WO2018205467A1 - Automobile damage part recognition method, system and electronic device and storage medium - Google Patents

Abstract

Disclosed in the present invention are an automobile damage part recognition method, system and electronic device, and a storage medium. The method comprises: if an automobile insurance claim photo uploaded by a first terminal is received, recognizing a pixel area of an automobile damage part in the automobile insurance claim photo by utilizing a pre-trained recognition model, the predetermined recognition model being obtained through labeling and training in advance pixel areas in a preset number of automobile damage part sample pictures; and if the pixel area of an automobile damage part in the automobile insurance claim photo is recognized, labeling the recognized pixel area and sending the automobile insurance claim photo having the pixel area label to the first terminal and/or a predetermined second terminal, or intercepting the recognized pixel area and sending the intercepted pixel area to the first terminal and/or the predetermined second terminal. The present invention can more precisely recognize automobile damage parts of different area sizes in automobile insurance claim photos.

Description

Method, system, electronic device and storage medium for identifying vehicle damage parts

Priority claim

This application is based on the priority of the Paris Convention, which is entitled to the Chinese Patent Application No. 201710327373.3, entitled "Identification Method and System for Vehicle Damage Locations", which is filed on May 10, 2017, the entire contents of which are incorporated by reference. Combined in this application.

Technical field

The present invention relates to the field of computer technologies, and in particular, to a method, system, electronic device and storage medium for identifying a vehicle damage portion.

Background technique

At present, in the field of auto insurance claims, in order to improve the efficiency of claims, many auto insurance companies use image classification and recognition technology in the auto insurance claims system to automatically identify the vehicles and damaged parts in the uploaded claims photo. However, the existing image classification and recognition technology can only recognize the outline of the damaged part of the vehicle, and cannot accurately identify the damaged part of the vehicle. When one damaged part involves two vehicle parts at the same time, in the prior art, The identified outline of a substantially damaged vehicle location may only point to one vehicle location. For example, if the vehicle scratches are mostly on one vehicle part X1 and only a small part is on the edge of the other vehicle part X2, the vehicle damage identified by the prior art may be only X1. Therefore, how to more accurately identify the damage parts has become a technical problem to be solved.

Summary of the invention

The main object of the present invention is to provide a method, system, electronic device and storage medium for identifying a vehicle damage portion, which aim to accurately identify a vehicle damage portion of a vehicle.

To achieve the above object, the present invention provides a method for identifying a vehicle damage portion, the method comprising the following steps:

If the car insurance claim photo uploaded by the first terminal is received, the pixel area of the car damage part is identified by using the pre-trained recognition model; wherein the predetermined recognition model is pre-adjusted to the preset a number of pixel regions in each sample of the vehicle damage part are marked and trained to obtain a recognition model;

If the pixel area of the car damage part in the car insurance claim photo is identified, the identified pixel area is identified, and the car insurance claim photo with the pixel area identifier is sent to the first terminal and/or predetermined Or the second terminal, or the identified pixel area is intercepted and sent to the first terminal and/or the predetermined second terminal.

In addition, in order to achieve the above object, the present invention also provides a vehicle damage part identification system, and the vehicle damage part identification system includes:

The identification module is configured to use the pre-training if the car insurance claim photo uploaded by the first terminal is received The recognition model identifies the pixel area of the vehicle damage part of the car insurance claim photo; wherein the predetermined recognition model is to mark and train the pixel area in the preset number of each car damage part sample picture in advance The obtained recognition model;

a sending module, configured to: identify the pixel area of the vehicle damage part in the car insurance claim photo, and identify the identified pixel area, and send the car insurance claim photo with the pixel area identifier to the first terminal And/or a predetermined second terminal, or intercepting the identified pixel area for transmission to the first terminal and/or the predetermined second terminal.

In addition, in order to achieve the above object, the present invention further provides an electronic device including a memory and a processor connected to the memory, wherein the memory stores a vehicle damage portion operable on the processor The identification system, when the vehicle damage part identification system is executed by the processor, implements the following steps:

If the car insurance claim photo uploaded by the first terminal is received, the pixel area of the car damage part is identified by using the pre-trained recognition model; wherein the predetermined recognition model is pre-adjusted to the preset a number of pixel regions in each sample of the vehicle damage part are marked and trained to obtain a recognition model;

If the pixel area of the car damage part in the car insurance claim photo is identified, the identified pixel area is identified, and the car insurance claim photo with the pixel area identifier is sent to the first terminal and/or predetermined Or the second terminal, or the identified pixel area is intercepted and sent to the first terminal and/or the predetermined second terminal.

In addition, in order to achieve the above object, the present invention also provides a computer readable storage medium having a vehicle damage part identification system stored thereon, and the vehicle damage part identification system is implemented by the processor The following steps:

If the car insurance claim photo uploaded by the first terminal is received, the pixel area of the car damage part is identified by using the pre-trained recognition model; wherein the predetermined recognition model is pre-adjusted to the preset a number of pixel regions in each sample of the vehicle damage part are marked and trained to obtain a recognition model;

If the pixel area of the car damage part in the car insurance claim photo is identified, the identified pixel area is identified, and the car insurance claim photo with the pixel area identifier is sent to the first terminal and/or predetermined Or the second terminal, or the identified pixel area is intercepted and sent to the first terminal and/or the predetermined second terminal.

The method, system, electronic device and storage medium for identifying a vehicle damage part proposed by the present invention are to be recognized by a recognition model which is marked and trained by a preset number of pixel areas in each sample of the vehicle damage part. The car insurance claim photo is used to identify the pixel area of the car damage part. If the pixel area of the car damage part is recognized in the car insurance claim photo, the identified pixel area is marked or intercepted as the identified car damage part and then sent to The corresponding terminal. Since the specific vehicle damage area can be identified by identifying the pixel area of the vehicle damage part in the auto insurance claim photo, instead of identifying the vehicle damage part only by recognizing the outline of the damaged part of the vehicle, the car insurance claim photo can be more accurately identified. The car damage parts of different sizes in different areas.

