WO2021000489A1 - Vehicle image classification method and apparatus, and computer device and storage medium - Google Patents

Abstract

A vehicle image classification method and apparatus, and a computer device and a storage medium. The method comprises: acquiring a vehicle loss assessment data set, wherein the vehicle loss assessment data set comprises at least two vehicle loss assessment images (S10); identifying each vehicle loss assessment image in the vehicle loss assessment data set to obtain attribute information and damaged point information of each vehicle loss assessment image (S20); classifying, according to the attribute information, the vehicle loss assessment image to obtain a global image set and a local image set (S30); setting an initial clustering cluster according to attribute information and damaged point information of each global image, and performing, according to the initial clustering cluster, clustering analysis on the vehicle loss assessment data set to obtain different vehicle classification sets (S40); and performing an image similarity calculation on vehicle classification images in each vehicle classification set, and eliminating repeated vehicle classification images according to the image similarity (S50). By means of the method, vehicle loss assessment images in a vehicle loss assessment data set are intelligently classified, the image classification efficiency is improved, and the image classification accuracy is ensured.

Description

Vehicle image classification method, device, computer equipment and storage medium

This application is based on the Chinese invention patent application filed on July 3, 2019 with the application number 201910596017.0, titled "Vehicle image classification method, device, computer equipment and storage medium", and claims its priority.

Technical field

This application relates to the field of image processing, and in particular to a vehicle image classification method, device, computer equipment and storage medium.

Background technique

With the increasing popularity of various vehicles, the proportion of vehicles on the road is increasing. Correspondingly, various traffic accidents also happen occasionally. When a vehicle is involved in a traffic accident, it is often necessary to investigate and determine the damage of the vehicle in the traffic accident. In the process of vehicle damage assessment, image acquisition of vehicle damage is required.

In the current damage assessment process, the damage assessor will use image acquisition tools to capture the damage of the vehicle, and the number of images related to the damage assessment of the vehicle collected every day is very large. Because of the need to classify the collected images of different traffic accidents. Manual classification by the loss assessor will take a lot of time. Moreover, the manual classification method is easily affected by various subjective factors and cannot guarantee the accuracy of vehicle image classification.

Summary of the invention

The embodiments of the present application provide a vehicle image classification method, device, computer equipment, and storage medium to solve the problem of low vehicle image classification efficiency.

A vehicle image classification method includes:

Acquiring a vehicle loss assessment data set, where the vehicle loss assessment data set includes at least two vehicle loss assessment images;

Recognize each vehicle damage assessment image in the vehicle damage assessment data set, and obtain the attribute information and damage point information of each vehicle damage assessment image;

Classify the vehicle loss-specific image according to the attribute information to obtain a global image set and a partial image set;

Setting an initial cluster cluster according to the attribute information and damaged point information of each of the global images, and performing a cluster analysis on the vehicle damage data set according to the initial cluster cluster to obtain different vehicle classification sets;

Perform image similarity calculation on the vehicle classification images in each vehicle classification set, and eliminate duplicate vehicle classification images based on the image similarity.

A vehicle image classification device, including:

The first data set acquisition module is configured to acquire a vehicle loss assessment data set, where the vehicle loss assessment data set includes at least two vehicle loss assessment images;

The first image recognition module is used for recognizing each vehicle's loss-fixing image in the vehicle's loss-fixing data set to obtain attribute information and damaged point information of each vehicle's loss-fixing image;

The first image classification module is configured to classify the vehicle loss-specific image according to the attribute information to obtain a global image set and a partial image set;

The first classification set acquisition module is used to set initial clusters according to the attribute information and damaged point information of each of the global images, and perform cluster analysis on the vehicle damage data set according to the initial clusters to obtain different Vehicle classification set;

The first image elimination module is used to calculate the image similarity of the vehicle classification images in each vehicle classification set, and eliminate the repeated vehicle classification images according to the image similarity.

