WO2018166116A1

WO2018166116A1 - Car damage recognition method, electronic apparatus and computer-readable storage medium

Info

Publication number: WO2018166116A1
Application number: PCT/CN2017/091373
Authority: WO
Inventors: 黄章成; 马进; 王健宗; 肖京
Original assignee: 平安科技（深圳）有限公司
Priority date: 2017-03-13
Filing date: 2017-06-30
Publication date: 2018-09-20
Also published as: CN107092922A; CN107092922B

Abstract

Disclosed are a car damage recognition method, a server and a storage medium. The method comprises: receiving a loss assessment request sent by a user by means of a first terminal, using a pre-set first pre-set type model to analyse a loss assessment picture to obtain corresponding first car damage part classification information, and returning the first car damage part classification information to the first terminal for display (S10); if a rejection instruction, sent by the user by means of the first terminal, regarding the first car damage part classification information is received, re-using the pre-set first pre-set type model to analyse the loss assessment picture to obtain corresponding second car damage part classification information, and returning the second car damage part classification information to the first terminal for display (S20); and if a rejection instruction, sent by the user by means of the first terminal, regarding the second car damage part classification information is received, sending, to a pre-determined second terminal, an instruction to carry out manual car damage part recognition on the loss assessment picture (S30). The solution improves the precision and recall rate of car damage recognition.

Description

Vehicle damage recognition method, electronic device and computer readable storage medium

Priority claim

This application is based on the priority of the Paris Convention, which is entitled to the Chinese Patent Application No. CN201710147701.1, entitled "Car Damage Identification Method and Server", filed on March 13, 2017, the entire contents of which are hereby incorporated by reference. Combined in this application.

Technical field

The present invention relates to the field of computer technologies, and in particular, to a vehicle damage recognition method, an electronic device, and a computer readable storage medium.

Background technique

At present, one of the difficulties in the auto insurance business is that it requires a lot of manpower and resources to invest in the auto insurance claims. In order to effectively reduce the manpower and material resources of the auto insurance claims, some insurance companies currently use the technology of image detection and recognition to automatically detect damaged parts. However, the existing automatic detection scheme is sometimes easier to detect missing or identification errors of the damaged part, and the accuracy and recall of the identification are not guaranteed.

Summary of the invention

The main object of the present invention is to provide a vehicle damage recognition method, an electronic device and a computer readable storage medium, which aim to improve the accuracy and recall rate of vehicle damage recognition.

The first method of the present application provides a vehicle damage recognition method, and the method includes the following steps:

A. The server receives the fixed loss request that is sent by the user through the first terminal, and uses the preset first preset type model to analyze the fixed loss photo to obtain the first car corresponding to the fixed loss photo. Losing the part classification information, and returning the first vehicle damage part classification information to the first terminal for display;

B. If the user rejects the rejection instruction for the first vehicle damage location information sent by the first terminal, the first predetermined type model is used to analyze the determined loss photo again. The second vehicle damage part classification information corresponding to the fixed loss photo, and returning the second vehicle damage part classification information to the first terminal for display;

C. If receiving a rejection instruction for the second vehicle damage part classification information sent by the first terminal, sending a manual identification of the vehicle damage location to the fixed loss photo to the predetermined second terminal Instructions to manually identify the damage location.

A second aspect of the present application provides a server, including a processing device and a storage device connected to the processing device, the storage device storing a vehicle damage recognition system, the vehicle damage recognition system including at least one computer readable instruction, the at least one Computer readable instructions are executable by the processing device to:

A. The server receives a fixed loss request that is sent by the user through the first terminal, and uses the preset first preset type model to analyze the fixed loss photo to obtain the corresponding loss photo. The first vehicle damage part classification information, and returning the first vehicle damage part classification information to the first terminal for display;

A third aspect of the present application provides a computer readable storage medium having stored thereon at least one computer readable instruction executable by a processing device to:

According to the technical solution of the present invention, the first loss type photo is analyzed by the preset first preset type model to obtain the first vehicle damage part classification information, and if the user denies the first vehicle damage part classification information, the preset first The preset type model analyzes the fixed loss photo to obtain the second vehicle damage part classification information, and if the user denies the second vehicle damage part classification information, sends the fixed loss photo to the predetermined second terminal. Manual identification of the vehicle damage location to manually identify the vehicle damage location. Because the automatic identification of the vehicle damage is carried out with the user, the first preset type model is used to automatically identify the fixed loss photo twice, which improves the recognition accuracy and the passing rate, and saves manpower and material resources. Moreover, when the vehicle damage portion cannot be confirmed by two automatic recognitions, the vehicle damage portion is manually recognized for the fixed-loss photograph, thereby avoiding the occurrence of the missing portion of the damaged portion or the recognition error due to the inability to automatically identify the vehicle damage portion, and improving The accuracy and recall rate of vehicle damage identification.

DRAWINGS

1 is a schematic diagram of an application environment of an embodiment of a vehicle damage recognition method according to the present invention;

2 is a schematic flow chart of a first embodiment of a vehicle damage recognition method according to the present invention;

3 is a schematic flow chart of a second embodiment of a vehicle damage recognition method according to the present invention;

4 is a schematic diagram of functional modules of a first embodiment of a vehicle damage recognition system according to the present invention;

FIG. 5 is a schematic diagram of functional modules of a second embodiment of the vehicle damage recognition system of the present invention.

The implementation, functional features, and advantages of the present invention will be further described in conjunction with the embodiments.

detailed description

The present invention will be further described in detail below with reference to the accompanying drawings and embodiments, in order to make the present invention. It is understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

The invention provides a vehicle damage recognition method.

Referring to FIG. 1 , it is a schematic diagram of an application environment of an embodiment of a vehicle damage recognition method according to the present invention. The application environment diagram includes a server 1 and a terminal device 2. The server 1 can perform data interaction with the terminal device 2 through a suitable technology such as a network or a near field communication technology.

The terminal device 2 includes, but is not limited to, any electronic product that can interact with a user through a keyboard, a mouse, a remote controller, a touch pad, or a voice control device, for example, a personal computer, a tablet computer, a smart phone, or an individual. Personal Digital Assistant (PDA), game console, Internet Protocol Television (IPTV), smart wearable device, etc.

The server 1 is a device capable of automatically performing numerical calculation and/or information processing in accordance with an instruction set or stored in advance. The server 1 may be a computer, a single network server, a server group composed of a plurality of network servers, or a cloud-based cloud composed of a large number of hosts or network servers, wherein the cloud computing is a kind of distributed computing, and is loosely distributed by a group. A super virtual computer consisting of a set of coupled computers.

