology (AREA)
  • Computational Linguistics (AREA)
  • Accounting & Taxation (AREA)
  • Finance (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Economics (AREA)
  • Marketing (AREA)
  • Strategic Management (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • General Business, Economics & Management (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Technology Law (AREA)
  • Development Economics (AREA)
  • Multimedia (AREA)
  • Human Resources & Organizations (AREA)
  • Primary Health Care (AREA)
  • Tourism & Hospitality (AREA)
  • Image Analysis (AREA)
  • Financial Or Insurance-Related Operations Such As Payment And Settlement (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

一种车辆定损的方法、装置、设备和存储介质。该方法包括:获取待处理的车辆事故图像集合以及多个车辆事故案件的理赔金额(101);将车辆事故图像集合以及所述多个车辆事故案件的理赔金额输入搭建的语义分割网络PSPnet卷积神经网络模型,以在多组训练多组超参数下的训练PSPnet(102);通过损失函数以及正则化系数计算不同超参数下所述PSPnet的误差评估每一组超参数下卷积神经网络模型的误差,将误差最小的一组超参数下的所述PSPnet作为目标卷积神经网络模型PSPnet(103);接收到用户在终端上发送的定损请求,向客户端发送指示消息(104);接收所述客户端发送的车损照片后,则将客户端发送的车损照片输入至目标PSPnet,构建生成定损策略,并将定损策略发送给所述客户端(105)。将AI图片识别定损技术运用在车险事故理赔场景,提供便捷的客户端,让用户可以快速方便的完成理赔的全部流程。