A vehicle image classification method includes:

Acquiring a vehicle loss assessment data set, the vehicle loss assessment data set including a global image and a local image;

Performing classification processing on the vehicle damage data set to obtain a global image set and a partial image set;

Use the preset recognition model to identify each vehicle damage image in the vehicle damage data set, and obtain the attribute information and damage point information of each vehicle damage image;

Setting an initial cluster cluster according to the attribute information and damaged point information of each of the global images, and performing cluster analysis on a local image set according to the initial cluster cluster to obtain different vehicle classification sets;

Carry out vehicle similarity calculation on the vehicle images in each vehicle classification set, and eliminate duplicate vehicle images according to the vehicle similarity.

A vehicle image classification device, including:

The second data set acquisition module is configured to acquire a vehicle loss assessment data set, the vehicle loss assessment data set including global images and partial images;

The second image classification module is used to classify the vehicle loss assessment data set to obtain a global image set and a local image set;

The second image recognition module is used to recognize each vehicle damage image in the vehicle damage data set by using a preset recognition model to obtain attribute information and damage point information of each vehicle damage image;

The second classification set acquisition module is used to set initial clusters according to the attribute information and damaged point information of each of the global images, and perform cluster analysis on the local image sets according to the initial clusters to obtain different vehicles Classification set

The second image elimination module is used to calculate the vehicle similarity of the vehicle images in each vehicle classification set, and eliminate the repeated vehicle images according to the vehicle similarity.

A computer device comprising a memory, a processor, and computer readable instructions stored in the memory and capable of running on the processor, and the processor implements the above vehicle image classification method when the computer readable instructions are executed .

One or more readable storage media storing computer readable instructions, and when the computer readable instructions are executed by one or more processors, the one or more processors execute the aforementioned vehicle image classification method.

The details of one or more embodiments of the present application are presented in the following drawings and description, and other features and advantages of the present application will become apparent from the description, drawings and claims.

Description of the drawings

In order to explain the technical solutions of the embodiments of the present application more clearly, the following will briefly introduce the drawings that need to be used in the description of the embodiments of the present application. Obviously, the drawings in the following description are only some embodiments of the present application. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative labor.

FIG. 1 is a schematic diagram of an application environment of a vehicle image classification method in an embodiment of the present application;

Fig. 2 is a flowchart of a vehicle image classification method in an embodiment of the present application;

FIG. 3 is another flowchart of a method for classifying vehicle images in an embodiment of the present application;

FIG. 4 is another flowchart of a vehicle image classification method in an embodiment of the present application;

FIG. 5 is another flowchart of a method for classifying vehicle images in an embodiment of the present application;

Fig. 6 is another flowchart of a vehicle image classification method in an embodiment of the present application;

Fig. 7 is a functional block diagram of a vehicle image classification device in an embodiment of the present application;

FIG. 8 is another principle block diagram of the vehicle image classification device in an embodiment of the present application;

FIG. 9 is another principle block diagram of the vehicle image classification device in an embodiment of the present application;

Fig. 10 is a schematic diagram of a computer device in an embodiment of the present application.

Detailed ways

The technical solutions in the embodiments of the present application will be described clearly and completely in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, rather than all of them. Based on the embodiments in this application, all other embodiments obtained by those of ordinary skill in the art without creative work shall fall within the protection scope of this application.

The vehicle image classification method provided by the embodiments of the present application can be applied in an application environment as shown in FIG. 1, in which a client (computer device) communicates with a server through a network. The server obtains a vehicle loss assessment data set, which includes at least two vehicle loss assessment images; recognizes each vehicle loss assessment image in the vehicle loss assessment data set to obtain the attributes of each vehicle loss assessment image Information and damaged point information; classify the vehicle damage-related images according to the attribute information to obtain a global image set and a partial image set; set initial clustering according to the attribute information and damaged point information of each global image Clusters, perform cluster analysis on the vehicle damage data set according to the initial cluster clusters to obtain different vehicle classification sets; perform image similarity calculations on the vehicle classification images in each vehicle classification set, and calculate the repeated Vehicle classification images are eliminated. Among them, the client (computer equipment) can be, but not limited to, various personal computers, notebook computers, smart phones, tablet computers, and portable wearable devices. The server can be implemented as an independent server or a server cluster composed of multiple servers.

In one embodiment, as shown in FIG. 2, a vehicle image classification method is provided, and the method is applied to the server in FIG. 1 as an example for description, including the following steps:

S10: Acquire a vehicle loss assessment data set, where the vehicle loss assessment data set includes at least two vehicle loss assessment images.