In the present embodiment, the server 1 includes, but is not limited to, a storage device 11, a processing device 12, and a network interface 13 that are communicably connected to each other through a system bus. It is pointed out that Figure 1 only shows the server 1 with the components 11-13, but it should be understood that not all illustrated components are required to be implemented, and more or fewer components may be implemented instead.

The storage device 11 includes a memory and at least one type of readable storage medium. The memory provides a cache for the operation of the server 1; the readable storage medium may be a non-volatile storage medium such as a flash memory, a hard disk, a multimedia card, a card type memory, or the like. In some embodiments, the readable storage medium may be an internal storage unit of the server 1, such as a hard disk of the server 1; in other embodiments, the non-volatile storage medium may also be an external storage device of the server 1, For example, a plug-in hard disk provided on the server 1, a smart memory card (SMC), a Secure Digital (SD) card, a flash card, and the like. In this embodiment, the readable storage medium of the storage device 11 is generally used to store an operating system installed on the server 1 and various types of application software, such as program codes of the vehicle damage recognition system 10 in an embodiment of the present application. Further, the storage device 11 can also be used to temporarily store various types of data that have been output or are to be output.

Processing device 12 may, in some embodiments, include one or more microprocessors, microcontrollers, Digital processor, etc. The processing device 12 is typically used to control the operation of the server 1, such as performing control and processing related to data interaction or communication with the terminal device 2. In the present embodiment, the processing device 12 is configured to run program code or process data stored in the storage device 11, such as running the vehicle damage recognition system 10 and the like.

The network interface 13 may comprise a wireless network interface or a wired network interface, which is typically used to establish a communication connection between the server 1 and other electronic devices. In this embodiment, the network interface 13 is mainly used to connect the server 1 with one or more terminal devices 2, and establish a data transmission channel and a communication connection between the server 1 and one or more terminal devices 2.

The vehicle damage recognition system 10 is stored in the storage device 11 and includes at least one computer readable instructions executable by the processing device 12 to implement the control response area display control method of various embodiments of the present application. As described later, the at least one computer readable instruction can be classified into different logic modules depending on the functions implemented by its various parts.

In an embodiment, when the vehicle damage recognition system 10 is executed by the processing device 12, the following operations are performed: receiving a loss request including a fixed loss photo sent by the user through the first terminal, using a preset first preset type model pair The determined loss photo is analyzed to obtain the first vehicle damage part classification information corresponding to the fixed loss photo, and the first vehicle damage part classification information is returned to the first terminal for display; if the user receives the user pass The rejecting instruction of the first vehicle damage part classification information sent by the first terminal is used to analyze the fixed loss photo by using a preset first preset type model to obtain a corresponding corresponding to the fixed loss photo. And classifying the second vehicle damage part classification information, and returning the second vehicle damage part classification information to the first terminal for display; if receiving the classification information of the second vehicle damage part issued by the user through the first terminal When the instruction is rejected, an instruction for manually identifying the damage location of the fixed-loss photo is sent to the predetermined second terminal to manually identify the vehicle damage location.

Referring to FIG. 2, FIG. 2 is a schematic flow chart of a first embodiment of a vehicle damage recognition method according to the present invention.

In the first embodiment, the vehicle damage recognition method includes:

Step S10: The vehicle damage recognition system receives a fixed loss request that is sent by the user through the first terminal, and uses the preset first preset type model to analyze the fixed loss photo to obtain the fixed loss photo corresponding to The first vehicle damage part classification information, and returning the first vehicle damage part classification information to the first terminal for display.

In this embodiment, the server receives, by the first terminal (for example, a mobile phone, a tablet computer, a handheld device, etc.), a fixed-loss photo including a user-supplied part including a to-be-damaged car damage location (for example, a close-up photo of a car damage part). ) The loss request. In an embodiment, the auto insurance claim application APP may be pre-installed in the first terminal, the user opens the auto insurance claim application APP and sends a loss request to the server through the auto insurance claim application APP; in another embodiment A browser system is pre-installed in the first terminal, and the user can access the server through the browser system, and send a loss request to the server through the browser system.

After receiving the loss request including the loss-receiving photo sent by the user, the server analyzes the acquired fixed-loss photo by using the pre-generated first preset type model to analyze the first vehicle damage corresponding to the fixed-loss photo. Part classification information (for example, front, side, rear, overall, etc.) and analysis The first vehicle damage part classification information is returned to the first terminal and displayed on a predetermined operation interface of the first terminal (for example, the analyzed first vehicle damage part classification information is returned to the first terminal The auto insurance claim application APP is displayed on the operation interface generated by the auto insurance claim application APP).

Step S20: If receiving the rejection instruction of the first vehicle loss location classification information sent by the first terminal, the method further analyzes the fixed loss photo by using a preset first preset type model. The second vehicle damage part classification information corresponding to the fixed loss photograph is obtained, and the second vehicle damage part classification information is returned to the first terminal for display.

Receiving, by the server, feedback information of the first car damage part classification information sent by the first terminal, such as confirming that the first vehicle damage part classification information is correct, or negating the first vehicle damage part classification Rejection instruction for information. It should be noted that, in this embodiment, the server may receive an instruction sent by the user by using a button, a touch, a press, a shaking, a mobile phone, a fingerprint, or the like, such as displaying the first on a predetermined operation interface of the first terminal. After the vehicle damage part classification information, the user may send feedback information about the classification information of the first vehicle damage part to the server by long pressing the first terminal screen or short pressing the first terminal screen, such as a confirmation instruction or a rejection instruction. For example, in the embodiment, the user sends the feedback information to the server in the manner that the user clicks the button on the first terminal. The predetermined operation interface of the first terminal includes a vehicle damage part classification information display area, a vehicle damage part classification information confirmation button, and a vehicle damage part classification information rejection button; if the user confirms the vehicle by using the vehicle damage part classification information confirmation button If the first vehicle damage part classification information is described, the server ends the vehicle damage part identification process, or if the user rejects the first vehicle damage part classification information through the vehicle damage part classification information rejection button, the server reuses the generated information. The first preset type model analyzes the acquired fixed loss photo to analyze the second vehicle damage part classification information corresponding to the fixed loss photo, and the second vehicle damage part classification information is the first preset type model pair The obtained vehicle damage part classification information may be the same as the first vehicle damage part classification information, or may be different from the first vehicle damage part classification information. The analyzed second vehicle damage part classification information is returned to the first terminal and displayed on a predetermined operation interface of the first terminal.

Step S30, if receiving a rejection instruction for the second vehicle damage part classification information sent by the first terminal, sending a manual identification of the vehicle damage location to the fixed loss photo to the predetermined second terminal The instructions to manually identify the damage location.