Description

车辆定损方法、装置、设备和存储介质
本申请要求于2019年07月23日提交中国专利局、申请号为201910667158.7,发明名称为“车辆定损方法、装置、设备和存储介质”的中国专利申请的优先权,其全部内容通过引用结合在本申请中。
技术领域
本申请涉及图像检测领域,尤其涉及一种车辆定损的方法、装置、设备和存储介质。
背景技术
目前国内小汽车的总量在不断增多,中小型的交通意外越来越频发,发明人意识到,当客户发生小型的交通意外时,需要浪费大量的时间等待理赔人员确认以及要走繁杂的流程才能将理赔流程走完。这个过程中还需要理赔人员进行相关的鉴定,确认出赔偿的方案,导致处理的效率低下。
发明内容
本申请实例中提出一种新的理赔服务模式,将AI图片识别定损技术运用在车险事故理赔场景,提供便捷的客户端,让用户可以快速方便的完成理赔的全部流程。
第一方面,本申请提供一种车辆定损方法,包括:
获取待处理的车辆事故图像集合以及多个车辆事故案件的理赔金额。每个所述车辆事故案件对应至少一张所述车辆事故图像。
将所述车辆事故图像集合以及所述多个车辆事故案件的理赔金额输入语义分割网络PSPnet,以在多组超参数下训练所述PSPnet;训练所述PSPnet中的池化层以及卷积层通过
Figure PCTCN2020098767-appb-000001
计算。其中,
Figure PCTCN2020098767-appb-000002
代表根据所述PSPnet卷积层的多层感知器中第n-1层的输出,训练所述PSPnet卷积层的多层感知器中第n层中第k个神经元得到的权值,
Figure PCTCN2020098767-appb-000003
表示相应的偏置,
Figure PCTCN2020098767-appb-000004
表示第i个车辆事故案件的第j张车辆事故图像输入至所述PSPnet后在所述PSPnet的第n层的输出,i、j以及k为任意正整数,n为自然数;当n为0时,
Figure PCTCN2020098767-appb-000005
是指所述车辆事故图像。当n为所述PSPnet的最后一层时,
Figure PCTCN2020098767-appb-000006
是指所述车辆事故案件的理赔金额。
通过损失函数以及正则化系数计算不同超参数下所述PSPnet的误差,将误差最小的一组超参数下的所述PSPnet作为目标PSPnet;所述损失函数的函数值是指PSPnet测试集中的车辆事故图像输入至卷积神经网络后的实际赔付金额与测试集中的车辆事故图像输入至卷积神经网络后的期望赔付金额之间差值的平方和;所述正则化系数C通过以下公式计算得到:
Figure PCTCN2020098767-appb-000007
C 0是所述损失函数的初始取值,n'是指所述车辆事故图像集合中的图像数量;λ是指正则项与C 0项的比重,所述正则项是指
Figure PCTCN2020098767-appb-000008
的平方和。
接收到用户在终端上发送的定损请求,向客户端发送指示消息。
接收所述客户端发送的车损照片后,则将所述客户端发送的车损照片输入至所述目标 PSPnet,生成定损策略,并将所述定损策略发送给所述客户端。
第二方面,本申请提供一种车辆定损的装置,具有实现对应于上述第一方面提供的车辆定损的平台的方法的功能。所述功能可以通过硬件实现,也可以通过硬件执行相应的软件实现。硬件或软件包括一个或多个与上述功能相对应的模块,所述模块可以是软件和/或硬件。
所述车辆定损的装置包括:
获取模块,获取待处理的车辆事故图像集合以及多个车辆事故案件的理赔金额。每个所述车辆事故案件对应至少一张所述车辆事故图像。
处理模块,将所述车辆事故图像集合以及所述多个车辆事故案件的理赔金额输入语义分割网络PSPnet,以在多组超参数下训练所述PSPnet;训练所述PSPnet中的池化层以及卷积层通过
Figure PCTCN2020098767-appb-000009
计算。其中,
Figure PCTCN2020098767-appb-000010
代表根据所述PSPnet卷积层的多层感知器中第n-1层的输出,训练所述PSPnet卷积层的多层感知器中第n层中第k个神经元得到的权值,
Figure PCTCN2020098767-appb-000011
表示相应的偏置,
Figure PCTCN2020098767-appb-000012
表示第i个车辆事故案件的第j张车辆事故图像输入至所述PSPnet后在所述PSPnet的第n层的输出,i、j以及k为任意正整数,n为自然数;当n为0时,
Figure PCTCN2020098767-appb-000013
是指所述车辆事故图像。当n为所述PSPnet的最后一层时,
Figure PCTCN2020098767-appb-000014
是指所述车辆事故案件的理赔金额。
通过损失函数以及正则化系数计算不同超参数下所述PSPnet的误差,将误差最小的一组超参数下的所述PSPnet作为目标PSPnet;所述损失函数的函数值是指PSPnet测试集中的车辆事故图像输入至卷积神经网络后的实际赔付金额与测试集中的车辆事故图像输入至卷积神经网络后的期望赔付金额之间差值的平方和;所述正则化系数C通过以下公式计算得到:
Figure PCTCN2020098767-appb-000015
C 0是所述损失函数的初始取值,n'是指所述车辆事故图像集合中的图像数量;λ是指正则项与C 0项的比重,所述正则项是指
Figure PCTCN2020098767-appb-000016
的平方和。
接收到用户在终端上发送的定损请求,向客户端发送指示消息。
接收所述客户端发送的车损照片后,则将所述客户端发送的车损照片输入至所述目标PSPnet,生成定损策略,并将所述定损策略发送给所述客户端。
本申请又一方面提供了一种车辆定损的设备,其包括至少一个连接的处理器、存储器、输入输出单元,其中,所述存储器用于存储程序代码,所述处理器用于调用所述存储器中存储的程序代码来执行以下步骤:
获取待处理的车辆事故图像集合以及多个车辆事故案件的理赔金额;每个所述车辆事故案件对应至少一张所述车辆事故图像;
将所述车辆事故图像集合以及所述多个车辆事故案件的理赔金额输入语义分割网络PSPnet,以在多组超参数下训练所述PSPnet;训练所述PSPnet中的池化层以及卷积层通过
Figure PCTCN2020098767-appb-000017
计算;其中,
Figure PCTCN2020098767-appb-000018
代表根据所述PSPnet卷积层的多层感知器中第n-1层的输出,训练所述PSPnet卷积层的多层感知器中第n层中第k个神经元得到的权值,
Figure PCTCN2020098767-appb-000019
表示相应的偏置,
Figure PCTCN2020098767-appb-000020
表示第i个车辆事故案件的第 j张车辆事故图像输入至所述PSPnet后在所述PSPnet的第n层的输出,i、j以及k为任意正整数,n为自然数;当n为0时,
Figure PCTCN2020098767-appb-000021
是指所述车辆事故图像;当n为所述PSPnet的最后一层时,
Figure PCTCN2020098767-appb-000022
是指所述车辆事故案件的理赔金额;
通过损失函数以及正则化系数计算不同超参数下所述PSPnet的误差,将误差最小的一组超参数下的所述PSPnet作为目标PSPnet;所述损失函数的函数值是指PSPnet测试集中的车辆事故图像输入至卷积神经网络后的实际赔付金额与测试集中的车辆事故图像输入至卷积神经网络后的期望赔付金额之间差值的平方和;所述正则化系数C通过以下公式计算得到:
Figure PCTCN2020098767-appb-000023
C 0是所述损失函数的初始取值,n'是指所述车辆事故图像集合中的图像数量;λ是指正则项与C 0项的比重,所述正则项是指
Figure PCTCN2020098767-appb-000024
的平方和;
接收到用户在终端上发送的定损请求,向客户端发送指示消息;
接收所述客户端发送的车损照片后,则将所述客户端发送的车损照片输入至所述目标PSPnet,生成定损策略,并将所述定损策略发送给所述客户端。
本申请又一方面提供了一种计算机存储介质,所述计算机可读存储介质存储有计算机程序,所述计算机程序包括程序指令,所述程序指令被处理器执行时,用于实现以下步骤:
获取待处理的车辆事故图像集合以及多个车辆事故案件的理赔金额;每个所述车辆事故案件对应至少一张所述车辆事故图像;
将所述车辆事故图像集合以及所述多个车辆事故案件的理赔金额输入语义分割网络PSPnet,以在多组超参数下训练所述PSPnet;训练所述PSPnet中的池化层以及卷积层通过
Figure PCTCN2020098767-appb-000025
计算;其中,
Figure PCTCN2020098767-appb-000026
代表根据所述PSPnet卷积层的多层感知器中第n-1层的输出,训练所述PSPnet卷积层的多层感知器中第n层中第k个神经元得到的权值,
Figure PCTCN2020098767-appb-000027
表示相应的偏置,
Figure PCTCN2020098767-appb-000028
表示第i个车辆事故案件的第j张车辆事故图像输入至所述PSPnet后在所述PSPnet的第n层的输出,i、j以及k为任意正整数,n为自然数;当n为0时,
Figure PCTCN2020098767-appb-000029
是指所述车辆事故图像;当n为所述PSPnet的最后一层时,
Figure PCTCN2020098767-appb-000030
是指所述车辆事故案件的理赔金额;
通过损失函数以及正则化系数计算不同超参数下所述PSPnet的误差,将误差最小的一组超参数下的所述PSPnet作为目标PSPnet;所述损失函数的函数值是指PSPnet测试集中的车辆事故图像输入至卷积神经网络后的实际赔付金额与测试集中的车辆事故图像输入至卷积神经网络后的期望赔付金额之间差值的平方和;所述正则化系数C通过以下公式计算得到:
Figure PCTCN2020098767-appb-000031