Among them, the vehicle loss assessment data set is an image set composed of a large number of vehicle loss assessment images. Vehicle damage determination is when the insured vehicle has a traffic accident, the relevant unit conducts on-site inspection and determination of the damage. The vehicle damage determination involves many aspects such as maintenance, manufacturing, and loss of the owner. The vehicle damage assessment image is image information related to the damage assessment vehicle, which can include vehicle information (license plate number), images of damaged areas of the vehicle, and the environment in which the vehicle is located. Understandably, there are usually at least two vehicle damage images of a vehicle.

The vehicle damage data set can be all images collected by the same image collection tool, or all images collected by the same image collection tool within a predetermined time (for example, one day), or all images collected by at least two image collection tools. Or all images collected by at least two image collection tools within a predetermined time (for example, one day). An image collection tool may be a client, or at least two image collection tools both upload the vehicle's loss-specific images to the server through the corresponding client or the same client. The upload timing of the vehicle's fixed-loss images can be real-time or scheduled.

S20: Recognizing each vehicle loss-fixing image in the vehicle loss-fixing data set to obtain attribute information and damaged point information of each vehicle loss-fixing image.

Attribute information refers to the information of the corresponding vehicle in the vehicle damage image, such as the license plate number, vehicle color, and/or vehicle model. Further, the attribute information may also include related information of the vehicle damage-assessment image, such as shooting time, shooting location, shooting equipment, etc. This part of the attribute information can be obtained by identifying the vehicle damage-assessment image itself. Specifically, this part of the attribute information can be integrated into the vehicle's loss-making image by means of watermarking, and this part of the attribute information can be obtained by performing text recognition on the vehicle's loss-making image.

Damaged point information refers to information indicating the specific damaged part or damaged type of the vehicle. Damaged parts can be divided according to different parts of the vehicle, such as: hood, headlights, bumpers, rearview mirrors, front doors, rear doors, or window glass. Furthermore, each part can be further divided according to actual accuracy requirements. For example, the front door of the car is divided into a left front door and a right front door, and the rear of the car is divided into a left rear door and a right rear door.

In this step, the identification of each vehicle loss-fixing image in the vehicle loss-fixing data set can be realized through a preset recognition model. The attribute information includes license plate number information, vehicle color, vehicle model, etc. However, not all of the above attribute information can be identified in every vehicle damage image. If a vehicle damage image is only a partial detail image, it may be possible The vehicle model in the damage image of the vehicle cannot be recognized. If a vehicle damage image does not capture the license plate number, the license plate number information in the vehicle damage image cannot be identified. The color of the vehicle can generally be identified. Therefore, a recognition model can be used to identify the vehicle damage image to obtain the attribute information of the vehicle damage image. If one item of attribute information is successfully recognized, the specific information of the attribute information will be directly output, if the attribute information is recognized If it fails, the attribute information of the item is not output. Specifically, a license plate number recognition model, a vehicle color recognition model, and a vehicle model recognition model can be separately established in advance, and then each vehicle damage image is input into these three recognition models for recognition, and corresponding attribute information is obtained. If the identification of the attribute information fails, the attribute information can be left blank or represented by the same symbol, which will not be repeated here. Among them, the license plate number recognition model, the vehicle color recognition model, and the vehicle model recognition model can all be established by machine learning, and will not be repeated here.

Similarly, the damaged point information of the vehicle damage image can also be recognized by pre-setting a damaged point recognition model. However, some damaged images of vehicles may not contain damaged parts. Therefore, the damaged point information may be specific damaged parts or none. Understandably, there may be at least two damaged point information.

S30: Classify the vehicle loss-specific image according to the attribute information to obtain a global image set and a partial image set.

In this step, the vehicle damage image is classified by attribute information, specifically, the vehicle damage image is classified according to the amount of attribute information, or the specific one or at least two attributes in the attribute information are classified. The state of the information is used to classify the vehicle damage image. For example, if the number of attribute information is 5 items, the vehicle loss-making images with attribute information greater than or equal to 4 items can be classified into the global image set, and the vehicle loss-making images with attribute information less than 4 items can be classified into the local image set. Or, classify the vehicle loss-based images containing the two attribute information of the license plate number and the vehicle model into the global image set, and classify the remaining vehicle loss-based images into the local image set.