If the user confirms the second vehicle damage part classification information by using the vehicle damage part classification information confirmation button, the server ends the vehicle damage part identification process, or if the user rejects the vehicle through the vehicle damage part classification information rejection button The second vehicle damage part classification information is sent to the predetermined second terminal (for example, the terminal of the vehicle risk determination personnel) to send an instruction for manually identifying the vehicle damage part of the fixed loss photo to perform the vehicle damage part. Manual identification.

In this embodiment, the first damage type photo is analyzed by using a preset first preset type model to obtain the first vehicle damage part classification information. If the user denies the first vehicle damage part classification information, the preset first preset is used again. The type model analyzes the fixed loss photo to obtain the second vehicle damage part classification information, if When the user denies the second vehicle damage part classification information, the user sends a command for manually identifying the vehicle damage part to the predetermined second terminal to manually identify the vehicle damage part. Because the automatic identification of the vehicle damage is carried out with the user, the first preset type model is used to automatically identify the fixed loss photo twice, which improves the recognition accuracy and the passing rate, and saves manpower and material resources. Moreover, when the vehicle damage portion cannot be confirmed by two automatic recognitions, the vehicle damage portion is manually recognized for the fixed-loss photograph, thereby avoiding the occurrence of the missing portion of the damaged portion or the recognition error due to the inability to automatically identify the vehicle damage portion, and improving The accuracy and recall rate of vehicle damage identification.

As shown in FIG. 3, a second embodiment of the present invention provides a vehicle damage recognition method. On the basis of the foregoing embodiment, the above step S20 is replaced by:

Step S201: If the user manually receives the instruction of the vehicle damage location issued by the first terminal, the first terminal generates an area selection of the preset size and shape in the preset position of the display area of the fixed loss photo. a frame, the area selection box is configured for the user to adjust the current area selection frame to the preset direction to select the damaged photo feature area; and send the fixed loss photo feature area to the server;

Step S202: The server receives the fixed loss photo feature area, and analyzes the fixed loss photo feature area to obtain corresponding second car damage part classification information.

In this embodiment, the predetermined operation interface of the first terminal further includes a fixed photo display area and a vehicle frame manual button. If the user confirms the first vehicle damage part classification information by using the vehicle damage part classification information confirmation button, the server ends the vehicle damage part identification process, or if the user receives the manual framed button through the vehicle damage part The first terminal responds to the instruction by manually instructing or refusing the first vehicle damage part classification information through the vehicle damage part classification information rejection button (for example, the first terminal's automobile insurance claim application APP rings) It should be instructed to generate an area selection frame of a preset size and shape (for example, a rectangle of X*Y pixels) at a preset position (for example, a geometric center position) of the fixed-loss photo display area, the area selection box is used for The user manually adjusts the boundary line of the fixed-loss photo area included in the current area selection frame to the preset direction (for example, up, down, left, and right) to select the fixed-loss photo feature area selected by the user. Receiving, by the first terminal, a secondary identification instruction issued by the user that includes the fixed loss photo feature area selected based on the area selection box, the first terminal (eg, the first terminal's auto insurance claim application APP) rings The instruction should be identified twice and the fixed loss photo feature area sent to the server. After receiving the fixed-feature photo feature area, the server analyzes the fixed-loss photo feature area to analyze the second car-loss part classification information corresponding to the fixed-loss photo.

In this embodiment, when the first car damage part classification information obtained by analyzing the fixed loss photo is rejected by the user as the misclassification information, the user is first analyzed by the user before re-analysing the fixed loss photo. The identified feature area of the fixed loss photo is manually selected, and then the secondary damage analysis feature area is subjected to secondary analysis to obtain the corresponding second vehicle damage part classification information. In the secondary analysis, it is an analysis of the feature area of the fixed loss photo that is confirmed by the user, which effectively improves the accuracy of the secondary recognition.

Further, in other embodiments, the generating step of the first preset type model includes: obtaining, according to a preset vehicle damage part classification, each preset vehicle damage part from a preset vehicle risk claim database. The claim photo corresponding to the class is preprocessed for the claim photo corresponding to each preset car damage part classification, so as to convert the format of the claim photo into a preset format; and using the converted preset preset car damage part classification corresponding pre A convolutional neural network model of the preset model structure is trained to generate a convolutional neural network model corresponding to each preset car damage location classification.

The first preset type model is a convolutional neural network (CNN) model, and the first preset type model generating rule is: acquiring each preset car from a preset car insurance claim database according to a preset car damage part classification The claim photo corresponding to the damage part is classified, and the obtained claim photo of the preset car damage part classification is preprocessed to convert the obtained claim photo format into a preset format (for example, leveldb format); Each of the preset car damage parts is classified into a preset format picture, and the CNN model of the preset model structure is trained to generate a CNN model corresponding to each preset car damage part classification. The purpose of training is to optimize the values of the weights in the CNN model, so that the CNN model as a whole can be well applied to the classification and identification of vehicle damage parts in practical applications. The specific training process is as follows: Before the training starts, the system randomly and uniformly generates the initial values of the weights in the CNN model (for example, -0.05 to 0.05). The CNN model was trained using a stochastic gradient descent method. The entire training process can be divided into two stages: forward propagation and backward propagation. In the forward propagation phase, the system randomly samples the samples from the training data set, inputs them into the CNN network for calculation, and obtains the actual calculation results. In the backward propagation process, the difference between the actual result and the expected result is calculated, and then the value of each weight is inversely adjusted by the error minimization positioning method, and the effective error generated by the adjustment is calculated at the same time. The training process is iterated several times (for example, 100 times), and the training ends when the overall effective error of the CNN model is less than a predetermined threshold (for example, plus or minus 0.01).

Further, in other embodiments, after receiving the confirmation instruction of the first vehicle damage part classification information or the second vehicle damage part classification information sent by the first terminal, the method further includes:

The server analyzes the fixed loss photo by using a preset second preset type model, determines a car damage level corresponding to the fixed loss photo, and maps according to pre-stored car damage parts, vehicle damage levels and repair methods. Relationship, finding a repair method corresponding to the determined vehicle damage location and the vehicle damage level, and returning the determined vehicle damage location, the vehicle damage level, and the corresponding repair mode to the first terminal for display;

Receiving, by the first terminal, a rejection instruction for the vehicle damage level or repair mode issued by the first terminal, the server sends a manual identification or repair manner of the vehicle damage level to the fixed loss terminal to the predetermined second terminal. Manually identified instructions for manual identification of vehicle damage levels or repair methods.