DRAWINGS

1 is a schematic diagram of an optional application environment of each embodiment of the present invention;

2 is a schematic flow chart of an embodiment of a method for identifying a vehicle damage portion according to the present invention;

FIG. 3 is a schematic diagram of functional modules of an embodiment of a vehicle damage part identification system according to 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.

Please refer to FIG. 1. FIG. 1 is a schematic diagram of an optional application environment according to various embodiments of the present invention. The application environment diagram includes an electronic device 1, a first terminal 2, and a second terminal 3. The electronic device 1 can perform data interaction with the first terminal 2 and the second terminal 3 through a suitable technology such as a network or a near field communication technology.

The first terminal 2 or the second terminal 3 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 panel, or a voice control device, for example, a personal computer or a tablet. Mobile devices such as computers, smart phones, personal digital assistants (PDAs), game consoles, Internet Protocol Television (IPTV), smart wearable devices, navigation devices, etc., or such as digital TV Fixed terminals for desktop computers, notebooks, servers, etc.

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

In this embodiment, the electronic device 1 may include, but is not limited to, a memory 11 communicably connected to each other through a system bus, a processor 12, and a network interface 13, and the memory 11 stores a vehicle loss portion that can be operated on the processor 12. recognition system. It should be noted that FIG. 1 only shows the electronic device 1 having 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 memory 11 includes a memory and at least one type of readable storage medium. The memory provides a cache for the operation of the electronic device 1; the readable storage medium may be, for example, a flash memory, a hard disk, a multimedia card, a card type memory (eg, SD or DX memory, etc.), a random access memory (RAM), a static random access memory (SRAM). A non-volatile storage medium such as a read only memory (ROM), an electrically erasable programmable read only memory (EEPROM), a programmable read only memory (PROM), a magnetic memory, a magnetic disk, an optical disk, or the like. In some embodiments, the readable storage medium may be an internal storage unit of the electronic device 1, such as a hard disk of the electronic device 1; in other embodiments, the non-volatile storage medium may also be external to the electronic device 1. a storage device, such as a plug-in type on the electronic device 1 Hard disk, smart memory card (SMC), Secure Digital (SD) card, flash card, etc. In the present embodiment, the readable storage medium of the memory 11 is generally used to store an operating system and various types of application software installed in the electronic device 1, such as a program code of a vehicle damage part identification system in an embodiment of the present invention. Further, the memory 11 can also be used to temporarily store various types of data that have been output or are to be output.

The processor 12 may be a Central Processing Unit (CPU), controller, microcontroller, microprocessor, or other data processing chip in some embodiments. The processor 12 is typically used to control the overall operation of the electronic device 1, such as performing control and processing related to data interaction or communication with the first terminal 2 or the second terminal 3. In this embodiment, the processor 12 is configured to run program code or processing data stored in the memory 11, such as running a vehicle damage part identification system or 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 electronic device 1 and other electronic devices. In this embodiment, the network interface 13 is mainly used to connect the electronic device 1 with the first terminal 2 and the second terminal 3, and establish a data transmission channel and a communication connection between the electronic device 1 and the first terminal 2 and the second terminal 3. .

The vehicle damage location identification system is stored in the memory 11 and includes at least one computer readable instruction stored in the memory 11, the at least one computer readable instruction being executable by the processor 12 to implement the embodiments of the present application. The method; and the at least one computer readable instruction can be classified into different logic modules depending on the functions implemented by the various portions thereof.

Referring to FIG. 2, FIG. 2 is a schematic flow chart of an embodiment of a method for identifying a vehicle damage portion according to the present invention.

In an embodiment, the method for identifying the vehicle damage location includes:

Step S10: If the car insurance claim photo uploaded by the first terminal is received, the pixel area of the car damage part is identified by using the pre-trained recognition model; wherein the predetermined recognition model is pre-determined A recognition model in which a pixel area in a predetermined number of sample images of each vehicle loss portion is marked and trained.

In this embodiment, the vehicle damage part identification system can receive the vehicle damage part identification request sent by the user and include the car damage claim photo to be identified, for example, receiving the user through the first terminal (for example, the user's handheld terminal, the car insurance inspector's handheld terminal) Or the vehicle damage part identification request sent by the background operator of the loss-receiving staff, for example, receiving the vehicle damage part identification request sent by the user on the pre-installed client in the terminal such as the mobile phone, the tablet computer, the self-service terminal device, or receiving A vehicle loss part identification request sent by a user on a browser system in a terminal such as a mobile phone, a tablet computer, or a self-service terminal device.

After receiving the vehicle damage part identification request sent by the user, the vehicle damage part identification system uses the pre-trained identification model to identify the received auto insurance claim photo to be identified. The recognition model can be continuously labeled, and the pixel regions in the sample images of different vehicle damage parts are marked in advance, and the pixel regions marked in the sample images of different vehicle damage parts are continuously identified, thereby continuously training, learning, verifying, optimizing, etc. Train it into a model that accurately identifies the pixel areas of different vehicle damage locations. For example, the recognition model may employ a Convolutional Neural Network (CNN) model or the like.

Step S20, if the pixel area of the vehicle damage part in the car insurance claim photo is identified, the identification is performed The pixel area is identified, and the car insurance claim photo with the pixel area identifier is sent to the first terminal and/or the predetermined second terminal, or the identified pixel area is intercepted and sent to the first a terminal and/or a predetermined second terminal.