S40: Set an initial cluster cluster according to the attribute information and damaged point information of each of the global images, and perform a cluster analysis on the vehicle damage data set according to the initial cluster cluster to obtain different vehicle classification sets.

Specifically, the attribute information and damaged point information of each global image is converted into a feature vector, and then the feature vector corresponding to the global image is set as an initial cluster cluster. The number of initial cluster clusters can be equal to the number of global images. equal. Then perform cluster analysis on the vehicle damage data set according to the set initial cluster clusters to obtain different vehicle classification sets. Specifically, a clustering algorithm can be used to perform clustering analysis on the vehicle damage data set.

S50: Perform image similarity calculation on the vehicle classification images in each vehicle classification set, and eliminate the repeated vehicle classification images according to the image similarity.

By calculating the image similarity of the vehicle classification images in the same vehicle classification set, if the image similarity of two vehicle classification images exceeds the preset similarity threshold, it will face any vehicle in the two vehicle classification images Categorize the image to remove.

In this embodiment, first obtain a vehicle loss assessment data set, the vehicle loss assessment data set includes at least two vehicle loss assessment images; identify each vehicle loss assessment image in the vehicle loss assessment data set, and obtain each vehicle Attribute information and damaged point information of the fixed-loss image; classify the vehicle fixed-loss image according to the attribute information to obtain a global image set and a partial image set; according to the attribute information and damaged point of each of the global images The information sets the initial cluster cluster, and performs cluster analysis on the vehicle damage data set according to the initial cluster cluster to obtain different vehicle classification sets; image similarity calculation is performed on the vehicle classification images in each vehicle classification set, according to the image The similarity eliminates duplicate vehicle classification images. The intelligent classification of the vehicle loss-fixing images in the vehicle loss-fixing data set improves the efficiency of image classification and ensures the accuracy of image classification.

In an embodiment, the vehicle damage assessment image includes a damaged area.

Specifically, the damaged area is the area where the damaged part on the vehicle damage image is located. The damaged part in the image can be marked when the vehicle damage image is collected, for example, a rectangular frame is added to the area where the damaged part is located.

In this embodiment, as shown in FIG. 3, the identification of each vehicle damage image in the vehicle damage data set to obtain the attribute information and damaged point information of each vehicle damage image includes:

S21: Use a preset damaged point recognition model to identify the damaged area of each vehicle's damage assessment image, and obtain the damaged point information of each vehicle's damage assessment image.

The damaged point identification model is a model for identifying specific vehicle parts in the damaged area. A large number of vehicle images with damaged parts can be collected in advance to train the convolutional neural network to obtain the damaged point recognition model. The damaged point information may be a specific damaged part. Understandably, the damaged point information may be at least two, that is, the damaged part may be at least two.

S22: Use the semantic segmentation model to process each vehicle's fixed-loss image to obtain image information of each vehicle's fixed-loss image.

The semantic segmentation model is used to process the vehicle's fixed-loss image to obtain the image information of each vehicle's fixed-loss image. Specifically, a semantic segmentation model is obtained through training in advance, and the segmentation model is used to determine whether there is a clearly recognizable license plate area in the vehicle damage image and whether there is a complete vehicle image available for vehicle model recognition to obtain image information. The image information is used to indicate whether the vehicle license plate number can be recognized and whether the vehicle model can be recognized in the vehicle damage image. Specifically, a simple number can be used to identify whether the vehicle license plate number can be recognized and whether the vehicle model can be recognized in the corresponding vehicle damage image. For example, "00", "01", "10" and "11" are used to identify different situations respectively. Among them, 1 means available for identification, 0 means unidentifiable. The two positions can be customized to indicate whether the license plate number can be recognized and whether the vehicle model can be recognized. There is no specific limitation here.

S23: Acquire a corresponding target recognition model according to the image information of each vehicle's damage-fixing image, and use the target recognition model to perform recognition processing on the vehicle's damage-fixing image to obtain attribute information of the vehicle's damage-fixing image.