In this embodiment, the predetermined operation interface of the first terminal further includes a vehicle damage level information display area and a repair mode information display area, the vehicle damage part classification information display area, the vehicle damage level information display area, and the repair mode information. The display areas correspond to one selection item. After the user confirms the first vehicle damage part classification information or the second vehicle damage part classification information by using the vehicle damage part classification information confirmation button, the server pairs the fixed loss type image by using a second preset type model generated in advance. Perform an analysis to determine the vehicle damage level corresponding to the fixed loss photo, and find out the determined vehicle damage location and the vehicle damage level according to the mapping relationship between the pre-stored vehicle damage location, the vehicle damage level and the repair mode. The repair method should be applied (for example, for sheet metal parts, the repair method includes only full spray, light sheet metal, light sheet metal + full spray, heavy sheet metal + full spray, replacement, etc.), and the determined A vehicle damage part classification information and its corresponding vehicle damage level and repair method are returned to the first terminal and displayed on a predetermined operation interface of the first terminal, or the determined second vehicle damage part classification information and The corresponding vehicle damage level and repair mode are returned to the first terminal and displayed on a predetermined operation interface of the first terminal.

If the user selects the determined vehicle damage level by the selection item corresponding to the vehicle damage level information display area, and the user rejects the selected vehicle damage level by using the vehicle loss part classification information rejection button, the user believes that the current automatic recognition is performed. If the vehicle damage level is incorrect, the server sends an instruction to identify the vehicle damage level to the predetermined loss photo to the predetermined second terminal (for example, the terminal of the car insurance loss person) to manually identify the vehicle damage level.

If the user selects the determined repair mode through the selection item corresponding to the repair mode information display area, and the user rejects the selected repair mode by using the vehicle damage part classification information rejection button, the user automatically believes that the repair method is currently recognized. If there is an error, the server sends an instruction to identify the repair mode to the predetermined second terminal (for example, the terminal of the car insurance loss person) to manually identify the repair mode.

In this embodiment, after the vehicle damage portion corresponding to the fixed-loss photo is correctly recognized, the determined vehicle damage level and the repair mode corresponding to the determined vehicle damage portion are further automatically recognized, and the identified When the vehicle damage level and repair method are wrong, manual identification can be carried out to more comprehensively identify the vehicle damage, so that the subsequent vehicle damage treatment can be carried out more conveniently and quickly.

Further, in other embodiments, the generating step of the second preset type model includes:

According to the preset car damage level classification, a predetermined number of fixed loss photos corresponding to each preset car damage level are obtained from the preset car insurance claim database; each of the acquired car damage parts corresponds to each preset car damage level. The classified fixed loss photo is preprocessed to convert the fixed loss photo into a preset size and a preset format; and the preset pre-format image corresponding to each preset car damage level is converted by using each converted car damage part, and the training pre- The convolutional neural network model of the model structure is set to generate a convolutional neural network model corresponding to each preset vehicle damage level.

In this embodiment, the second preset type model is a convolutional neural network (CNN) model, and the generating step of the second preset type model includes: the server classifies according to a preset vehicle loss level, for example, the pre- The classification of vehicle damage levels includes primary damage (for example, damage without deformation, no rupture), secondary damage (for example, 2 or less slight recoverable deformation, damage without rupture), and tertiary damage (1) More than one serious recoverable deformation or more than three minor recoverable deformations, no rupture damage), four-level damage (for example, damage that cannot be repaired manually), etc., from a preset auto insurance claim database (for example, The auto insurance claim database stores the mapping relationship or tag data of the preset car damage level classification, the car damage part and the fixed loss photo, and the fixed loss photo refers to the photo taken by the repair shop at the time of the loss). The part corresponds to a preset number (for example, 100,000 sheets) of the predetermined preset vehicle damage level classification, for example, obtaining 100,000 corresponding left front doors, and is a fixed-loss photo of the first-level damage. The server generates rules according to the preset model, and each of the obtained vehicle damage parts is corresponding to each a fixed-length photo of the preset car damage level classification, generating a second preset type model for analyzing the preset car damage level classification corresponding to the fixed-loss photo (for example, based on the car damage parts corresponding to the first-level damage) The preset number of fixed loss photos is generated, and a second preset type model for analyzing the car damage level corresponding to the fixed loss photo is generated.

The preset model generation rule is:

Pre-processing the fixed loss photos of each of the obtained vehicle damage parts corresponding to each preset vehicle damage level to convert the obtained fixed loss photos into preset sizes, and convert the format of the fixed-length photos converted into preset sizes into For the preset format (for example, leveldb format), the CNN model of the preset model structure is trained by using the preset preset image of each of the preset car damage levels corresponding to each car damage part to generate each car damage part corresponding to each The CNN model of the preset car damage level classification. The purpose of the training is to optimize the values of the weights in the CNN model, so that the CNN model as a whole can be well applied to the classification of each car damage location corresponding to each preset car damage level in practical applications. The CNN model can have seven layers, five convolutional layers, one downsampled layer, and one fully connected layer. The convolutional layer is formed by a feature map constructed by a plurality of feature vectors, and the function of the feature map is to extract key features by using a convolution filter. The function of the downsampling layer is to remove the feature points of repeated expression and reduce the number of feature extractions by sampling method, thereby improving the efficiency of data communication between network layers. The available sampling methods include maximum sampling method, mean sampling method and random sampling method. The role of the fully connected layer is to connect the previous convolutional layer with downsampling and calculate the weight matrix for subsequent actual classification. After entering the CNN model, each image undergoes two processes: forward iteration and backward iteration. Each iteration generates a probability distribution. The probability distributions after multiple iterations are superimposed, and the system selects the probability distribution to obtain the maximum value. The category is the final classification result.

The present invention further provides a vehicle damage recognition system that operates in the server 1 described above.

Referring to FIG. 4, FIG. 4 is a schematic diagram of functional modules of the first embodiment of the vehicle damage recognition system 10 of the present invention.

In the first embodiment, the vehicle damage recognition system 10 includes:

The first analysis module 01 is configured to receive a fixed loss request that is sent by the user through the first terminal, and use the preset first preset type model to analyze the fixed loss photo to obtain the fixed loss photo. Corresponding first vehicle damage part classification information, and returning the first vehicle damage part classification information to the first terminal for display;

In this embodiment, the first analysis module 01 receives a fixed-loss photo (for example, a vehicle damage) that is sent by the user through the first terminal (for example, a mobile phone, a tablet computer, a handheld device, etc.) and includes a user-supplied part including a to-be-determined damage. A close-up photo of the part) of the damage request. In an embodiment, the auto insurance claim application APP may be pre-installed in the first terminal, and the user opens the auto insurance claim application APP and sends a loss request to the first analysis module 01 through the auto insurance claim application APP; In an embodiment, a browser system is pre-installed in the first terminal, and the user can access the first analysis module 01 of the vehicle damage recognition system 10 through the browser system, and send the first analysis module 01 to the first analysis module 01 through the browser system. Fixed loss request.