C 0是所述损失函数的初始取值,n'是指所述车辆事故图像集合中的图像数量;λ是指正则项与C 0项的比重,所述正则项是指
Figure PCTCN2020098767-appb-000032
的平方和;
接收到用户在终端上发送的定损请求,向客户端发送指示消息;
接收所述客户端发送的车损照片后,则将所述客户端发送的车损照片输入至所述目标PSPnet,生成定损策略,并将所述定损策略发送给所述客户端。
相较于现有技术,本申请提出一种新的理赔服务模式,将AI图片识别定损技术运用在车险事故理赔场景,提供好车主加小程序等便捷客户端,旨在客户能够通过简单的报案登记、拍照上传损伤部位、确认赔达联系专付方案、录入支付账户等操作,即可完成理赔全流程。此外,考虑客户操作过程有可能存在疑问,提供直人服务,客户自助理赔过程中随时可联系理赔员寻求帮助。同时从风险管控角度出发,AI自助报案规则,筛选过滤风险案件;增加案件赔偿金额上限控制,超额案件强制转化,由理赔员介入。
附图说明
图1为本申请实施例中车辆定损的方法的流程示意图;
图2为本申请实施例中车辆定损的装置的结构示意图;
图3为本申请实施例中计算机设备的结构示意图。
具体实施方式
请参照图1,以下对本申请提供一种车辆定损的方法进行举例说明,所述方法包括:
101、获取待处理的车辆事故图像集合以及多个车辆事故案件的理赔金额。
每个所述车辆事故案件对应至少一张所述车辆事故图像。
本申请中,所述车辆事故图像集合包含多张已标注的车辆事故图像。车辆事故图像标注区域包括车头、车尾、大灯、车尾灯以及后视镜等车辆特征信息。将车辆事故图像集合分割为测试集和训练集。所述车辆事故是指机动车在道路上发生的,不涉及人员伤亡,仅造成价值5000元以下的财产损失的交通事故,如仅车身前后保险杠,车灯,引擎盖,门窗,后视镜等外部部件损坏,车辆可以继续驾驶的事故。
102、将所述车辆事故图像集合以及所述多个车辆事故案件的理赔金额输入语义分割网络PSPnet,以在多组超参数下训练PSPnet。
训练所述PSPnet中的池化层以及卷积层通过
Figure PCTCN2020098767-appb-000033
计算;其中,
Figure PCTCN2020098767-appb-000034
代表根据所述PSPnet卷积层的多层感知器中第n-1层的输出,训练所述PSPnet卷积层的多层感知器中第n层中第k个神经元得到的权值,
Figure PCTCN2020098767-appb-000035
表示相应的偏置,
Figure PCTCN2020098767-appb-000036
表示第i个车辆事故案件的第j张车辆事故图像输入至所述PSPnet后在所述PSPnet的第n层的输出,i、j以及k为任意正整数,n为自然数;当n为0时,
Figure PCTCN2020098767-appb-000037
是指所述车辆事故图像;当n为所述PSPnet的最后一层时,
Figure PCTCN2020098767-appb-000038
是指所述车辆事故案件的理赔金额;
所述超参数至少包括卷积核的尺寸、学习率η、batch参数、神经网络层数、激活函数、优化器、批大小以及训练的epoch数量中的至少一项。
所述超参数通过自定义的方式调整。通过调整所述超参数,得到多组超参数,分别计算每组超参数下的测试值和理论值,该测试值和理论值可以计算模型的误差,通过所述误差对不同参数下的神经网络模型进行评估,以获得误差最小的神经网络模型参数。
103、通过损失函数以及正则化系数计算不同超参数下所述PSPnet的误差,将误差最小的一组超参数下的所述PSPnet作为目标PSPnet。
所述损失函数的函数值是指PSPnet测试集中的车辆事故图像输入至卷积神经网络后的实际赔付金额与测试集中的车辆事故图像输入至卷积神经网络后的期望赔付金额之间差值的平方和;所述正则化系数C通过以下公式计算得到:
Figure PCTCN2020098767-appb-000039
C 0是所述损失函数的初始取值,n'是指所述车辆事故图像集合中的图像数量;λ是指正则项与C 0项的比重,所述正则项是指
Figure PCTCN2020098767-appb-000040
的平方和
通过加入正则化解决模型过拟合的问题,并提高模型的泛化能力。
104、接收到用户在终端上发送的定损请求,向客户端发送指示消息。
所述指示消息是指用于命令客户端发送车损照片。所述客户端是指提供客户自助理赔服务的程序。
其中,通过客户端携带的手机号或微信号识别投保人车牌,并通过GPS定位出险地点,根据手机号或者微信号获取关联车辆信息、关联保单信息,并在根据获取的车辆信息以及关联保单信息判断可以办理自助理赔时,自动生成报案号从而自助完成理赔流程。所述理赔流程是指报案登记、拍照上传损伤部位、确认赔达联系专付方案以及录入支付账户等流程。
105、接收所述客户端发送的车损照片后,则将所述客户端发送的车损照片输入至所述目标PSPnet,生成定损策略,并将所述定损策略发送给所述客户端。
一些实施方式中,获取待处理的车辆事故图像集合以及多个车辆事故案件的理赔金额之后,所述方法还包括:
对收集的车辆事故集合进行数据清洗。所述数据清洗至少包括数据标准化、特征提取和消除重复值中的至少一项。所述消除重复值是指通过计算车辆事故图像集合的车辆事故图像之间的相似度,将相似度高于阈值的车辆事故图像进行剔除。
所述数据标准化是指将数据集中的数据按照一定比例缩放,将所述数据集中的数据映射到一个同一个平面中。所述特征提取指的是使用计算机提取图像信息,决定每个图像的点是否属于一个图像特征。特征提取的结果是把图像上的点分为不同的子集,这些子集往往属于孤立的点、连续的曲线或者连续的区域。是从原始特征中选择出一些最有效特征以降低数据集维度的过程,是提高学习算法性能。其中主要的方法至少包括:主成分分析法(Principal Component Analysis,PCA),线性判别分析(Latent Dirichlet Allocation,LDA),奇异值分解(Singular Value Decomposition,SVD)。
一些实施方式中,所述数据标准化,所述方法还包括:
所述数据标准化通过以下数学公式实现:
Figure PCTCN2020098767-appb-000041
所述
Figure PCTCN2020098767-appb-000042
是指第j个卷积面上在位置(x,y)的值,响应标准化的值,
Figure PCTCN2020098767-appb-000043
是通过若干相邻卷积面在位置(x,y)的值计算,N是卷积面的总数,j是自然数,i是相邻面的个数,k、α、β是可调参数。
以卷积神经网络有96层为例,取卷积神经网络的8、9、10、11以及12层的车损图像的特征之和,并计算第10层中车损图像的特征在所述特征之和的占比。当i=10,N=96, n=4,k=2,α=e -4以及β=0.75时,该获取车辆事故图像中的像素数据
Figure PCTCN2020098767-appb-000044
将所述像素数据导入公式(1)中,得到
Figure PCTCN2020098767-appb-000045
是车损图像的特征在所述特征之和的占比,
Figure PCTCN2020098767-appb-000046
以增加卷积神经网络的训练速度和卷积神经网络的精确度。
一些实施方式中,所述将所述车辆事故图像集合以及所述多个车辆事故案件的理赔金额输入语义分割网络PSPnet之前,所述方法还包括:
神经元的权值和偏置的通过以下方法进行:若神经元y j与n个车辆事故图像输入x 1,x 2,…,x n对应连接的权值为w 1j,w 2j,…,w nj,则对权值进行初始化,所述初始化的方法是指权值w nj满足以下正太分布:
Figure PCTCN2020098767-appb-000047
其中b为常数。
通过此方法可以加快神经网络的收敛速度。
一些实施方式中,所述获取待处理的车辆事故图像集合以及多个车辆事故案件的理赔金额之后,所述方法还包括:
还可以对获得车辆事故图像集合进行数据扩充,所述数据扩充是指对车辆事故图像集合旋转、翻转变换、平移变换、尺度变换颜色变化、以及缩放变换操作。通过以下数学公式实现数据扩充:
Figure PCTCN2020098767-appb-000048
其中x,y是指车辆事故图像的像素坐标,θ是指旋转的角度,a是指沿车辆事故图像像素x轴平移的单位距离,b是指车辆事故图像像素沿y轴平移的单位距离。通过此公式可以对车辆事故图像进行翻转,旋转和平移,以增加车辆事故图像集合里车辆事故图像数据。
通过上述实施方式,可以增加车辆事故图像集合中的数据。车辆事故图像集合经过所述数据扩充以后,车辆事故图像集合里的图像数据将会增多,以提高车辆事故图像集合里数据的数量从而提高模型预测的准确率。
一些实施方式中,所述向客户端发送指示消息,所述客户端在收到所述指示消息之后,所述方法还包括:
所述客户端提示用户拍摄车损照片,所述拍摄车损照片的步骤还包括:
在提示用户拍摄车损照片时,显示各个拍摄要求的图框,在相应图框被触发时,触发所述客户端开启相机应用。
在收到所述指示消息后,提示用户拍摄车损照片。
在拍摄完一张照片时,所述客户端对所述照片预处理。所述预处理是指提取车头、车尾、大灯、车尾灯以及后视镜等车辆特征信息以及对所述照片进行压缩。
对完成预处理的照片通过所述客户端判断是否符合拍摄预设规则,如不符合预设规则,所述客户端提示用户重新拍摄。如预设规则,则对用户拍摄车损照片上传至服务端。所述预设规则是指用于筛选满足分辨率高于阈值、图片尺寸满足阈值以及图片格式为jpg或png的车损照片。
所述构建定损策略是指根据车辆定损系统根据获取到的所述车辆受损部件的图像信息确定事故车辆的定损信息,并根据定所述车辆的定损信息以及所述目标PSPnet的核算规则计算定损金额。
一些实施方式中,所述并将所述定损策略发送给所述客户端,所述方法还包括:
设置赔偿金额上限控制,若检测到目标神经网络模型的核算规则计算的定损金额超额过设置赔偿金额,则让客户端发出提示此案件无法自助理赔;例如设置赔偿金额上限5000的案子无法被客户自助处理,当客户申报的时候,案件将直接移交给理赔员进行处理。
若客户操作过程有可能存在疑问,则提供直人服务。所述直人服务是指客户自助理赔过程中,客户端收到用户发出请求帮助信号后,向服务端发送用户请求,服务端在接收到请求后,将所述客户端之间建立通讯。从而实现理赔员在线的答疑,进而解决客户的疑问。