If the pixel area of the car damage claim in the car insurance claim photo is recognized by the recognition model, it indicates that there is a car damage part in the car insurance claim photo, and the specific car damage part area is recognized, that is, the recognized pixel area And identifying the identified pixel area (for example, according to a color identifier corresponding to a preset different vehicle damage part, performing corresponding color identification on the pixel area of the identified vehicle damage part, for example, setting a door damage identifier Yellow, bumper damage is marked as red, etc.; according to the corresponding digital identification of different vehicle damage parts, the corresponding digital identification of each pixel in the identified pixel area of the vehicle damage part, such as Setting each pixel point in the pixel area where the door is damaged is identified as "1", each pixel point in the pixel area damaged by the bumper is identified as "2", etc., and the photo with the pixel area identifier is sent to the first a terminal and/or a predetermined second terminal, so that the user accurately and comprehensively obtains the identified pixel area according to the pixel area identifier in the car insurance claim photo Said vehicle damage various parts of the region size photo auto insurance claims. In another embodiment, the identified pixel region may also be intercepted for transmission to the first terminal and/or the predetermined second terminal. Wherein, if the first terminal is a handheld terminal of the user or a handheld terminal of the vehicle inspector, the second terminal may be an office terminal of the background loss maker or the like.

If the pixel area of the car insurance claim photo is not recognized, the car damage claim photo does not exist in the car insurance claim photo or the car damage part is not successfully identified, and the unidentified car is sent to the first terminal. Reminder of the damage location to identify or transfer to manual identification.

In this embodiment, the pixel area of the vehicle damage part is identified by the identification model of the preset number of pixel areas in the sample picture of each vehicle damage part and is trained, if the vehicle insurance claims are compensated If the pixel area of the vehicle damage part is recognized in the photo, the identified pixel area is marked or intercepted as the identified vehicle damage part, and then transmitted to the corresponding terminal. Since the specific vehicle damage area can be identified by identifying the pixel area of the vehicle damage part in the auto insurance claim photo, instead of identifying the vehicle damage part only by recognizing the outline of the damaged part of the vehicle, the car insurance claim photo can be more accurately identified. The car damage parts of different sizes in different areas.

Further, in other embodiments, the recognition model is a deep convolutional neural network model without a fully connected layer, the deep convolutional neural network model including an input layer, a convolution layer, a pooling layer, and an upsampling Layer and cutting layer.

In a specific embodiment, the deep convolutional neural network model is composed of one input layer, 16 convolution layers, 5 pooling layers, 1 upsampling layer, and 1 cropping layer. The detailed structure of the deep convolutional neural network model is shown in Table 1 below:

Layer Name	Channel	Kernel Size	Stride Size	Pad Size
Input	3	N/A	N/A	N/A
Conv1_1	64	3	1	100
Conv1_2	64	3	1	1

MaxPool1		Computable value	2	2	0
Conv2_1	128	3	1	1
Conv2_2	128	3	1	1
MaxPool2		Computable value	2	2	0
Conv3_1	256	3	1	1
Conv3_2	256	3	1	1
Conv3_3	256	3	1	1
MaxPool3		Computable value	2	2	0
Conv4_1	512	3	1	1
Conv4_2	512	3	1	1
Conv4_3	512	3	1	1
MaxPool4		Computable value	2	2	0
Conv5_1	512	3	1	1
Conv5_2	512	3	1	1
Conv5_3	512	3	1	1
MaxPool5		Computable value	2	2	0
Conv6	4096	7	1	0
Conv7	4096	1	1	0
Conv8	Class num+1	5	1	0
Upscore	Class num+1	64	32	0
Crop	N/A	N/A	N/A	N/A
SoftmaxWithLoss	N/A	N/A	N/A	N/A

Table 1

Among them: Layer Name column indicates the name of each layer, Channel indicates the number of channels output, Kernel Size indicates the scale of the current layer convolution kernel (for example, Kernel Size can be equal to 3, indicating that the scale of the convolution kernel is 3x 3), Stride Size indicates the moving step size of the convolution kernel, that is, the distance moved to the next convolution position after one convolution is completed; Pad Size indicates the size of the image fill in the current network layer. Input represents the data input layer of the network, Conv represents the convolutional layer of the model, Conv1 represents the first convolutional layer of the model, MaxPool represents the maximum pooled layer of the model, and MaxPool1 represents the first maximum pooled layer of the model. SoftmaxWithLoss is the Softmax layer used in the training phase to calculate Loss. Unlike the SoftMax layer, it only calculates Loss; Upscore represents the upsampling layer and implements deconvolution; Crop represents the cropping layer, and the Upscore layer is cropped to the same size as the original image. Big. It should be noted that the pooling mode of the pooling layer in this embodiment includes, but is not limited to, Mean pooling, Max pooling, Overlapping, L2pooling, Local Contrast. Normalization, Stochasticpooling, Def-pooling, and more.

In the traditional classification problem, it is generally necessary to use the fully connected layer to output the probability that each picture belongs to each class. However, in the semantic segmentation problem, using this method to predict which class each pixel belongs to will inevitably lead to efficiency. low. Therefore, the recognition model in this embodiment is a deep convolutional neural network model without a fully connected layer, and the deep convolutional neural network model only needs to be on Conv8. A convolutional layer to output the classification score for each pixel. At this level, each point on the feature map has scores of different classifications in class num+1 categories, so the output channel is also class num+1, and the recognition efficiency is greatly improved.