The first analysis module 01 uses the pre-determination after receiving the request for the loss of the fixed-length photo sent by the user. The first preset type model generated first analyzes the acquired fixed loss photo to analyze the first car damage part classification information corresponding to the fixed loss photo (for example, front, side, rear, overall, etc.) And returning the analyzed first vehicle damage part classification information to the first terminal and displaying it on a predetermined operation interface of the first terminal (for example, returning the analyzed first vehicle damage part classification information to the first A terminal auto insurance claim application APP is displayed on the operation interface generated by the auto insurance claim application APP).

The second analysis module 02 is configured to: if the user rejects the rejection instruction for the first vehicle damage location information sent by the first terminal, use the preset first preset type model to Performing analysis on the damage photo, obtaining the second vehicle damage part classification information corresponding to the fixed loss photograph, and returning the second vehicle damage part classification information to the first terminal for display;

The second analysis module 02 receives the feedback information of the first vehicle damage part classification information sent by the user through the first terminal, such as confirming that the first vehicle damage part classification information is correct or confirming the first Rejection instruction for classification information of vehicle damage parts. It should be noted that, in this embodiment, the second analysis module 02 can receive an instruction sent by the user by using a button, a touch, a press, a shaking mobile phone, a fingerprint, and the like, such as on a predetermined operation interface of the first terminal. After displaying the first vehicle damage part classification information, the user may send the classification information of the first vehicle damage part to the second analysis module 02 by long pressing the first terminal screen or short pressing the first terminal screen. The feedback information, such as the acknowledgment command, the refusal command, and the like, is not limited herein. In this embodiment, only the user sends the feedback information to the second analysis module 02 by clicking the button on the first terminal. The predetermined operation interface of the first terminal includes a vehicle damage part classification information display area, a vehicle damage part classification information confirmation button, and a vehicle damage part classification information rejection button; if the user confirms the vehicle by using the vehicle damage part classification information confirmation button When the first vehicle damage part classification information is described, the vehicle damage recognition system 10 ends the vehicle damage part identification process, or if the user rejects the first vehicle damage part classification information by the vehicle damage part classification information rejection button, The second analysis module 02 analyzes the acquired fixed loss photo by using the generated first preset type model to analyze the second vehicle damage part classification information corresponding to the fixed loss photo, and the second vehicle damage part classification information is After the re-analysis of the obtained fixed-loss photo by using the first preset type model, the second vehicle-loss part classification information may be the same as the first vehicle-loss part classification information, or may be the first The classification information of the vehicle damage parts is different. The analyzed second vehicle damage part classification information is returned to the first terminal and displayed on a predetermined operation interface of the first terminal.

The manual identification module 03 is configured to: if the user receives the rejection instruction for the second vehicle damage part classification information sent by the first terminal, send the vehicle to the predetermined second terminal The manual identification of the damage part is to manually identify the damage part.

If the user confirms the second vehicle damage part classification information by using the vehicle damage part classification information confirmation button, the vehicle damage recognition system 10 ends the vehicle damage part identification process, or if the user passes the vehicle damage part classification information When the reject button rejects the second vehicle damage part classification information, the manual identification module 03 sends a command for manually identifying the vehicle damage part to the fixed loss photo to the predetermined second terminal (for example, the terminal of the vehicle risk determination personnel). , to manually identify the car damage parts.

In this embodiment, the first damage type photo is analyzed by using a preset first preset type model to obtain the first vehicle damage part classification information. If the user denies the first vehicle damage part classification information, the preset first preset is used again. The type model analyzes the fixed loss photo to obtain the second vehicle damage part classification information, and if the user denies the second vehicle damage part classification information, sends the vehicle damage part to the fixed loss photo to the predetermined second terminal. Manually recognized instructions to manually identify the damage location. Because the automatic identification of the vehicle damage is carried out with the user, the first preset type model is used to automatically identify the fixed loss photo twice, which improves the recognition accuracy and the passing rate, and saves manpower and material resources. Moreover, when the vehicle damage portion cannot be confirmed by two automatic recognitions, the vehicle damage portion is manually recognized for the fixed-loss photograph, thereby avoiding the occurrence of the missing portion of the damaged portion or the recognition error due to the inability to automatically identify the vehicle damage portion, and improving The accuracy and recall rate of vehicle damage identification.

Further, in other embodiments, the foregoing second analysis module 02 is further configured to:

And receiving the fixed loss photo feature area sent by the first terminal, and analyzing the fixed loss photo feature area to obtain corresponding second car damage part classification information, wherein the fixed loss photo feature area is obtained by: Receiving, by the first terminal, a frame selection frame of a preset size and shape in a preset position of the display area of the fixed-loss photo, if the user manually receives a command for the vehicle damage location issued by the first terminal, The area selection box is used for the user to adjust the current area selection box to the preset direction to select the damaged photo feature area.

In this embodiment, the predetermined operation interface of the first terminal further includes a fixed photo display area and a vehicle frame manual button. If the user confirms the first vehicle damage part classification information by using the vehicle damage part classification information confirmation button, the vehicle damage recognition system 10 ends the vehicle damage part identification process, or if the user receives the vehicle damage through the vehicle damage part The first terminal responds to the instruction (for example, the first terminal's auto insurance claim) by manually instructing the vehicle damage portion issued by the frame button or rejecting the first vehicle damage portion classification information through the vehicle damage portion classification information rejection button. The application APP responds to the instruction to generate an area selection frame of a preset size and shape (for example, a rectangle of X*Y pixels) at a preset position (for example, a geometric center position) of the fixed-loss photo display area, the area The selection box is used for the user to manually adjust the boundary line of the fixed loss photo area included in the current area selection box to the preset direction (for example, up, down, left, and right) to select the selected loss photo feature selected by the user. region. Receiving, by the first terminal, a secondary identification instruction issued by the user that includes the fixed loss photo feature area selected based on the area selection box, the first terminal (eg, the first terminal's auto insurance claim application APP) rings The instruction should be recognized twice, and the fixed loss photo feature area is sent to the second analysis module 02. After receiving the fixed loss photo feature area, the second analysis module 02 analyzes the fixed loss photo feature area to analyze the second car damage part classification information corresponding to the fixed loss photo.