自助处理理赔流程重增加AI自助报案规则。所述AI自助报案规则是指在接收到用户在客户端的定损请求时,将用户信息发送至服务端,所述服务端判断用户信息是否满足报案规则,若满足规则,则命令所述客户端发送照片,若不满足则不允许客户端完成自助理赔。
如图2所示的一种车辆定损的装置20的结构示意图,其可应用于车辆定损。本申请实施例中的车辆定损的装置能够实现对应于上述图1所对应的实施例中所执行的车辆定损的方法的步骤。车辆定损的装置20实现的功能可以通过硬件实现,也可以通过硬件执行相应的软件实现。硬件或软件包括一个或多个与上述功能相对应的模块,所述模块可以是软件和/或硬件。所述车辆定损的装置可包括输入输出模块201和处理模块202,所述处理模块202和输入输出模块201的功能实现可参考图1所对应的实施例中所执行的操作,此处不作赘述。输入输出模块201可用于控制所述输入输出模块201的输入、输出以及获取操作。
一些实施方式中,所述输入输出模块201可用于获取待处理的车辆事故图像集合以及多个车辆事故案件的理赔金额。每个所述车辆事故案件对应至少一张所述车辆事故图像。
所述处理模块202可用于
将所述车辆事故图像集合以及所述多个车辆事故案件的理赔金额输入语义分割网络PSPnet,以在多组超参数下训练所述PSPnet。训练所述PSPnet中的池化层以及卷积层通过
Figure PCTCN2020098767-appb-000049
计算;其中,
Figure PCTCN2020098767-appb-000050
代表根据所述PSPnet卷积层的多层感知器中第n-1层的输出,训练所述PSPnet卷积层的多层感知器中第n层中第k个神经元得到的权值,
Figure PCTCN2020098767-appb-000051
表示相应的偏置,
Figure PCTCN2020098767-appb-000052
表示第i个车辆事故案件的第 j张车辆事故图像输入至所述PSPnet后在所述PSPnet的第n层的输出,i、j以及k为任意正整数,n为自然数;当n为0时,
Figure PCTCN2020098767-appb-000053
是指所述车辆事故图像;当n为所述PSPnet的最后一层时,
Figure PCTCN2020098767-appb-000054
是指所述车辆事故案件的理赔金额。
通过损失函数以及正则化系数计算不同超参数下所述PSPnet的误差,将误差最小的一组超参数下的所述PSPnet作为目标PSPnet。所述损失函数的函数值是指PSPnet测试集中的车辆事故图像输入至卷积神经网络后的实际赔付金额与测试集中的车辆事故图像输入至卷积神经网络后的期望赔付金额之间差值的平方和;所述正则化系数C通过以下公式计算得到:
Figure PCTCN2020098767-appb-000055
C 0是所述损失函数的初始取值,n'是指所述车辆事故图像集合中的图像数量;λ是指正则项与C 0项的比重,所述正则项是指
Figure PCTCN2020098767-appb-000056
的平方和。
接收到用户在终端上发送的定损请求,向客户端发送指示消息;
接收所述客户端发送的车损照片后,则将所述客户端发送的车损照片输入至所述目标PSPnet,生成定损策略,并将所述定损策略发送给所述客户端。
一些实施方式中,所述处理模块202还用于:
对收集的车辆事故集合进行数据清洗。所述数据清洗至少包括数据标准化、特征提取和消除重复值中的至少一项。所述消除重复值是指通过计算车辆事故图像集合的车辆事故图像之间的相似度,将相似度高于阈值的车辆事故图像进行剔除。
一些实施方式中,所述数据标准化通过以下数学公式实现:
Figure PCTCN2020098767-appb-000057
所述
Figure PCTCN2020098767-appb-000058
是指第j个卷积面上在位置(x,y)的值,响应标准化的值,
Figure PCTCN2020098767-appb-000059
是通过若干相邻卷积面在位置(x,y)的值计算,N是卷积面的总数,j是自然数,i是相邻面的个数,k、α、β是可调参数。
一些实施方式中,所述处理模块202还用于:
神经元的权值和偏置的通过以下方法进行:若神经元y j与n个车辆事故图像输入x 1,x 2,…,x n对应连接的权值为w 1j,w 2j,…,w nj,则对权值进行初始化,所述初始化的方法是指权值w nj满足以下正太分布:
Figure PCTCN2020098767-appb-000060
其中b为常数。
一些实施方式中,所述处理模块202还用于:
还可以对获得车辆事故图像集合进行数据扩充,所述数据扩充是指对车辆事故图像集合旋转、翻转变换、平移变换、尺度变换颜色变化、以及缩放变换操作。通过以下数学公式实现数据扩充:
Figure PCTCN2020098767-appb-000061
其中x,y是指车辆事故图像的像素坐标,θ是指旋转的角度,a是指沿车辆事故图像像素x轴平移的单位距离,b是指车辆事故图像像素沿y轴平移的单位距离。通过此公式可以对车辆事故图像进行翻转,旋转和平移,以增加车辆事故图像集合里车辆事故图像数据。
一些实施方式中,所述处理模块202还用于:
所述客户端提示用户拍摄车损照片,所述拍摄车损照片的步骤还包括:
在提示用户拍摄车损照片时,显示各个拍摄要求的图框,在相应图框被触发时,触发所述客户端开启相机应用。
在收到所述指示消息后,提示用户拍摄车损照片。
在拍摄完一张照片时,所述客户端对所述照片预处理。所述预处理是指提取车头、车尾、大灯、车尾灯以及后视镜等车辆特征信息以及对所述照片进行压缩。
对完成预处理的照片通过所述客户端判断是否符合拍摄预设规则,如不符合预设规则,所述客户端提示用户重新拍摄。如预设规则,则对用户拍摄车损照片上传至服务端。所述预设规则是指用于筛选满足分辨率高于阈值、图片尺寸满足阈值以及图片格式为jpg或png的车损照片。
所述构建定损策略是指根据车辆定损系统根据获取到的所述车辆受损部件的图像信息确定事故车辆的定损信息,并根据定所述车辆的定损信息以及所述目标PSPnet的核算规则计算定损金额。
一些实施方式中,所述处理模块202还用于:
设置赔偿金额上限控制,若检测到目标神经网络模型的核算规则计算的定损金额超额过设置赔偿金额,则让客户端发出提示此案件无法自助理赔;
若客户操作过程有可能存在疑问,则提供直人服务。所述直人服务是指客户自助理赔过程中,客户端收到用户发出请求帮助信号后,向服务端发送用户请求,服务端在接收到请求后,将所述客户端之间建立通讯。
自助处理理赔流程重增加AI自助报案规则。所述AI自助报案规则是指在接收到用户在客户端的定损请求时,将用户信息发送至服务端,所述服务端判断用户信息是否满足报案规则,若满足规则,则命令所述客户端发送照片,若不满足则不允许客户端完成自助理赔。
上面从模块化功能实体的角度分别介绍了本申请实施例中的创建装置,以下从硬件角度介绍一种计算机设备,如图3所示,其包括:处理器、存储器、输入输出单元(也可以是收发器,图3中未标识出)以及存储在所述存储器中并可在所述处理器上运行的计算机程序。例如,该计算机程序可以为图1所对应的实施例中车辆定损的方法对应的程序。例 如,当计算机设备实现如图2所示的车辆定损的装置20的功能时,所述处理器执行所述计算机程序时实现上述图2所对应的实施例中由车辆定损的装置20执行的车辆定损的方法中的各步骤。或者,所述处理器执行所述计算机程序时实现上述图2所对应的实施例的车辆定损的装置20中各模块的功能。又例如,该计算机程序可以为图1所对应的实施例中车辆定损的方法对应的程序。
所称处理器可以是中央处理单元(Central Processing Unit,CPU),还可以是其他通用处理器、数字信号处理器(Digital Signal Processor,DSP)、专用集成电路(Application Specific Integrated Circuit,ASIC)、现成可编程门阵列(Field-Programmable Gate Array,FPGA)或者其他可编程逻辑器件、分立门或者晶体管逻辑器件、分立硬件组件等。通用处理器可以是微处理器或者该处理器也可以是任何常规的处理器等,所述处理器是所述计算机装置的控制中心,利用各种接口和线路连接整个计算机装置的各个部分。
所述存储器可用于存储所述计算机程序和/或模块,所述处理器通过运行或执行存储在所述存储器内的计算机程序和/或模块,以及调用存储在存储器内的数据,实现所述计算机装置的各种功能。所述存储器可主要包括存储程序区和存储数据区,其中,存储程序区可存储操作系统、至少一个功能所需的应用程序(比如声音播放功能、图像播放功能等)等;存储数据区可存储根据手机的使用所创建的数据(比如音频数据、视频数据等)等。此外,存储器可以包括高速随机存取存储器,还可以包括非易失性存储器,例如硬盘、内存、插接式硬盘,智能存储卡(Smart Media Card,SMC),安全数字(Secure Digital,SD)卡,闪存卡(Flash Card)、至少一个磁盘存储器件、闪存器件、或其他易失性固态存储器件。
所述输入输出单元也可以用接收器和发送器代替,可以为相同或者不同的物理实体。为相同的物理实体时,可以统称为输入输出单元。该输入输出可以为收发器。
所述存储器可以集成在所述处理器中,也可以与所述处理器分开设置。
通过以上的实施方式的描述,本领域的技术人员可以清楚地了解到上述实施例方法可借助软件加必需的通用硬件平台的方式来实现,当然也可以通过硬件,但很多情况下前者是更佳的实施方式。基于这样的理解,本申请的技术方案本质上或者说对现有技术做出贡献的部分可以以软件产品的形式体现出来,该计算机软件产品存储在一个存储介质(如ROM/RAM)中,包括若干指令用以使得一台终端(可以是手机,计算机,服务器或者网络设备等)执行本申请各个实施例所述的方法。其中,所述计算机可读存储介质可以是非易失性,也可以是易失性。
以上所述,仅为本申请的具体实施方式,但本申请的保护范围并不局限于此,任何熟悉本技术领域的技术人员在本申请揭露的技术范围内,可轻易想到变化或替换,都应涵盖在本申请的保护范围之内。因此,本申请的保护范围应以所述权利要求的保护范围为准。