Further, in other embodiments, the training process of the predetermined recognition model is as follows:

A. Prepare a corresponding preset number of sample pictures for preset vehicle damage parts (for example, left front door, right front door, left fender, right fender, front bumper, rear bumper, etc.);

B. Perform image preprocessing on each sample picture to obtain a training picture to be trained by the model. For example, the model training can be performed after performing image preprocessing such as scaling, cropping, flipping, and/or distorting on each sample picture to effectively perform model training. Improve the authenticity and accuracy of model training. Then, according to the preset mapping relationship between the vehicle damage part and the label color, the pixel color of the vehicle damage part of each training picture is changed to the corresponding label color, and the training picture for each label color is changed according to a preset conversion rule. Converting and generating a corresponding pixel area labeling matrix of the vehicle damage part;

C. The training pictures with all the pixel area labeling matrix of the vehicle loss part are divided into a training set of a first ratio (for example, 70%) and a verification set of a second ratio (for example, 30%);

D. training the predetermined recognition model by using the training set;

E. verifying the accuracy of the training recognition model by using the verification set, if the accuracy rate is greater than or equal to the preset accuracy rate, the training ends, or if the accuracy rate is less than the preset accuracy rate, increasing the corresponding position of each vehicle damage part The number of sample pictures is re-executed in steps B, C, D, and E until the accuracy of the trained recognition model is greater than or equal to the preset accuracy.

Further, in other embodiments, the preset conversion rule includes:

Recognizing a label color pixel area and a corresponding vehicle damage part thereof in each training picture in which the label color is changed;

Determining, according to a predetermined mapping relationship between the vehicle damage location and the identification data, identification data corresponding to the label color pixel region in each training image in which the label color is changed;

Converts all other pixels except the label color pixel area in the training picture whose label color has been changed to preset data (for example, 0), and in the label color pixel area in the training picture in which each label color is changed Each pixel is converted into corresponding identification data (for example, the identification data corresponding to the label color pixel area of the left front door is 1, and the identification data corresponding to each pixel in the label color pixel area of the left front door is 1; the label color of the right front door The identification data corresponding to the pixel area is 2, and the identification data corresponding to each pixel in the label color pixel area of the right front door is 2), to obtain the pixel area labeling matrix of the vehicle loss part corresponding to each training picture whose label color is changed, so that The subsequent recognition model trains the recognition of the pixel areas of different vehicle damage parts.

Further, in other embodiments, the step of performing image pre-processing on each sample picture to obtain a training picture to be model-trained includes:

Adjusting each sample picture to a first picture of a first preset size (eg, 384*384 pixels), and randomly cropping a second preset size (eg, 256*256 pixels) on each first picture Two pictures;

Performing a flip of a preset direction (for example, a horizontal direction and a vertical direction) for each of the second pictures, and performing a warping operation on each of the second pictures according to a preset twist angle to obtain a third picture corresponding to each of the second pictures. Among them, the function of the flipping and twisting operation is to simulate various forms of images in the actual business scene. Through these flipping and twisting operations, the scale of the data set can be increased, thereby improving the authenticity and practicability of the model training.

An average pixel picture of all the second picture and the third picture corresponding to each sample picture is calculated, and each pixel of the average pixel picture is an average pixel of the corresponding second picture and the third picture corresponding pixel. For example, the pixel point X of the average pixel picture corresponds to the pixel point X1 of the second picture and the pixel point X2 of the third picture, respectively, and the pixel of the pixel point X is the average pixel of the pixels of all the pixel points X1 and the pixel point X2. .

Based on the average pixel picture, a training picture corresponding to each sample picture can be obtained. For example, each of the second picture and the average picture picture of the third picture corresponding to each sample picture can be directly used as the training picture corresponding to each sample picture, and each pixel in each of the second picture and the third picture corresponding to each sample picture can also be used. The corresponding pixels in the corresponding average pixel image are respectively subtracted to obtain a training picture corresponding to each sample picture, which is not limited herein.

The invention further provides a vehicle damage part identification system. In the embodiment, the vehicle damage part identification system is installed and operated in the electronic device 1.

Please refer to FIG. 3, which is a functional block diagram of a preferred embodiment of the vehicle damage part identification system of the present invention. In this embodiment, the vehicle damage part identification system may be divided into one or more modules, the one or more modules being stored in the memory 11 and being processed by one or more processors (this Embodiments are performed by the processor 12) to complete the present invention. For example, in FIG. 3, the vehicle damage part identification system may be divided into an identification module 01 and a transmission module 02. The term "module" as used in the present invention refers to a series of computer program instruction segments capable of performing a specific function, which is more suitable than the program for describing the execution process of the vehicle damage part identification system in the electronic device 1. The following description will specifically describe the functions of the identification module 01 and the transmission module 02.

The identification module 01 is configured to: if the car insurance claim photo uploaded by the first terminal is received, identify, by using a pre-trained recognition model, the pixel area of the car damage part of the car insurance claim photo; wherein the predetermined recognition model is A recognition model obtained by pre-labeling and training a pixel area in a preset number of sample images of respective vehicle damage parts.

In this embodiment, the vehicle damage part identification system can receive the vehicle damage part identification request sent by the user and include the car damage claim photo to be identified, for example, receiving the user through the first terminal (for example, the user's handheld terminal, the car insurance inspector's handheld terminal) Or the vehicle damage part identification request sent by the background operator of the loss-receiving staff, for example, receiving the vehicle damage part identification request sent by the user on the pre-installed client in the terminal such as the mobile phone, the tablet computer, the self-service terminal device, or receiving A vehicle loss part identification request sent by a user on a browser system in a terminal such as a mobile phone, a tablet computer, or a self-service terminal device.

After receiving the vehicle damage part identification request sent by the user, the vehicle damage part identification system uses the pre-trained identification model to identify the received auto insurance claim photo to be identified. The recognition model can be pre- Firstly, the pixel area in the sample picture of a large number of different vehicle damage parts is marked, and the pixel area marked in the sample picture of different vehicle damage parts is continuously identified, and training, learning, verification, optimization, etc. are continuously trained to train A model that accurately identifies pixel regions of different vehicle damage locations. For example, the recognition model may employ a Convolutional Neural Network (CNN) model or the like.