After the image information is obtained, the corresponding target recognition model is obtained according to the image information. Specifically, when the corresponding image information is recognizable, the corresponding recognition model is acquired. For example, license plate number recognition model, vehicle color recognition model and vehicle model recognition model. If the image information indicates that the license plate number can be recognized and the vehicle model can be recognized, then the license plate number recognition model and the vehicle model recognition model are acquired as the target recognition model, and the target recognition model is used to recognize the vehicle damage image , To obtain the attribute information of the vehicle damage image. Further, it can be assumed that each vehicle damage image can perform vehicle color recognition, that is, the target recognition model must include the vehicle color recognition model.

In this embodiment, the sequence of step S21 may be executed before steps S22-S23, or may be executed after steps S22-S23.

In this embodiment, by using a preset damaged point recognition model to identify the damaged area of each vehicle's damage assessment image, the damaged point information of each vehicle's damage assessment image is obtained; the semantic segmentation model is used for each Carrying out the processing of the vehicle's loss-fixing image to obtain the image information of each vehicle's loss-fixing image; finally obtain the corresponding target recognition model according to the image information of each vehicle's loss-fixing image, and use the target recognition model to assess the damage of the vehicle The image is subjected to recognition processing to obtain the attribute information of the damage-oriented image of the vehicle. The semantic segmentation model is used to recognize the vehicle damage image, and then the corresponding target recognition model is used to recognize the attribute information according to the recognition result, which avoids unnecessary recognition process and further improves the recognition efficiency.

In one embodiment, as shown in FIG. 4, the classification of the vehicle loss-specific image according to the attribute information to obtain a global image set and a partial image set includes:

S31: Obtain preset classification information, and match the attribute information of each vehicle's loss-specific image according to the classification information.

Specifically, the classification information can be embodied in different ways, and the classification information can be a specific value or a vector. If the classification information is a specific numerical value, then the vehicle damage image is classified by the number of attribute information. That is, matching the attribute information of each of the vehicle's damage-quality images according to the classification information is a match between the number of attribute information and the classification information. For example, if the number of attribute information is 5 items, and if the classification information is set to 4, then the vehicle loss-assessment images with attribute information greater than or equal to 4 items can be classified into the global image set, and the vehicles with attribute information less than 4 items can be classified The images are classified into local image sets.

Further, if the classification information is a vector, the one-hot vector conversion is performed in advance for the attribute information of each vehicle's damage image. If the corresponding attribute information exists, the value of the corresponding position in the one-hot vector is 1, otherwise it is 0. At this time, the vector in the classification information is to classify the vehicle damage image according to the state of a specific item or at least two items of the attribute information in the attribute information. For example, if the classification information is A=[1, 1, 0, 0, 0]. It indicates that the attribute value of the first item and the second item in the attribute information is 1, and the vehicle loss-based image will be classified into the global image set, otherwise the vehicle loss-based image will be classified into the local image set. At this time, the classification information and the one-hot vector of the attribute information of each vehicle's damage-related image can be calculated for the vector distance, and then the matching result can be obtained according to the calculation result.

S32: If the attribute information and the classification information are matched successfully, put the corresponding vehicle lossy image into the global image set.

S33: If the attribute information and the classification information fail to match, put the corresponding vehicle damage image into the partial image set.

In this embodiment, a preset classification information is matched with the attribute information of each vehicle loss-making image, and the vehicle loss-making image is classified according to the matching result, which improves the flexibility and accuracy of image classification.

In one embodiment, as shown in FIG. 5, the initial clusters are set according to the attribute information and damaged point information of each of the global images, and the local image sets are clustered according to the initial clusters. , Get different vehicle classification sets, including:

S41: Transform the attribute information and damaged point information of each vehicle's damage-fixing image into feature vector to obtain the feature vector of each vehicle's damage-fixing image.