Claims (20)

  1. 一种车辆定损方法,其中,所述方法包括:
    获取待处理的车辆事故图像集合以及多个车辆事故案件的理赔金额;每个所述车辆事故案件对应至少一张所述车辆事故图像;
    将所述车辆事故图像集合以及所述多个车辆事故案件的理赔金额输入语义分割网络PSPnet,以在多组超参数下训练所述PSPnet;训练所述PSPnet中的池化层以及卷积层通过
    Figure PCTCN2020098767-appb-100001
    计算;其中,
    Figure PCTCN2020098767-appb-100002
    代表根据所述PSPnet卷积层的多层感知器中第n-1层的输出,训练所述PSPnet卷积层的多层感知器中第n层中第k个神经元得到的权值,
    Figure PCTCN2020098767-appb-100003
    表示相应的偏置,
    Figure PCTCN2020098767-appb-100004
    表示第i个车辆事故案件的第j张车辆事故图像输入至所述PSPnet后在所述PSPnet的第n层的输出,i、j以及k为任意正整数,n为自然数;当n为0时,
    Figure PCTCN2020098767-appb-100005
    是指所述车辆事故图像;当n为所述PSPnet的最后一层时,
    Figure PCTCN2020098767-appb-100006
    是指所述车辆事故案件的理赔金额;
    通过损失函数以及正则化系数计算不同超参数下所述PSPnet的误差,将误差最小的一组超参数下的所述PSPnet作为目标PSPnet;所述损失函数的函数值是指PSPnet测试集中的车辆事故图像输入至卷积神经网络后的实际赔付金额与测试集中的车辆事故图像输入至卷积神经网络后的期望赔付金额之间差值的平方和;所述正则化系数C通过以下公式计算得到:
    Figure PCTCN2020098767-appb-100007
    C 0是所述损失函数的初始取值,n'是指所述车辆事故图像集合中的图像数量;λ是指正则项与C 0项的比重,所述正则项是指
    Figure PCTCN2020098767-appb-100008
    的平方和;
    接收到用户在终端上发送的定损请求,向客户端发送指示消息;
    接收所述客户端发送的车损照片后,则将所述客户端发送的车损照片输入至所述目标PSPnet,生成定损策略,并将所述定损策略发送给所述客户端。
  2. 根据权利要求1所述的方法,其中,所述获取待处理的车辆事故图像集合以及多个车辆事故案件的理赔金额之后,所述方法还包括:
    对收集的车辆事故集合进行数据清洗;所述数据清洗至少包括数据标准化、特征提取和消除重复值中的至少一项;所述消除重复值是指通过计算车辆事故图像集合的车辆事故图像之间的相似度,将相似度高于阈值的车辆事故图像进行剔除。
  3. 根据权利要求2所述的方法,其中,所述数据标准化,所述方法还包括:
    所述数据标准化通过以下数学公式实现:
    Figure PCTCN2020098767-appb-100009
    所述
    Figure PCTCN2020098767-appb-100010
    是指第j个卷积面上在位置(x,y)的值,响应标准化的值,
    Figure PCTCN2020098767-appb-100011
    是通过若干相邻卷积面在位置(x,y)的值计算,N是卷积面的总数,j是自然数,i是相邻面的个数, k、α、β是可调参数。
  4. 根据权利要求1所述的方法,其中,所述将所述车辆事故图像集合以及所述多个车辆事故案件的理赔金额输入语义分割网络PSPnet之前,所述方法还包括:
    神经元的权值和偏置的通过以下方法进行:若神经元y j与n个车辆事故图像输入x 1,x 2,…,x n对应连接的权值为w 1j,w 2j,…,w nj,则对权值进行初始化,所述初始化的方法是指权值w nj满足以下正太分布:
    Figure PCTCN2020098767-appb-100012
    其中b为常数。
  5. 根据权利要求1所述的方法,其中,所述获取待处理的车辆事故图像集合以及多个车辆事故案件的理赔金额之后,所述方法还包括:
    还可以对获得车辆事故图像集合进行数据扩充,所述数据扩充是指对车辆事故图像集合旋转、翻转变换、平移变换、尺度变换颜色变化、以及缩放变换操作;通过以下数学公式实现数据扩充:
    Figure PCTCN2020098767-appb-100013
    其中x,y是指车辆事故图像的像素坐标,θ是指旋转的角度,a是指沿车辆事故图像像素x轴平移的单位距离,b是指车辆事故图像像素沿y轴平移的单位距离,通过此公式可以对车辆事故图像进行翻转,旋转和平移,以增加车辆事故图像集合里车辆事故图像数据。
  6. 根据权利要求1所述的方法,其中,所述向客户端发送指示消息,所述客户端在收到所述指示消息之后,所述方法还包括:
    所述客户端提示用户拍摄车损照片,所述拍摄车损照片的步骤还包括:
    在提示用户拍摄车损照片时,显示各个拍摄要求的图框,在相应图框被触发时,触发所述客户端开启相机应用;
    在收到所述指示消息后,提示用户拍摄车损照片;
    在拍摄完一张照片时,所述客户端对所述照片预处理,所述预处理是指提取车头、车尾、大灯、车尾灯以及后视镜等车辆特征信息以及对所述照片进行压缩;
    对完成预处理的照片通过所述客户端判断是否符合拍摄预设规则,如不符合预设规则,所述客户端提示用户重新拍摄;如预设规则,则对用户拍摄车损照片上传至服务端;所述预设规则是指用于筛选满足分辨率高于阈值、图片尺寸满足阈值以及图片格式为jpg或png的车损照片;
    所述构建定损策略是指根据车辆定损系统根据获取到的所述车辆受损部件的图像信息确定事故车辆的定损信息,并根据定所述车辆的定损信息以及所述目标PSPnet的核算规则计算定损金额。
  7. 根据权利要求1所述的方法,其中,所述并将所述定损策略发送给所述客户端,所述方法还包括:
    设置赔偿金额上限控制,若检测到目标神经网络模型的核算规则计算的定损金额超额过设置赔偿金额,则让客户端发出提示此案件无法自助理赔;
    若客户操作过程有可能存在疑问,则提供直人服务;所述直人服务是指客户自助理赔过程中,客户端收到用户发出请求帮助信号后,向服务端发送用户请求,服务端在接收到请求后,将所述客户端之间建立通讯;
    自助处理理赔流程重增加AI自助报案规则;所述AI自助报案规则是指在接收到用户在客户端的定损请求时,将用户信息发送至服务端,所述服务端判断用户信息是否满足报案规则,若满足规则,则命令所述客户端发送照片,若不满足则不允许客户端完成自助理赔。
  8. 一种车辆定损的装置,其中,所述装置包括:
    输入输出模块,获取待处理的车辆事故图像集合以及多个车辆事故案件的理赔金额,所述车辆事故案件对应至少一张所述车辆事故图像;
    处理模块,将所述车辆事故图像集合以及所述多个车辆事故案件的理赔金额输入语义分割网络PSPnet,以在多组超参数下训练所述PSPnet;训练所述PSPnet中的池化层以及卷积层通过
    Figure PCTCN2020098767-appb-100014
    计算;其中,
    Figure PCTCN2020098767-appb-100015
    代表根据所述PSPnet卷积层的多层感知器中第n-1层的输出,训练所述PSPnet卷积层的多层感知器中第n层中第k个神经元得到的权值,
    Figure PCTCN2020098767-appb-100016
    表示相应的偏置,
    Figure PCTCN2020098767-appb-100017
    表示第i个车辆事故案件的第j张车辆事故图像输入至所述PSPnet后在所述PSPnet的第n层的输出,i、j以及k为任意正整数,n为自然数;当n为0时,
    Figure PCTCN2020098767-appb-100018
    是指所述车辆事故图像,当n为所述PSPnet的最后一层时,
    Figure PCTCN2020098767-appb-100019
    是指所述车辆事故案件的理赔金额;
    通过损失函数以及正则化系数计算不同超参数下所述PSPnet的误差,将误差最小的一组超参数下的所述PSPnet作为目标PSPnet;所述损失函数的函数值是指PSPnet测试集中的车辆事故图像输入至卷积神经网络后的实际赔付金额与测试集中的车辆事故图像输入至卷积神经网络后的期望赔付金额之间差值的平方和;所述正则化系数C通过以下公式计算得到:
    Figure PCTCN2020098767-appb-100020
    C 0是所述损失函数的初始取值,n'是指所述车辆事故图像集合中的图像数量;λ是指正则项与C 0项的比重,所述正则项是指
    Figure PCTCN2020098767-appb-100021
    的平方和;
    接收到用户在终端上发送的定损请求,向客户端发送指示消息;
    接收所述客户端发送的车损照片后,则将所述客户端发送的车损照片输入至所述目标PSPnet,生成定损策略,并将所述定损策略发送给所述客户端。
  9. 一种计算机设备,其中,所述计算机设备包括:至少一个处理器、存储器和输入输出单元,其中,所述存储器用于存储程序代码,所述处理器用于调用所述存储器中存储的程序代码来执行以下步骤:
    获取待处理的车辆事故图像集合以及多个车辆事故案件的理赔金额;每个所述车辆事故案件对应至少一张所述车辆事故图像;
    将所述车辆事故图像集合以及所述多个车辆事故案件的理赔金额输入语义分割网络PSPnet,以在多组超参数下训练所述PSPnet;训练所述PSPnet中的池化层以及卷积层通过
    Figure PCTCN2020098767-appb-100022
    计算;其中,
    Figure PCTCN2020098767-appb-100023
    代表根据所述PSPnet卷积层的多层感知器中第n-1层的输出,训练所述PSPnet卷积层的多层感知器中第n层中第k个神经元得到的权值,
    Figure PCTCN2020098767-appb-100024
    表示相应的偏置,
    Figure PCTCN2020098767-appb-100025
    表示第i个车辆事故案件的第j张车辆事故图像输入至所述PSPnet后在所述PSPnet的第n层的输出,i、j以及k为任意正整数,n为自然数;当n为0时,
    Figure PCTCN2020098767-appb-100026
    是指所述车辆事故图像;当n为所述PSPnet的最后一层时,