In this embodiment, when the first car damage part classification information obtained by analyzing the fixed loss photo is rejected by the user as the misclassification information, the user is first analyzed by the user before re-analysing the fixed loss photo. The identified feature area of the fixed loss photo is manually selected, and then the secondary damage analysis feature area is subjected to secondary analysis to obtain the corresponding second vehicle damage part classification information. In the secondary analysis, it is an analysis of the feature area of the fixed loss photo that is confirmed by the user, which effectively improves the secondary knowledge. Other accuracy.

Further, in other embodiments, the generating step of the first preset type model includes: obtaining, according to a preset vehicle damage part classification, a claim photo corresponding to each preset car damage part classification from a preset car insurance claim database, Pre-processing the claim photo corresponding to each preset car damage part classification to convert the format of the claim photo into a preset format; using the preset preset format picture corresponding to each preset car damage part classification, training pre- A convolutional neural network model of the model structure is set to generate a convolutional neural network model corresponding to each preset car damage location classification.

As shown in FIG. 5, a second embodiment of the present invention provides a vehicle damage recognition system 10. Based on the foregoing embodiments, the method further includes:

The third analysis module 04 is configured to: after receiving the confirmation instruction for the first vehicle damage part classification information or the second vehicle damage part classification information sent by the first terminal, the preset second preset The type model analyzes the fixed loss photo, determines the vehicle damage level corresponding to the fixed loss photo, and finds the determined vehicle damage location according to the mapping relationship between the pre-stored vehicle damage location, the vehicle damage level and the repair mode. And the repairing method corresponding to the vehicle damage level, and returning the determined vehicle damage part, the vehicle damage level and the corresponding repairing manner to the first terminal for display;

The manual identification module 03 is further configured to: if receiving, by the first terminal, a rejection instruction for the vehicle damage level or the repair mode, send the determined loss photo to the predetermined second terminal. The manual identification of the vehicle damage level or the manual identification of the repair method to manually identify the damage level or repair method.

In this embodiment, the predetermined operation interface of the first terminal further includes a vehicle damage level information display area and a repair mode information display area, the vehicle damage part classification information display area, the vehicle damage level information display area, and the repair mode information. The display areas correspond to one selection item. In the user pass After the vehicle damage part classification information confirmation button confirms the first vehicle damage part classification information or the second vehicle damage part classification information, the third analysis module 04 corrects the fixed loss by using a second preset type model generated in advance. The photo is analyzed to determine the vehicle damage level corresponding to the fixed loss photo, and the determined vehicle damage location and the vehicle damage level are determined according to the mapping relationship between the pre-stored vehicle damage location, the vehicle damage level and the repair mode. Repair method (for example, for sheet metal parts, repair methods include only full spray, light sheet metal, light sheet metal + full spray, heavy sheet metal + full spray, replacement, etc.), and the first car will be determined The damage part classification information and its corresponding vehicle damage level and repair mode are returned to the first terminal and displayed on a predetermined operation interface of the first terminal, or the determined second vehicle damage part classification information and corresponding The vehicle damage level and repair mode are returned to the first terminal and displayed at a predetermined operational interface of the first terminal.

If the user selects the determined vehicle damage level by the selection item corresponding to the vehicle damage level information display area, and the user rejects the selected vehicle damage level by using the vehicle loss part classification information rejection button, the user believes that the current automatic recognition is performed. If the vehicle damage level is incorrect, the manual identification module 03 sends a command for identifying the damage level of the fixed loss photo to the predetermined second terminal (for example, the terminal of the vehicle risk-determining person) to perform the vehicle damage level. Manual identification.

If the user selects the determined repair mode through the selection item corresponding to the repair mode information display area, and the user rejects the selected repair mode by using the vehicle damage part classification information rejection button, the user automatically believes that the repair method is currently recognized. If there is an error, the manual identification module 03 sends an instruction to identify the repair mode of the fixed-loss photo to a predetermined second terminal (for example, the terminal of the car-losing person) to manually identify the repair mode.

In this embodiment, the second preset type model is a convolutional neural network (CNN) model, and the generating step of the second preset type model includes: classifying according to a preset vehicle damage level, for example, the preset The classification of vehicle damage levels includes primary damage (for example, damage without deformation, no rupture), secondary damage (for example, 2 or less slight recoverable deformation, damage without rupture), and tertiary damage (1 The above severe recoverable deformation or more than three minor recoverable deformations, no rupture damage), four-level damage (for example, damage that cannot be repaired manually), etc., from a preset auto insurance claim database (for example, the auto insurance) The claim database stores the preset car damage level classification, the car damage location and the fixed The mapping relationship or label data of the photo loss, the photo of the fixed loss refers to the photo taken by the repair shop at the time of the loss determination) the preset number of each car damage location corresponding to each preset car damage level classification (for example, 10 10,000 sheets) Fixed loss photos, for example, 100,000 corresponding left front doors, and are damage photos of the first level damage. According to the preset model generation rule, a second preset type model for analyzing the preset vehicle damage level classification corresponding to the fixed loss photo is generated based on the obtained fixed loss photos corresponding to the respective preset vehicle damage levels. (For example, based on a predetermined number of fixed-loss photos that have occurred in each vehicle damage portion corresponding to the first-level damage, a second preset type model for analyzing the vehicle damage level corresponding to the fixed-loss photo is generated).

The preset model generation rule is:

It is to be understood that the term "comprises", "comprising", or any other variants thereof, is intended to encompass a non-exclusive inclusion, such that a process, method, article, or device comprising a series of elements includes those elements. It also includes other elements that are not explicitly listed, or elements that are inherent to such a process, method, article, or device. An element that is defined by the phrase "comprising a ..." does not exclude the presence of additional equivalent elements in the process, method, item, or device that comprises the element.

The preferred embodiments of the present invention have been described above with reference to the drawings, and are not intended to limit the scope of the invention. The serial numbers of the embodiments of the present invention are merely for the description, and do not represent the advantages and disadvantages of the embodiments. Additionally, although logical sequences are shown in the flowcharts, in some cases the steps shown or described may be performed in a different order than the ones described herein.

A person skilled in the art can implement the invention in various variants without departing from the scope and spirit of the invention. For example, the features of one embodiment can be used in another embodiment to obtain a further embodiment. Any modifications, equivalent substitutions and improvements made within the technical concept of the invention are intended to be included within the scope of the invention.