The sending module 02 is configured to: if the pixel area of the car damage part in the car insurance claim photo is identified, identify the identified pixel area, and send the car insurance claim photo with the pixel area identifier to the first The terminal and/or the predetermined second terminal, or the identified pixel area is intercepted and sent to the first terminal and/or the predetermined second terminal.

If the pixel area of the car damage claim in the car insurance claim photo is recognized by the recognition model, it indicates that there is a car damage part in the car insurance claim photo, and the specific car damage part area is recognized, that is, the recognized pixel area And identifying the identified pixel area (for example, according to a color identifier corresponding to a preset different vehicle damage part, performing corresponding color identification on the pixel area of the identified vehicle damage part, for example, setting a door damage identifier Yellow, bumper damage is marked as red, etc.; according to the corresponding digital identification of different vehicle damage parts, the corresponding digital identification of each pixel in the identified pixel area of the vehicle damage part, such as Setting each pixel point in the pixel area where the door is damaged is identified as "1", each pixel point in the pixel area damaged by the bumper is identified as "2", etc., and the photo with the pixel area identifier is sent to the first a terminal and/or a predetermined second terminal, so that the user accurately and comprehensively obtains the identified pixel area according to the pixel area identifier in the car insurance claim photo Said vehicle damage various parts of the region size photo auto insurance claims. In another embodiment, the identified pixel region may also be intercepted for transmission to the first terminal and/or the predetermined second terminal. Wherein, if the first terminal is a handheld terminal of the user or a handheld terminal of the vehicle inspector, the second terminal may be an office terminal of the background loss maker or the like.

If the pixel area of the car insurance claim photo is not recognized, the car damage claim photo does not exist in the car insurance claim photo or the car damage part is not successfully identified, and the unidentified car is sent to the first terminal. Reminder of the damage location to identify or transfer to manual identification.

In this embodiment, the pixel area of the vehicle damage part is identified by the identification model of the preset number of pixel areas in the sample picture of each vehicle damage part and is trained, if the vehicle insurance claims are compensated If the pixel area of the vehicle damage part is recognized in the photo, the identified pixel area is marked or intercepted as the identified vehicle damage part, and then transmitted to the corresponding terminal. Since the specific vehicle damage area can be identified by identifying the pixel area of the vehicle damage part in the auto insurance claim photo, instead of identifying the vehicle damage part only by recognizing the outline of the damaged part of the vehicle, the car insurance claim photo can be more accurately identified. The car damage parts of different sizes in different areas.

Further, in other embodiments, the recognition model is a deep convolutional neural network model without a fully connected layer, the deep convolutional neural network model including an input layer, a convolution layer, a pooling layer, and an upsampling Layer and cutting layer.

In a specific embodiment, the deep convolutional neural network model consists of one input layer. 16 convolution layers, 5 pooling layers, 1 upsampling layer, and 1 cropping layer. The detailed structure of the deep convolutional neural network model is shown in Table 1 below:

Layer Name	Channel	Kernel Size	Stride Size	Pad Size
Input	3	N/A	N/A	N/A
Conv1_1	64	3	1	100
Conv1_2	64	3	1	1
MaxPool1		Computable value	2	2	0
Conv2_1	128	3	1	1
Conv2_2	128	3	1	1
MaxPool2		Computable value	2	2	0
Conv3_1	256	3	1	1
Conv3_2	256	3	1	1
Conv3_3	256	3	1	1
MaxPool3		Computable value	2	2	0
Conv4_1	512	3	1	1
Conv4_2	512	3	1	1
Conv4_3	512	3	1	1
MaxPool4		Computable value	2	2	0
Conv5_1	512	3	1	1
Conv5_2	512	3	1	1
Conv5_3	512	3	1	1
MaxPool5		Computable value	2	2	0
Conv6	4096	7	1	0
Conv7	4096	1	1	0
Conv8	Class num+1	5	1	0
Upscore	Class num+1	64	32	0
Crop	N/A	N/A	N/A	N/A
SoftmaxWithLoss	N/A	N/A	N/A	N/A

Table 1

Among them: Layer Name column indicates the name of each layer, Channel indicates the number of channels output, Kernel Size indicates the scale of the current layer convolution kernel (for example, Kernel Size can be equal to 3, indicating that the scale of the convolution kernel is 3x 3), Stride Size indicates the moving step size of the convolution kernel, that is, the distance moved to the next convolution position after one convolution is completed; Pad Size indicates the size of the image fill in the current network layer. Input represents the data input layer of the network, Conv represents the convolutional layer of the model, Conv1 represents the first convolutional layer of the model, MaxPool represents the maximum pooled layer of the model, and MaxPool1 represents the first maximum pooled layer of the model. SoftmaxWithLoss is the Softmax layer used in the training phase to calculate Loss. Unlike the SoftMax layer, it only calculates Loss; Upscore represents the upsampling layer and implements deconvolution; Crop represents the cropping layer, and the Upscore layer is cropped to the same size as the original image. Big. It should be noted that the pooling mode of the pooling layer in this embodiment includes, but is not limited to, Mean pooling, Max pooling, Overlapping, and L2pooling. Square sampling), Local Contrast Normalization, Stochasticpooling, Def-pooling, and more.

In the traditional classification problem, it is generally necessary to use the fully connected layer to output the probability that each picture belongs to each class. However, in the semantic segmentation problem, using this method to predict which class each pixel belongs to will inevitably lead to efficiency. low. Therefore, the recognition model in this embodiment is a deep convolutional neural network model without a fully connected layer. The deep convolutional neural network model only needs to output a classification score for each pixel on a Conv8 layer. . At this level, each point on the feature map has scores of different classifications in class num+1 categories, so the output channel is also class num+1, and the recognition efficiency is greatly improved.