    Figure PCTCN2020098767-appb-100027
    是指所述车辆事故案件的理赔金额;
    通过损失函数以及正则化系数计算不同超参数下所述PSPnet的误差,将误差最小的一组超参数下的所述PSPnet作为目标PSPnet;所述损失函数的函数值是指PSPnet测试集中的车辆事故图像输入至卷积神经网络后的实际赔付金额与测试集中的车辆事故图像输入至卷积神经网络后的期望赔付金额之间差值的平方和;所述正则化系数C通过以下公式计算得到:
    Figure PCTCN2020098767-appb-100028
    C 0是所述损失函数的初始取值,n'是指所述车辆事故图像集合中的图像数量;λ是指正则项与C 0项的比重,所述正则项是指
    Figure PCTCN2020098767-appb-100029
    的平方和;
    接收到用户在终端上发送的定损请求,向客户端发送指示消息;
    接收所述客户端发送的车损照片后,则将所述客户端发送的车损照片输入至所述目标PSPnet,生成定损策略,并将所述定损策略发送给所述客户端。
  10. 根据权利要求9所述的计算机设备,其中,所述处理器用于调用所述存储器中存储的程序代码来执行以下步骤:
    对收集的车辆事故集合进行数据清洗;所述数据清洗至少包括数据标准化、特征提取和消除重复值中的至少一项;所述消除重复值是指通过计算车辆事故图像集合的车辆事故图像之间的相似度,将相似度高于阈值的车辆事故图像进行剔除。
  11. 根据权利要求10所述的计算机设备,其中,所述处理器用于调用所述存储器中存储的程序代码来执行以下步骤:
    所述数据标准化通过以下数学公式实现:
    Figure PCTCN2020098767-appb-100030
    所述
    Figure PCTCN2020098767-appb-100031
    是指第j个卷积面上在位置(x,y)的值,响应标准化的值,
    Figure PCTCN2020098767-appb-100032
    是通过若干相邻卷积面在位置(x,y)的值计算,N是卷积面的总数,j是自然数,i是相邻面的个数,k、α、β是可调参数。
  12. 根据权利要求9所述的计算机设备,其中,所述处理器用于调用所述存储器中存储的程序代码来执行以下步骤:
    神经元的权值和偏置的通过以下计算机设备进行:若神经元y j与n个车辆事故图像输入x 1,x 2,…,x n对应连接的权值为w 1j,w 2j,…,w nj,则对权值进行初始化,所述初始化的计算机设备是指权值w nj满足以下正太分布:
    Figure PCTCN2020098767-appb-100033
    其中b为常数。
  13. 根据权利要求9所述的计算机设备,其中,所述处理器用于调用所述存储器中存储的程序代码来执行以下步骤:
    对获得车辆事故图像集合进行数据扩充,所述数据扩充是指对车辆事故图像集合旋转、翻转变换、平移变换、尺度变换颜色变化、以及缩放变换操作;通过以下数学公式实现数据扩充:
    Figure PCTCN2020098767-appb-100034
    其中x,y是指车辆事故图像的像素坐标,θ是指旋转的角度,a是指沿车辆事故图像像素x轴平移的单位距离,b是指车辆事故图像像素沿y轴平移的单位距离;过此公式可以对车辆事故图像进行翻转,旋转和平移,以增加车辆事故图像集合里车辆事故图像数据。
  14. 根据权利要求9所述的计算机设备,其中,所述处理器用于调用所述存储器中存储的程序代码来执行以下步骤:
    所述客户端提示用户拍摄车损照片,所述拍摄车损照片的步骤还包括:
    在提示用户拍摄车损照片时,显示各个拍摄要求的图框,在相应图框被触发时,触发所述客户端开启相机应用;
    在收到所述指示消息后,提示用户拍摄车损照片;
    在拍摄完一张照片时,所述客户端对所述照片预处理,所述预处理是指提取车头、车尾、大灯、车尾灯以及后视镜等车辆特征信息以及对所述照片进行压缩;
    对完成预处理的照片通过所述客户端判断是否符合拍摄预设规则,如不符合预设规则,所述客户端提示用户重新拍摄;如预设规则,则对用户拍摄车损照片上传至服务端;所述预设规则是指用于筛选满足分辨率高于阈值、图片尺寸满足阈值以及图片格式为jpg或png的车损照片;
    所述构建定损策略是指根据车辆定损系统根据获取到的所述车辆受损部件的图像信息确定事故车辆的定损信息,并根据定所述车辆的定损信息以及所述目标PSPnet的核算规则计算定损金额。
  15. 根据权利要求9所述的计算机设备,其中,所述处理器用于调用所述存储器中存储的程序代码来执行以下步骤:
    设置赔偿金额上限控制,若检测到目标神经网络模型的核算规则计算的定损金额超额 过设置赔偿金额,则让客户端发出提示此案件无法自助理赔;
    若客户操作过程有可能存在疑问,则提供直人服务;所述直人服务是指客户自助理赔过程中,客户端收到用户发出请求帮助信号后,向服务端发送用户请求,服务端在接收到请求后,将所述客户端之间建立通讯;
    自助处理理赔流程重增加AI自助报案规则;所述AI自助报案规则是指在接收到用户在客户端的定损请求时,将用户信息发送至服务端,所述服务端判断用户信息是否满足报案规则,若满足规则,则命令所述客户端发送照片,若不满足则不允许客户端完成自助理赔。
  16. 一种计算机可读存储介质,其中,所述计算机可读存储介质存储有计算机程序,所述计算机程序包括程序指令,所述程序指令被处理器执行时,用于实现以下步骤:
    获取待处理的车辆事故图像集合以及多个车辆事故案件的理赔金额;每个所述车辆事故案件对应至少一张所述车辆事故图像;
    将所述车辆事故图像集合以及所述多个车辆事故案件的理赔金额输入语义分割网络PSPnet,以在多组超参数下训练所述PSPnet;训练所述PSPnet中的池化层以及卷积层通过
    Figure PCTCN2020098767-appb-100035
    计算;其中,
    Figure PCTCN2020098767-appb-100036
    代表根据所述PSPnet卷积层的多层感知器中第n-1层的输出,训练所述PSPnet卷积层的多层感知器中第n层中第k个神经元得到的权值,
    Figure PCTCN2020098767-appb-100037
    表示相应的偏置,
    Figure PCTCN2020098767-appb-100038
    表示第i个车辆事故案件的第j张车辆事故图像输入至所述PSPnet后在所述PSPnet的第n层的输出,i、j以及k为任意正整数,n为自然数;当n为0时,
    Figure PCTCN2020098767-appb-100039
    是指所述车辆事故图像;当n为所述PSPnet的最后一层时,
    Figure PCTCN2020098767-appb-100040
    是指所述车辆事故案件的理赔金额;
    通过损失函数以及正则化系数计算不同超参数下所述PSPnet的误差,将误差最小的一组超参数下的所述PSPnet作为目标PSPnet;所述损失函数的函数值是指PSPnet测试集中的车辆事故图像输入至卷积神经网络后的实际赔付金额与测试集中的车辆事故图像输入至卷积神经网络后的期望赔付金额之间差值的平方和;所述正则化系数C通过以下公式计算得到:
    Figure PCTCN2020098767-appb-100041
    C 0是所述损失函数的初始取值,n'是指所述车辆事故图像集合中的图像数量;λ是指正则项与C 0项的比重,所述正则项是指
    Figure PCTCN2020098767-appb-100042
    的平方和;
    接收到用户在终端上发送的定损请求,向客户端发送指示消息;
    接收所述客户端发送的车损照片后,则将所述客户端发送的车损照片输入至所述目标PSPnet,生成定损策略,并将所述定损策略发送给所述客户端。
  17. 根据权利要求16所述的计算机可读存储介质,其中,所述程序指令被处理器执行时,还用于实现以下步骤:
    对收集的车辆事故集合进行数据清洗;所述数据清洗至少包括数据标准化、特征提取和消除重复值中的至少一项;所述消除重复值是指通过计算车辆事故图像集合的车辆事故图像之间的相似度,将相似度高于阈值的车辆事故图像进行剔除。
  18. 根据权利要求17所述的计算机可读存储介质,其中,所述程序指令被处理器执行时,还用于实现以下步骤:
    所述数据标准化通过以下数学公式实现:
    Figure PCTCN2020098767-appb-100043
    所述
    Figure PCTCN2020098767-appb-100044
    是指第j个卷积面上在位置(x,y)的值,响应标准化的值,
    Figure PCTCN2020098767-appb-100045
    是通过若干相邻卷积面在位置(x,y)的值计算,N是卷积面的总数,j是自然数,i是相邻面的个数,k、α、β是可调参数。
  19. 根据权利要求16所述的计算机可读存储介质,其中,其中,所述程序指令被处理器执行时,还用于实现以下步骤:
    神经元的权值和偏置的通过以下计算机可读存储介质进行:若神经元y j与n个车辆事故图像输入x 1,x 2,…,x n对应连接的权值为w 1j,w 2j,…,w nj,则对权值进行初始化,所述初始化的计算机可读存储介质是指权值w nj满足以下正太分布:
    Figure PCTCN2020098767-appb-100046
    其中b为常数。
  20. 根据权利要求16所述的计算机可读存储介质,其中,所述程序指令被处理器执行时,还用于实现以下步骤:
    对获得车辆事故图像集合进行数据扩充,所述数据扩充是指对车辆事故图像集合旋转、翻转变换、平移变换、尺度变换颜色变化、以及缩放变换操作;通过以下数学公式实现数据扩充:
    Figure PCTCN2020098767-appb-100047
    其中x,y是指车辆事故图像的像素坐标,θ是指旋转的角度,a是指沿车辆事故图像像素x轴平移的单位距离,b是指车辆事故图像像素沿y轴平移的单位距离通过此公式可以对车辆事故图像进行翻转,旋转和平移,以增加车辆事故图像集合里车辆事故图像数据。
PCT/CN2020/098767 2019-07-23 2020-06-29 车辆定损方法、装置、设备和存储介质 WO2021012891A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201910667158.7A CN110502998B (zh) 2019-07-23 2019-07-23 车辆定损方法、装置、设备和存储介质
CN201910667158.7 2019-07-23