The feature vector transformation of both the attribute information and the damaged point information in the vehicle damage image can be achieved by word vector. The role of word vectors is to convert words in natural language into dense vectors that computers can understand. Specifically, the input text is first formed by the attribute information and damaged point information of all vehicle loss-specific images, and then a vocabulary is generated using the input text. The word frequency of each word is counted, and the word frequency is sorted from high to low, and the most frequent V Words form a vocabulary. There is a one-hot vector for each word, and the dimension of the vector is V. If the word has appeared in the vocabulary, the corresponding position in the vocabulary in the vector is 1, and the other positions are all 0. If it does not appear in the vocabulary, the vector is all zeros. The one-hot vector is expressed as a feature vector of an attribute, that is, there is only one activation point (not 0) at the same time, this vector has only one feature that is not 0, and the others are 0.

In this step, a word vector can be constructed separately for each attribute information and each damaged point information to better characterize the accuracy of each word vector. Then, each attribute information and the word vector of each damaged point information form the loss-fixing feature vector of the vehicle's loss-fixing image.

S42: Set the loss-fixing feature vector of each of the global images as initial clustering points, and perform a cluster analysis on the set of vehicle loss-fixing images according to the initial clustering points using a clustering algorithm to obtain different vehicles Cluster clusters.

In this step, the initial clustering points are set by the loss-based feature vector of the global image. Since the global image contains the most information, the initial clustering points can be set by the loss-based feature vector corresponding to the global image for subsequent Perform cluster analysis.

Specifically, among them, cluster analysis is also called cluster analysis, which is a statistical analysis method for studying (sample or index) classification problems, and it is also an important analysis method for data mining. Optionally, the clustering algorithm may be implemented by algorithms such as K-Means (K-means) clustering, mean shift clustering, density-based clustering method, or agglomerated hierarchical clustering.

Preferably, a K-Means (K-means) clustering algorithm is used to perform cluster analysis on the partial image set to obtain the target center point. Specifically, the K value is set according to the number of global images, and the loss-constant feature vector of each global image is set as the initial cluster point. After all the points (vehicle damage image) have been allocated, recalculate (for example, calculate the average value) of all points in the cluster to obtain the new center point of the cluster. Then, the steps of assigning center points and updating the center points of the clusters are performed in an iterative manner until the center points of the clusters change very little or reach the specified number of iterations. The cluster clusters meeting the preset number are regarded as the vehicle cluster clusters.

S43: Calculate the vector distance from the loss-fixing feature vector of the loss-fixing image of each vehicle in each vehicle cluster to the center of the cluster cluster.

The distance algorithm can be used to calculate the vector distance from the loss-fixing feature vector of the loss-fixing image of each vehicle in each vehicle cluster to the center of the cluster cluster, for example: Euclidean distance algorithm, Manhattan distance algorithm, Chebyshev distance algorithm, Minkov Skye distance algorithm, standardized Euclidean distance algorithm, Mahalanobis distance or Hamming distance algorithm. Further, different weights can be set for different information in the loss-based feature vector to better calculate the vector distance. For example, the weights of the license plate number information and damaged point information can be set higher, and the weight of other information can be set higher. The weight is appropriately lowered, or adjusted according to actual needs.

S44: Determine the vehicle's loss-specific image whose vector distance exceeds a preset vector threshold as an image to be determined.

The vector threshold is a preset value. According to the preset vector threshold, it can be judged whether a vehicle damage image belongs to the vehicle cluster. The vector threshold can be obtained after testing with multiple samples, or it can be set based on empirical values. If the vector distance between a vehicle damage image and a vehicle cluster cluster exceeds the preset vector threshold, it means that the vehicle damage image may not belong to the vehicle cluster cluster, so the corresponding vehicle damage image is determined as the image to be determined .

S45: Send the to-be-determined image to the client, and obtain the classification information returned by the client.

The classification information is the information of the vehicle cluster cluster to which the corresponding to-be-determined image is allocated, which is fed back by the client. In this step, the image to be determined is sent to the client, and the classification information returned by the client is obtained.

S46: Assign each of the to-be-determined images to corresponding vehicle clusters according to the classification information.

With the help of classification information, the server assigns each to-be-determined image to the corresponding vehicle cluster.

S47: Combine the vehicle damage-related images in each vehicle cluster into a vehicle classification set.

Finally, the vehicle damage assessment images in each vehicle cluster are formed into a vehicle classification set. Each vehicle classification set represents a vehicle damage assessment image corresponding to a different vehicle damage assessment case.