Further, in other embodiments, the training process of the predetermined recognition model is as follows:

A. Prepare a corresponding preset number of sample pictures for preset vehicle damage parts (for example, left front door, right front door, left fender, right fender, front bumper, rear bumper, etc.);

B. Perform image preprocessing on each sample picture to obtain a training picture to be trained by the model. For example, the model training can be performed after performing image preprocessing such as scaling, cropping, flipping, and/or distorting on each sample picture to effectively perform model training. Improve the authenticity and accuracy of model training. Then, according to the preset mapping relationship between the vehicle damage part and the label color, the pixel color of the vehicle damage part of each training picture is changed to the corresponding label color, and the training picture for each label color is changed according to a preset conversion rule. Converting and generating a corresponding pixel area labeling matrix of the vehicle damage part;

C. The training pictures with all the pixel area labeling matrix of the vehicle loss part are divided into a training set of a first ratio (for example, 70%) and a verification set of a second ratio (for example, 30%);

D. training the predetermined recognition model by using the training set;

E. verifying the accuracy of the training recognition model by using the verification set, if the accuracy rate is greater than or equal to the preset accuracy rate, the training ends, or if the accuracy rate is less than the preset accuracy rate, increasing the corresponding position of each vehicle damage part The number of sample pictures is re-executed in steps B, C, D, and E until the accuracy of the trained recognition model is greater than or equal to the preset accuracy.

Further, in other embodiments, the preset conversion rule includes:

Recognizing a label color pixel area and a corresponding vehicle damage part thereof in each training picture in which the label color is changed;

Determining, according to a predetermined mapping relationship between the vehicle damage location and the identification data, identification data corresponding to the label color pixel region in each training image in which the label color is changed;

Converts all other pixels except the label color pixel area in the training picture whose label color has been changed to preset data (for example, 0), and in the label color pixel area in the training picture in which each label color is changed Each pixel is converted into corresponding identification data (for example, the identification data corresponding to the label color pixel area of the left front door is 1, and the identification data corresponding to each pixel in the label color pixel area of the left front door is 1; the label color of the right front door The identification data corresponding to the pixel area is 2, and the identification data corresponding to each pixel in the label color pixel area of the right front door is 2), to obtain the pixel area labeling matrix of the vehicle loss part corresponding to each training picture whose label color is changed, The subsequent recognition model identifies the training of the pixel areas of different vehicle damage parts.

Further, in other embodiments, the step of performing image pre-processing on each sample picture to obtain a training picture to be model-trained includes:

Adjusting each sample picture to a first picture of a first preset size (eg, 384*384 pixels), and randomly cropping a second preset size (eg, 256*256 pixels) on each first picture image;

Performing a flip of a preset direction (for example, a horizontal direction and a vertical direction) for each of the second pictures, and performing a warping operation on each of the second pictures according to a preset twist angle to obtain a third picture corresponding to each of the second pictures. Among them, the function of the flipping and twisting operation is to simulate various forms of images in the actual business scene. Through these flipping and twisting operations, the scale of the data set can be increased, thereby improving the authenticity and practicability of the model training.

An average pixel picture of all the second picture and the third picture corresponding to each sample picture is calculated, and each pixel of the average pixel picture is an average pixel of the corresponding second picture and the third picture corresponding pixel. For example, the pixel point X of the average pixel picture corresponds to the pixel point X1 of the second picture and the pixel point X2 of the third picture, respectively, and the pixel of the pixel point X is the average pixel of the pixels of all the pixel points X1 and the pixel point X2. .

Based on the average pixel picture, a training picture corresponding to each sample picture can be obtained. For example, each of the second picture and the average picture picture of the third picture corresponding to each sample picture can be directly used as the training picture corresponding to each sample picture, and each pixel in each of the second picture and the third picture corresponding to each sample picture can also be used. The corresponding pixels in the corresponding average pixel image are respectively subtracted to obtain a training picture corresponding to each sample picture, which is not limited herein.

The present invention also provides a computer readable storage medium having stored thereon a vehicle damage location identification system that implements the steps of the method described above when executed by a processor.

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.

Through the description of the above embodiments, those skilled in the art can clearly understand that the foregoing embodiment method can be implemented by means of software plus a necessary general hardware platform, and can also be implemented by hardware, but in many cases, the former is A better implementation. Based on such understanding, the technical solution of the present invention, which is essential or contributes to the prior art, may be embodied in the form of a software product stored in a storage medium (such as ROM/RAM, disk, The optical disc includes a number of instructions for causing a terminal device (which may be a cell phone, a computer, a server, an air conditioner, or a network device, etc.) to perform the methods described in various embodiments of the present invention.

The preferred embodiments of the present invention have been described above with reference to the accompanying drawings, and are not intended to limit the scope of the invention range. 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 (20)

1. A method for identifying a vehicle damage location, characterized in that the method comprises the following steps:

   If the car insurance claim photo uploaded by the first terminal is received, the pixel area of the car damage part is identified by using the pre-trained recognition model; wherein the predetermined recognition model is pre-adjusted to the preset a number of pixel regions in each sample of the vehicle damage part are marked and trained to obtain a recognition model;

   If the pixel area of the car damage part in the car insurance claim photo is identified, the identified pixel area is identified, and the car insurance claim photo with the pixel area identifier is sent to the first terminal and/or predetermined Or the second terminal, or the identified pixel area is intercepted and sent to the first terminal and/or the predetermined second terminal.
2. The method for identifying a vehicle damage portion according to claim 1, wherein the recognition model is a deep convolutional neural network model without a fully connected layer, and the deep convolutional neural network model includes an input layer and a volume. Stack, pooling layer, upsampling layer and cutting layer.
3. The method for identifying a vehicle damage portion according to claim 1 or 2, wherein the training process of the recognition model is as follows:

   A. Preparing a corresponding preset number of sample pictures for each preset vehicle damage part;

   B. Perform image preprocessing on each sample image to obtain a training picture to be trained by the model, and change the pixel color of the car damage part of each training picture to a corresponding label according to a preset mapping relationship between the car damage part and the label color. The color, and the training picture for each of the changed label colors is converted according to a preset conversion rule to generate a corresponding pixel area labeling matrix of the vehicle damage part;

   C. The training pictures with all the pixel area labeling matrix of the vehicle loss part are divided into a training set of a first ratio and a verification set of a second ratio;

   D. training the predetermined recognition model by using the training set;

   E. verifying the accuracy of the training recognition model by using the verification set, if the accuracy rate is greater than or equal to the preset accuracy rate, the training ends, or if the accuracy rate is less than the preset accuracy rate, increasing the corresponding position of each vehicle damage part The number of sample pictures and re-execute steps B, C, D, E above.
4. The method for identifying a vehicle damage portion according to claim 3, wherein the preset conversion rule comprises:

   Recognizing a label color pixel area and a corresponding vehicle damage part thereof in each training picture in which the label color is changed;

   Determining, according to a predetermined mapping relationship between the vehicle damage location and the identification data, identification data corresponding to the label color pixel region in each training image in which the label color is changed;

   Converts all the pixels except the label color pixel area in the training picture whose label color has been changed to the preset data, and converts each pixel in the label color pixel area in the training picture whose label color is changed to Corresponding identification data to get individual changed labels The color training image corresponds to the pixel area labeling matrix of the vehicle damage part.
5. The method for identifying a vehicle damage portion according to claim 3, wherein the step of performing image pre-processing on each sample picture to obtain a training picture to be model-trained comprises:

   Adjusting each sample picture to a first picture of a first preset size, and randomly cutting a second picture of a second preset size on each first picture;

   Performing a preset direction flipping on each second picture, and performing a twisting operation on each second picture according to a preset twist angle to obtain a third picture corresponding to each second picture;

   Calculating an average pixel picture of all second pictures and third pictures corresponding to each sample picture;

   A training picture corresponding to each sample picture is obtained based on the average pixel picture.
6. A vehicle damage part identification system, characterized in that the vehicle damage part identification system comprises:

   The identification module is configured to: if the car insurance claim photo uploaded by the first terminal is received, identify, by using a pre-trained recognition model, the pixel area of the car damage part of the car insurance claim photo; wherein the predetermined recognition model is a recognition model obtained by labeling and training a pixel area in a predetermined number of sample images of each vehicle damage part in advance;

   a sending module, configured to: identify the pixel area of the vehicle damage part in the car insurance claim photo, and identify the identified pixel area, and send the car insurance claim photo with the pixel area identifier to the first terminal And/or a predetermined second terminal, or intercepting the identified pixel area for transmission to the first terminal and/or the predetermined second terminal.
7. The vehicle damage part identification system according to claim 6, wherein said recognition model is a deep convolutional neural network model without a fully connected layer, said deep convolutional neural network model comprising an input layer, convolution Layer, pooling layer, upsampling layer, and cutting layer.
8. The vehicle damage part identification system according to claim 6 or 7, wherein the training process of the recognition model is as follows:

   A. Preparing a corresponding preset number of sample pictures for each preset vehicle damage part;

   B. Perform image preprocessing on each sample image to obtain a training picture to be trained by the model, and change the pixel color of the car damage part of each training picture to a corresponding label according to a preset mapping relationship between the car damage part and the label color. The color, and the training picture for each of the changed label colors is converted according to a preset conversion rule to generate a corresponding pixel area labeling matrix of the vehicle damage part;

   C. The training pictures with all the pixel area labeling matrix of the vehicle loss part are divided into a training set of a first ratio and a verification set of a second ratio;

   D. training the predetermined recognition model by using the training set;

   E. verifying the accuracy of the training recognition model by using the verification set, if the accuracy rate is greater than or equal to the preset accuracy rate, the training ends, or if the accuracy rate is less than the preset accuracy rate, increasing the corresponding position of each vehicle damage part The number of sample pictures and re-execute steps B, C, D, E above.
9. The vehicle damage part identification system according to claim 8, wherein the preset conversion rule comprises:

   Recognizing a label color pixel area and a corresponding vehicle damage part thereof in each training picture in which the label color is changed;

   Determining, according to a predetermined mapping relationship between the vehicle damage location and the identification data, identification data corresponding to the label color pixel region in each training image in which the label color is changed;

   Converts all the pixels except the label color pixel area in the training picture whose label color has been changed to the preset data, and converts each pixel in the label color pixel area in the training picture whose label color is changed to The corresponding identification data is used to obtain a pixel area labeling matrix of the vehicle damage part corresponding to each training picture whose label color is changed.
10. The vehicle damage part identification system according to claim 8, wherein the step of performing image pre-processing on each sample picture to obtain a training picture to be model-trained comprises:

    Adjusting each sample picture to a first picture of a first preset size, and randomly cutting a second picture of a second preset size on each first picture;

    Performing a preset direction flipping on each second picture, and performing a twisting operation on each second picture according to a preset twist angle to obtain a third picture corresponding to each second picture;

    Calculating an average pixel picture of all second pictures and third pictures corresponding to each sample picture;

    A training picture corresponding to each sample picture is obtained based on the average pixel picture.
11. An electronic device, comprising: a memory and a processor connected to the memory, wherein the memory stores a vehicle damage part identification system operable on the processor, the vehicle damage The part identification system is implemented by the processor to implement the following steps:

    If the car insurance claim photo uploaded by the first terminal is received, the pixel area of the car damage part is identified by using the pre-trained recognition model; wherein the predetermined recognition model is pre-adjusted to the preset a number of pixel regions in each sample of the vehicle damage part are marked and trained to obtain a recognition model;