Publications (1)

Publication Number Publication Date
WO2021012891A1 true WO2021012891A1 (zh) 2021-01-28

Family

ID=68586277

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2020/098767 WO2021012891A1 (zh) 2019-07-23 2020-06-29 车辆定损方法、装置、设备和存储介质

Country Status (2)

Country Link
CN (1) CN110502998B (zh)
WO (1) WO2021012891A1 (zh)

Cited By (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113341894A (zh) * 2021-05-27 2021-09-03 河钢股份有限公司承德分公司 事故规律数据的生成方法、装置和终端设备
CN115631002A (zh) * 2022-12-08 2023-01-20 邦邦汽车销售服务(北京)有限公司 基于计算机视觉的车险智能定损方法及系统
CN115914286A (zh) * 2022-11-03 2023-04-04 中国人民财产保险股份有限公司 一种海量照片的异步上传方法、系统、电子设备及介质
CN115937681A (zh) * 2022-12-05 2023-04-07 中铁第四勘察设计院集团有限公司 一种遥感影像样本数据清洗方法
CN117094781A (zh) * 2023-08-25 2023-11-21 国任财产保险股份有限公司 一种智能车险定价和理赔处理方法及系统
CN117348741A (zh) * 2023-12-06 2024-01-05 深圳市金政软件技术有限公司 金额输入框的金额显示方法及相关设备
CN117455759A (zh) * 2023-10-23 2024-01-26 重庆科技学院 基于单目相机的车辆前方三维检测系统
US11935127B2 (en) 2021-02-09 2024-03-19 State Farm Mutual Automobile Insurance Company User interface associated with holistic digital claims platform