In this embodiment, the attribute information and damaged point information of each vehicle's loss-based image are transformed into feature vectors to obtain the loss-based feature vector of each vehicle's loss-based image; The feature vector is set as the initial clustering point, and the clustering algorithm is used to perform cluster analysis on the set of vehicle damage image sets according to the initial clustering point to obtain different vehicle clusters; then each vehicle cluster is calculated The vector distance from the loss-fixing feature vector of each vehicle loss-fixing image to the cluster center; determining the loss-fixing image of the vehicle whose vector distance exceeds the preset vector threshold as the image to be determined; sending the image to be determined To the client, obtain the classification information returned by the client; assign each of the to-be-determined images to the corresponding vehicle cluster according to the classification information; finally, the vehicle in each vehicle cluster is assessed for damage The images make up the vehicle classification set. It can intelligently classify the vehicle damage image, ensuring the efficiency and accuracy of image classification.

The embodiment of the present application also relates to a vehicle image classification method. The application of the method to the server in FIG. 1 is taken as an example for description. As shown in FIG. 6, the vehicle image classification method includes:

S10': Acquire a vehicle loss assessment data set, where the vehicle loss assessment data set includes a global image and a partial image.

Among them, the vehicle loss assessment data set is an image set composed of a large number of vehicle loss assessment images. Vehicle damage determination is when the insured vehicle has a traffic accident, the relevant unit conducts on-site inspection and determination of the damage. The vehicle damage determination involves many aspects such as maintenance, manufacturing, and loss of the owner. The vehicle damage assessment image is image information related to the damage assessment vehicle, which can include vehicle information (license plate number), images of damaged areas of the vehicle, and the environment in which the vehicle is located. Understandably, there are usually at least two vehicle damage images of a vehicle.

The vehicle damage data set can be all images collected by the same image collection tool, or all images collected by the same image collection tool within a predetermined time (for example, one day), or all images collected by at least two image collection tools. Or all images collected by at least two image collection tools within a predetermined time (for example, one day). An image collection tool may be a client, or at least two image collection tools both upload the vehicle's loss-based image to the server through the corresponding client or the same client. The upload timing of the vehicle's fixed-loss image can be real-time or regularly.

Specifically, the vehicle damage data set includes a global image and a local image. The global image is an image that reflects the global information of the corresponding vehicle, and the local image is an image that reflects the local information of the corresponding vehicle. In the vehicle loss assessment data set, each vehicle loss assessment image has a corresponding image label to indicate that the image is a global image or a partial image.

S20': Perform classification processing on the vehicle loss-fixing data set to obtain a global image set and a partial image set.

Specifically, according to the image label of each vehicle loss-fixing image, the vehicle loss-fixing data set is classified, the global image is divided into a global image set, and the local image is divided into a local image set to obtain a global image set and a local image set. Image set.

S30': Use the preset recognition model to identify each vehicle damage image in the vehicle damage data set, and obtain the attribute information and damage point information of each vehicle damage image.

Specifically, this step is the same as step S20 in the foregoing embodiment, and will not be repeated here.

S40': Set initial clusters according to the attribute information and damaged point information of each of the global images, and perform cluster analysis on the partial image sets according to the initial clusters to obtain different vehicle classification sets.

S50': Perform vehicle similarity calculation on the vehicle images in each vehicle classification set, and eliminate duplicate vehicle images according to the vehicle similarity.

The above steps S40' and S50' are the same as the steps S40 and S50 in the above embodiment, and will not be repeated here.

In this embodiment, different image labels are pre-configured for each image in the vehicle loss assessment data set to better classify the vehicle loss assessment data set, and to better improve the efficiency of vehicle classification.

It should be understood that the size of the sequence number of each step in the foregoing embodiment does not mean the order of execution. The execution sequence of each process should be determined by its function and internal logic, and should not constitute any limitation to the implementation process of the embodiment of the present application.

In one embodiment, a vehicle image classification device is provided, and the vehicle image classification device corresponds to the vehicle image classification method in the above-mentioned embodiment one-to-one. As shown in FIG. 7, the vehicle image classification device includes a first data set acquisition module 10, a first image recognition module 20, a first image classification module 30, a first classification set acquisition module 40 and a first image removal module 50. The detailed description of each functional module is as follows:

The first data set acquisition module 10 is configured to acquire a vehicle loss assessment data set, the vehicle loss assessment data set including at least two vehicle loss assessment images.