    If the pixel area of the car damage part in the car insurance claim photo is identified, the identified pixel area is identified, and the car insurance claim photo with the pixel area identifier is sent to the first terminal and/or predetermined Or the second terminal, or the identified pixel area is intercepted and sent to the first terminal and/or the predetermined second terminal.
12. The electronic device according to claim 11, wherein said recognition model is a deep convolutional neural network model without a fully connected layer, said deep convolutional neural network model comprising an input layer, a convolution layer, and a pool Layer, upsampling layer and cutting layer.
13. The electronic device according to claim 11 or 12, wherein the training process of the recognition model is as follows:

    A. Preparing a corresponding preset number of sample pictures for each preset vehicle damage part;

    B. Perform image preprocessing on each sample image to obtain a training picture to be trained by the model, and change the pixel color of the car damage part of each training picture to a corresponding label according to a preset mapping relationship between the car damage part and the label color. The color, and the training picture for each of the changed label colors is converted according to a preset conversion rule to generate a corresponding pixel area labeling matrix of the vehicle damage part;

    C. The training pictures with all the pixel area labeling matrix of the vehicle loss part are divided into a training set of a first ratio and a verification set of a second ratio;

    D. training the predetermined recognition model by using the training set;

    E. verifying the accuracy of the training recognition model by using the verification set, if the accuracy rate is greater than or equal to the preset accuracy rate, the training ends, or if the accuracy rate is less than the preset accuracy rate, increasing the corresponding position of each vehicle damage part The number of sample pictures and re-execute steps B, C, D, E above.
14. The electronic device according to claim 13, wherein the preset conversion rule comprises:

    Recognizing a label color pixel area and a corresponding vehicle damage part thereof in each training picture in which the label color is changed;

    Determining, according to a predetermined mapping relationship between the vehicle damage location and the identification data, identification data corresponding to the label color pixel region in each training image in which the label color is changed;

    Converts all the pixels except the label color pixel area in the training picture whose label color has been changed to the preset data, and converts each pixel in the label color pixel area in the training picture whose label color is changed to The corresponding identification data is used to obtain a pixel area labeling matrix of the vehicle damage part corresponding to each training picture whose label color is changed.
15. The electronic device according to claim 13, wherein the step of performing image pre-processing on each sample picture to obtain a training picture to be model-trained comprises:

    Adjusting each sample picture to a first picture of a first preset size, and randomly cutting a second picture of a second preset size on each first picture;

    Performing a preset direction flipping on each second picture, and performing a twisting operation on each second picture according to a preset twist angle to obtain a third picture corresponding to each second picture;

    Calculating an average pixel picture of all second pictures and third pictures corresponding to each sample picture;

    A training picture corresponding to each sample picture is obtained based on the average pixel picture.
16. A computer readable storage medium, characterized in that the computer readable storage medium stores a vehicle damage part identification system, and when the vehicle damage part identification system is executed by the processor, the following steps are implemented:

    If the car insurance claim photo uploaded by the first terminal is received, the pixel area of the car damage part is identified by using the pre-trained recognition model; wherein the predetermined recognition model is pre-adjusted to the preset a number of pixel regions in each sample of the vehicle damage part are marked and trained to obtain a recognition model;

    If the pixel area of the vehicle damage part in the car insurance claim photo is identified, the recognized pixel is The area is identified, and the car insurance claim photo with the pixel area identifier is sent to the first terminal and/or the predetermined second terminal, or the identified pixel area is intercepted and sent to the first terminal and / or a predetermined second terminal.
17. The computer readable storage medium of claim 16 wherein said recognition model is a deep convolutional neural network model without a fully connected layer, said deep convolutional neural network model comprising an input layer, convolution Layer, pooling layer, upsampling layer, and cutting layer.
18. The computer readable storage medium according to claim 16 or 17, wherein the training process of the recognition model is as follows:

    A. Preparing a corresponding preset number of sample pictures for each preset vehicle damage part;

    B. Perform image preprocessing on each sample image to obtain a training picture to be trained by the model, and change the pixel color of the car damage part of each training picture to a corresponding label according to a preset mapping relationship between the car damage part and the label color. The color, and the training picture for each of the changed label colors is converted according to a preset conversion rule to generate a corresponding pixel area labeling matrix of the vehicle damage part;

    C. The training pictures with all the pixel area labeling matrix of the vehicle loss part are divided into a training set of a first ratio and a verification set of a second ratio;

    D. training the predetermined recognition model by using the training set;

    E. verifying the accuracy of the training recognition model by using the verification set, if the accuracy rate is greater than or equal to the preset accuracy rate, the training ends, or if the accuracy rate is less than the preset accuracy rate, increasing the corresponding position of each vehicle damage part The number of sample pictures and re-execute steps B, C, D, E above.
19. The computer readable storage medium of claim 18, wherein the predetermined conversion rule comprises:

    Recognizing a label color pixel area and a corresponding vehicle damage part thereof in each training picture in which the label color is changed;

    Determining, according to a predetermined mapping relationship between the vehicle damage location and the identification data, identification data corresponding to the label color pixel region in each training image in which the label color is changed;

    Converts all the pixels except the label color pixel area in the training picture whose label color has been changed to the preset data, and converts each pixel in the label color pixel area in the training picture whose label color is changed to The corresponding identification data is used to obtain a pixel area labeling matrix of the vehicle damage part corresponding to each training picture whose label color is changed.
20. The computer readable storage medium according to claim 18, wherein the step of performing image preprocessing on each sample picture to obtain a training picture to be model trained comprises:

    Adjusting each sample picture to a first picture of a first preset size, and randomly cutting a second picture of a second preset size on each first picture;

    Performing a preset direction flipping on each second picture, and performing a twisting operation on each second picture according to a preset twist angle to obtain a third picture corresponding to each second picture;

    Calculating an average pixel picture of all second pictures and third pictures corresponding to each sample picture;

    A training picture corresponding to each sample picture is obtained based on the average pixel picture.

Images