Families Citing this family (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110502998B (zh) * 2019-07-23 2023-01-31 平安科技(深圳)有限公司 车辆定损方法、装置、设备和存储介质
CN111127227A (zh) * 2019-12-26 2020-05-08 深圳市元征科技股份有限公司 一种车辆定损方法、装置和相关设备
CN111260487A (zh) * 2020-01-20 2020-06-09 北京中科泽达科技有限公司 一种车险理赔的风险控制方法及装置
CN112017065B (zh) * 2020-08-27 2024-05-24 中国平安财产保险股份有限公司 车辆定损理赔方法、装置及计算机可读存储介质
CN114462553B (zh) * 2022-04-12 2022-07-15 之江实验室 一种面向车险反欺诈的图像标注及要素抽取方法与系统
CN115994910B (zh) * 2023-03-24 2023-06-06 邦邦汽车销售服务(北京)有限公司 基于数据处理的汽车损伤程度确定方法及系统

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106372651A (zh) * 2016-08-22 2017-02-01 平安科技(深圳)有限公司 图片品质的检测方法及装置
CN108446618A (zh) * 2018-03-09 2018-08-24 平安科技(深圳)有限公司 车辆定损方法、装置、电子设备及存储介质
CN109784170A (zh) * 2018-12-13 2019-05-21 平安科技(深圳)有限公司 基于图像识别的车险定损方法、装置、设备及存储介质
US20190212739A1 (en) * 2019-03-12 2019-07-11 Georges Pierre Pantanelli Method to investigate human activities with artificial intelligence analysis in combination with logic and contextual analysis using advanced mathematic
CN110502998A (zh) * 2019-07-23 2019-11-26 平安科技(深圳)有限公司 车辆定损方法、装置、设备和存储介质

Family Cites Families (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106204468B (zh) * 2016-06-27 2019-04-26 深圳市未来媒体技术研究院 一种基于ReLU卷积神经网络的图像去噪方法
CN107992842B (zh) * 2017-12-13 2020-08-11 深圳励飞科技有限公司 活体检测方法、计算机装置及计算机可读存储介质
CN109523389A (zh) * 2018-09-19 2019-03-26 平安科技(深圳)有限公司 基于图像识别的车损处理方法、装置、设备及介质
CN109657805B (zh) * 2018-12-07 2021-04-23 泰康保险集团股份有限公司 超参数确定方法、装置、电子设备及计算机可读介质
CN109829929A (zh) * 2018-12-30 2019-05-31 中国第一汽车股份有限公司 一种基于深度边缘检测的层次场景语义分割模型
CN109948532A (zh) * 2019-03-19 2019-06-28 桂林电子科技大学 基于深度卷积神经网络的超宽带雷达人体动作识别方法

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106372651A (zh) * 2016-08-22 2017-02-01 平安科技(深圳)有限公司 图片品质的检测方法及装置
CN108446618A (zh) * 2018-03-09 2018-08-24 平安科技(深圳)有限公司 车辆定损方法、装置、电子设备及存储介质
CN109784170A (zh) * 2018-12-13 2019-05-21 平安科技(深圳)有限公司 基于图像识别的车险定损方法、装置、设备及存储介质
US20190212739A1 (en) * 2019-03-12 2019-07-11 Georges Pierre Pantanelli Method to investigate human activities with artificial intelligence analysis in combination with logic and contextual analysis using advanced mathematic
CN110502998A (zh) * 2019-07-23 2019-11-26 平安科技(深圳)有限公司 车辆定损方法、装置、设备和存储介质

Cited By (12)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US11935127B2 (en) 2021-02-09 2024-03-19 State Farm Mutual Automobile Insurance Company User interface associated with holistic digital claims platform
CN113341894A (zh) * 2021-05-27 2021-09-03 河钢股份有限公司承德分公司 事故规律数据的生成方法、装置和终端设备
CN115914286A (zh) * 2022-11-03 2023-04-04 中国人民财产保险股份有限公司 一种海量照片的异步上传方法、系统、电子设备及介质
CN115937681A (zh) * 2022-12-05 2023-04-07 中铁第四勘察设计院集团有限公司 一种遥感影像样本数据清洗方法
CN115937681B (zh) * 2022-12-05 2024-04-19 中铁第四勘察设计院集团有限公司 一种遥感影像样本数据清洗方法
CN115631002A (zh) * 2022-12-08 2023-01-20 邦邦汽车销售服务(北京)有限公司 基于计算机视觉的车险智能定损方法及系统
CN115631002B (zh) * 2022-12-08 2023-11-17 邦邦汽车销售服务(北京)有限公司 基于计算机视觉的车险智能定损方法及系统
CN117094781A (zh) * 2023-08-25 2023-11-21 国任财产保险股份有限公司 一种智能车险定价和理赔处理方法及系统
CN117094781B (zh) * 2023-08-25 2024-02-09 国任财产保险股份有限公司 一种智能车险理赔处理方法及系统
CN117455759A (zh) * 2023-10-23 2024-01-26 重庆科技学院 基于单目相机的车辆前方三维检测系统
CN117348741A (zh) * 2023-12-06 2024-01-05 深圳市金政软件技术有限公司 金额输入框的金额显示方法及相关设备
CN117348741B (zh) * 2023-12-06 2024-02-06 深圳市金政软件技术有限公司 金额输入框的金额显示方法及相关设备

Also Published As

Publication number Publication date
CN110502998A (zh) 2019-11-26
CN110502998B (zh) 2023-01-31

Similar Documents

Publication Publication Date Title
WO2021012891A1 (zh) 车辆定损方法、装置、设备和存储介质
US20220284517A1 (en) Automobile Monitoring Systems and Methods for Detecting Damage and Other Conditions
CN112862702B (zh) 图像增强方法、装置、设备及存储介质
US9569778B2 (en) Efficient prevention of fraud
US10901967B2 (en) License plate matching systems and methods
US11594013B2 (en) System, device, and method for image anomaly detection
US11461298B1 (en) Scoring parameter generation for identity resolution
US11514526B1 (en) Systems and methods for property damage restoration predictions based upon processed digital images
KR20190021187A (ko) 딥 러닝에 기반한 차량번호판 분류 방법, 시스템, 전자장치 및 매체
WO2018166116A1 (zh) 车损识别方法、电子装置及计算机可读存储介质
US20140270409A1 (en) Efficient prevention of fraud
US11417208B1 (en) Systems and methods for fraud prevention based on video analytics
CN111445058B (zh) 数据分析方法、装置、设备及计算机可读存储介质
US11450150B2 (en) Signature verification
CN112365007B (zh) 模型参数确定方法、装置、设备及存储介质
CN114639092A (zh) 图像识别方法、装置、设备和存储介质
CN112990868B (zh) 车辆保险自动赔付方法、系统、设备及存储介质
CN111310751A (zh) 车牌识别方法、装置、电子设备和存储介质
WO2021174869A1 (zh) 用户图片数据的处理方法、装置、计算机设备及存储介质
US20220036467A1 (en) Machine learning system and method for quote generation
CN111192150A (zh) 车辆出险代理业务的处理方法、装置、设备及存储介质
RU2018132643A (ru) Способ и устройство для определения ставок страховых взносов и скидок со страховых взносов для клиентов страховой компании
CA3036260A1 (en) Database image matching using machine learning with output personalization
CN108416880A (zh) 一种基于视频的识别方法
KR20220126345A (ko) 신분증 및 안면 인식을 이용한 비대면 차량 대여 시스템 및 그 방법

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 20843714

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 20843714

Country of ref document: EP

Kind code of ref document: A1

122 Ep: pct application non-entry in european phase

Ref document number: 20843714

Country of ref document: EP

Kind code of ref document: A1