Claims

A vehicle damage recognition method, characterized in that the method comprises the following steps:

A. The server receives the fixed loss request that is sent by the user through the first terminal, and uses the preset first preset type model to analyze the fixed loss photo to obtain the first car corresponding to the fixed loss photo. Losing the part classification information, and returning the first vehicle damage part classification information to the first terminal for display;

B. If the user rejects the rejection instruction for the first vehicle damage location information sent by the first terminal, the first predetermined type model is used to analyze the determined loss photo again. The second vehicle damage part classification information corresponding to the fixed loss photo, and returning the second vehicle damage part classification information to the first terminal for display;

C. If receiving a rejection instruction for the second vehicle damage part classification information sent by the first terminal, sending a manual identification of the vehicle damage location to the fixed loss photo to the predetermined second terminal Instructions to manually identify the damage location.
The vehicle damage recognition method according to claim 1, wherein the generating step of the first preset type model comprises: acquiring each preset vehicle damage from a preset car insurance claim database according to a preset car damage part classification; The claim photo corresponding to the part classification is pre-processed for the claim photo corresponding to each preset car damage part classification, so as to convert the format of the claim photo into a preset format; and the corresponding preset car damage part classification corresponding to the converted The preset format picture is used to train the convolutional neural network model of the preset model structure to generate a convolutional neural network model corresponding to each preset car damage part classification.
The vehicle damage recognition method according to claim 1, wherein after receiving the confirmation command for the first vehicle damage part classification information or the second vehicle damage part classification information issued by the first terminal, The method also includes:

The server analyzes the fixed loss photo by using a preset second preset type model, determines a car damage level corresponding to the fixed loss photo, and maps according to pre-stored car damage parts, vehicle damage levels and repair methods. Relationship, finding a repair method corresponding to the determined vehicle damage location and the vehicle damage level, and returning the determined vehicle damage location, the vehicle damage level, and the corresponding repair mode to the first terminal for display;

Receiving, by the first terminal, a rejection instruction for the vehicle damage level or repair mode issued by the first terminal, the server sends a manual identification or repair manner of the vehicle damage level to the fixed loss terminal to the predetermined second terminal. Manually identified instructions for manual identification of vehicle damage levels or repair methods.
The vehicle damage recognition method according to claim 3, wherein the generating step of the second preset type model comprises:

According to the preset car damage level classification, a predetermined number of fixed loss photos corresponding to each preset car damage level are obtained from the preset car insurance claim database; each of the acquired car damage parts corresponds to each preset car damage level. The classified fixed loss photo is preprocessed to convert the fixed loss photo into a preset size and a preset format; and the preset pre-format image corresponding to each preset car damage level is converted by using each converted car damage part, and the training pre- Set the convolutional neural network model of the model structure to generate individual vehicle losses The convolutional neural network model corresponding to each preset car damage level is corresponding to the part.
The vehicle damage recognition method according to claim 1, wherein said step B is replaced by:

B1. If the manual frame instruction is received by the user through the first terminal, the first terminal generates an area selection frame of a preset size and shape at a preset position of the display area of the fixed loss photo. The area selection box is used for the user to adjust the current area selection box to the preset direction to select the damaged photo feature area; and send the fixed loss photo feature area to the server;

B2. The server receives the fixed loss photo feature area, and analyzes the fixed loss photo feature area to obtain corresponding second car damage part classification information.
The vehicle damage recognition method according to claim 5, wherein the generating step of the first preset type model comprises: obtaining each preset vehicle damage from a preset car insurance claim database according to a preset car damage part classification The claim photo corresponding to the part classification is pre-processed for the claim photo corresponding to each preset car damage part classification, so as to convert the format of the claim photo into a preset format; and the corresponding preset car damage part classification corresponding to the converted The preset format picture is used to train the convolutional neural network model of the preset model structure to generate a convolutional neural network model corresponding to each preset car damage part classification.
The vehicle damage recognition method according to claim 5, wherein after receiving the confirmation command for the first vehicle damage part classification information or the second vehicle damage part classification information issued by the first terminal, The method also includes:

The server analyzes the fixed loss photo by using a preset second preset type model, determines a car damage level corresponding to the fixed loss photo, and maps according to pre-stored car damage parts, vehicle damage levels and repair methods. Relationship, finding a repair method corresponding to the determined vehicle damage location and the vehicle damage level, and returning the determined vehicle damage location, the vehicle damage level, and the corresponding repair mode to the first terminal for display;

Receiving, by the first terminal, a rejection instruction for the vehicle damage level or repair mode issued by the first terminal, the server sends a manual identification or repair manner of the vehicle damage level to the fixed loss terminal to the predetermined second terminal. Manually identified instructions for manual identification of vehicle damage levels or repair methods.
The vehicle damage recognition method according to claim 7, wherein the generating step of the second preset type model comprises:

According to the preset car damage level classification, a predetermined number of fixed loss photos corresponding to each preset car damage level are obtained from the preset car insurance claim database; each of the acquired car damage parts corresponds to each preset car damage level. The classified fixed loss photo is preprocessed to convert the fixed loss photo into a preset size and a preset format; and the preset pre-format image corresponding to each preset car damage level is converted by using each converted car damage part, and the training pre- The convolutional neural network model of the model structure is set to generate a convolutional neural network model corresponding to each preset vehicle damage level.
A server, comprising: a processing device and a storage device connected to the processing device, the storage device storing a vehicle damage recognition system, the vehicle damage recognition system comprising at least one computer readable instruction, the at least one computer A read command can be executed by the processing device to:

A. The server receives the fixed loss request sent by the user through the first terminal and includes the fixed loss photo, and utilizes The preset first preset type model analyzes the fixed loss photo, obtains the first vehicle damage part classification information corresponding to the fixed loss photo, and returns the first vehicle damage part classification information to the first The terminal performs display;

B. If the user rejects the rejection instruction for the first vehicle damage location information sent by the first terminal, the first predetermined type model is used to analyze the determined loss photo again. The second vehicle damage part classification information corresponding to the fixed loss photo, and returning the second vehicle damage part classification information to the first terminal for display;

C. If receiving a rejection instruction for the second vehicle damage part classification information sent by the first terminal, sending a manual identification of the vehicle damage location to the fixed loss photo to the predetermined second terminal Instructions to manually identify the damage location.
The server according to claim 9, wherein the generating step of the first preset type model comprises: obtaining, according to a preset car damage part classification, a corresponding car damage part classification corresponding from a preset car insurance claim database The claim photo, pre-processing the claim photo corresponding to each preset car damage part classification, to convert the format of the claim photo into a preset format; using the preset preset format corresponding to each preset car damage part classification Picture, a convolutional neural network model of the preset model structure is trained to generate a convolutional neural network model corresponding to each of the preset vehicle loss parts.
The server according to claim 9, wherein after receiving the confirmation instruction for the first vehicle damage part classification information or the second vehicle damage part classification information issued by the first terminal, the at least one Computer readable instructions may also be executed by the processing device to implement the following steps:

The fixed loss photo is analyzed by a preset second preset type model, and the car damage level corresponding to the fixed loss photo is determined, according to the mapping relationship between the pre-stored car damage part, the car damage level and the repair mode. Finding a repair method corresponding to the determined vehicle damage location and the vehicle damage level, and returning the determined vehicle damage location, the vehicle damage level, and the corresponding repair mode to the first terminal for display;

Receiving, by the first terminal, a rejection instruction for the vehicle damage level or the repair mode issued by the first terminal, transmitting, to the predetermined second terminal, the vehicle damage level manual identification or the repair method manually The identified instructions are manually identified for the level of damage or repair.
The server according to claim 11, wherein the generating step of the second preset type model comprises:

According to the preset car damage level classification, a predetermined number of fixed loss photos corresponding to each preset car damage level are obtained from the preset car insurance claim database; each of the acquired car damage parts corresponds to each preset car damage level. The classified fixed loss photo is preprocessed to convert the fixed loss photo into a preset size and a preset format; and the preset pre-format image corresponding to each preset car damage level is converted by using each converted car damage part, and the training pre- The convolutional neural network model of the model structure is set to generate a convolutional neural network model corresponding to each preset vehicle damage level.
The server of claim 9 wherein said step B is replaced by:

B1. If the user manually receives a command from the user to send a car damage location by the first terminal, the first terminal generates a preset size and shape in a preset position of the display area of the fixed loss photo. An area selection box, wherein the area selection box is used for the user to adjust the current area selection box to the preset direction to select the damaged photo feature area; and send the fixed loss photo feature area to the server;

B2. The server receives the fixed loss photo feature area, and analyzes the fixed loss photo feature area to obtain corresponding second car damage part classification information.
The server according to claim 13, wherein the generating step of the first preset type model comprises: obtaining, according to a preset car damage part classification, a corresponding car damage part classification corresponding from a preset car insurance claim database The claim photo, pre-processing the claim photo corresponding to each preset car damage part classification, to convert the format of the claim photo into a preset format; using the preset preset format corresponding to each preset car damage part classification Picture, a convolutional neural network model of the preset model structure is trained to generate a convolutional neural network model corresponding to each of the preset vehicle loss parts.
The server according to claim 13, wherein the at least one after receiving the confirmation instruction of the first vehicle damage part classification information or the second vehicle loss part classification information sent by the user through the first terminal Computer readable instructions may also be executed by the processing device to implement the following steps:

The server analyzes the fixed loss photo by using a preset second preset type model, determines a car damage level corresponding to the fixed loss photo, and maps according to pre-stored car damage parts, vehicle damage levels and repair methods. Relationship, finding a repair method corresponding to the determined vehicle damage location and the vehicle damage level, and returning the determined vehicle damage location, the vehicle damage level, and the corresponding repair mode to the first terminal for display;

Receiving, by the first terminal, a rejection instruction for the vehicle damage level or repair mode issued by the first terminal, the vehicle damage recognition system sends a vehicle damage level manual identification to the fixed loss photo to the predetermined second terminal. Or manually identify the manual identification method to manually identify the damage level or repair method.
The server according to claim 15, wherein the generating step of the second preset type model comprises:

According to the preset car damage level classification, a predetermined number of fixed loss photos corresponding to each preset car damage level are obtained from the preset car insurance claim database; each of the acquired car damage parts corresponds to each preset car damage level. The classified fixed loss photo is preprocessed to convert the fixed loss photo into a preset size and a preset format; and the preset pre-format image corresponding to each preset car damage level is converted by using each converted car damage part, and the training pre- The convolutional neural network model of the model structure is set to generate a convolutional neural network model corresponding to each preset vehicle damage level.
A computer readable storage medium having stored thereon at least one computer readable instruction executable by a processing device to:

A. The server receives the fixed loss request that is sent by the user through the first terminal, and uses the preset first preset type model to analyze the fixed loss photo to obtain the first car corresponding to the fixed loss photo. Losing the part classification information, and returning the first vehicle damage part classification information to the first terminal for display;

B. If the user rejects the rejection instruction for the first vehicle damage part classification information sent by the first terminal, the preset first type type model is used to analyze the fixed loss photo again. Obtaining second car damage part classification information corresponding to the fixed loss photo, and returning the second vehicle damage part classification information to the first terminal for display;

C. If receiving a rejection instruction for the second vehicle damage part classification information sent by the first terminal, sending a manual identification of the vehicle damage location to the fixed loss photo to the predetermined second terminal Instructions to manually identify the damage location.
The storage medium according to claim 17, wherein the generating step of the first preset type model comprises: obtaining, according to a preset car damage part classification, obtaining a preset car damage part classification from a preset car insurance claim database. Corresponding claim photos, pre-processing the claim photos corresponding to each preset car damage part classification, to convert the format of the claim photo into a preset format; using the preset presets corresponding to the converted car damage parts The format picture, the convolutional neural network model of the preset model structure is trained to generate a convolutional neural network model corresponding to each preset car damage part classification.
The storage medium according to claim 17, wherein after receiving the confirmation instruction for the first vehicle damage part classification information or the second vehicle damage part classification information issued by the user through the first terminal, Computer readable instructions are also used to:

The server analyzes the fixed loss photo by using a preset second preset type model, determines a car damage level corresponding to the fixed loss photo, and maps according to pre-stored car damage parts, vehicle damage levels and repair methods. Relationship, finding a repair method corresponding to the determined vehicle damage location and the vehicle damage level, and returning the determined vehicle damage location, the vehicle damage level, and the corresponding repair mode to the first terminal for display;

Receiving, by the first terminal, a rejection instruction for the vehicle damage level or repair mode issued by the first terminal, the vehicle damage recognition system sends a vehicle damage level manual identification to the fixed loss photo to the predetermined second terminal. Or manually identify the manual identification method to manually identify the damage level or repair method.
The storage medium according to claim 19, wherein the generating step of the second preset type model comprises:

According to the preset car damage level classification, a predetermined number of fixed loss photos corresponding to each preset car damage level are obtained from the preset car insurance claim database; each of the acquired car damage parts corresponds to each preset car damage level. The classified fixed loss photo is preprocessed to convert the fixed loss photo into a preset size and a preset format; and the preset pre-format image corresponding to each preset car damage level is converted by using each converted car damage part, and the training pre- The convolutional neural network model of the model structure is set to generate a convolutional neural network model corresponding to each preset vehicle damage level.