The first image recognition module 20 is used to identify each vehicle damage image in the vehicle damage data set to obtain attribute information and damaged point information of each vehicle damage image.

The first image classification module 30 is configured to classify the vehicle loss-specific image according to the attribute information to obtain a global image set and a partial image set.

The first classification set acquisition module 40 is configured to set an initial cluster cluster according to the attribute information and damaged point information of each of the global images, and perform cluster analysis on the vehicle loss assessment data set according to the initial cluster cluster to obtain Different vehicle classification set.

The first image culling module 50 is used to perform image similarity calculation on the vehicle classification images in each vehicle classification set, and to eliminate duplicate vehicle classification images according to the image similarity.

Preferably, as shown in FIG. 8, the first classification set acquisition module 40 includes:

The vector conversion unit 41 is used to transform the attribute information and the damaged point information of each vehicle's damage assessment image into feature vector to obtain the damage feature vector of each vehicle's damage assessment image.

The cluster cluster acquisition unit 42 is configured to set the loss-specific feature vector of each global image as an initial clustering point, and use a clustering algorithm to cluster the set of vehicle loss-based images according to the initial clustering point. Class analysis to obtain different vehicle clusters.

The vector distance calculation unit 43 is used to calculate the vector distance from the loss-fixing feature vector of each vehicle loss-making image in each vehicle cluster to the center of the cluster cluster.

The image dividing unit 44 is configured to determine the vehicle's loss-of-loss image whose vector distance exceeds a preset vector threshold as the image to be determined.

The classification information obtaining unit 45 is configured to send the to-be-determined image to the client, and obtain the classification information returned by the client.

The image allocation unit 46 is configured to allocate each of the to-be-determined images to the corresponding vehicle cluster according to the classification information.

The classification set composing unit 47 is used to compose the vehicle classification set of the vehicle damage-related images in each vehicle cluster.

Preferably, the vehicle damage assessment image includes a damaged area. The first image recognition module 20 includes:

The damaged area recognition unit is used to recognize the damaged area of each vehicle's damage assessment image by using a preset damaged point recognition model to obtain the damaged point information of each vehicle's damage assessment image.

The image information acquisition unit is used to process each vehicle's fixed-loss image by using the semantic segmentation model to obtain image information of each vehicle's fixed-loss image.

The attribute information acquisition unit is configured to acquire the corresponding target recognition model according to the image information of each vehicle damage image, and use the target recognition model to identify the vehicle damage image to obtain the vehicle damage image Attribute information.

Preferably, the first image classification module 30 includes:

The classification information acquisition unit is configured to acquire preset classification information, and match the attribute information of each vehicle's loss-based image according to the classification information.

The classification unit is configured to put the corresponding vehicle loss-specific image into the global image collection when the attribute information and the classification information match successfully; when the attribute information and the classification information fail to match, then the corresponding The vehicle's fixed-loss image is put into the partial image set.

In one embodiment, a vehicle image classification device is provided, and the vehicle image classification device corresponds to the vehicle image classification method in the above-mentioned embodiment one-to-one. As shown in FIG. 9, the vehicle image classification device includes a second data set acquisition module 10', a second image classification module 20', a second image recognition module 30', a second classification set acquisition module 40', and a second image removal module. Module 50'. The detailed description of each functional module is as follows:

The second data set acquisition module 10' is used to acquire a vehicle loss assessment data set. The vehicle loss assessment data set includes global images and partial images.

The second image classification module 20' is used to classify the vehicle damage data set to obtain a global image set and a local image set.

The second image recognition module 30' is used to identify each vehicle damage image in the vehicle damage data set using a preset recognition model to obtain attribute information and damage point information of each vehicle damage image.

The second classification set acquisition module 40' is configured to set initial clusters according to the attribute information and damaged point information of each of the global images, and perform cluster analysis on the local image sets according to the initial clusters to obtain different Set of vehicle classifications.

The second image rejection module 50' is used to calculate the vehicle similarity of the vehicle images in each vehicle classification set, and eliminate the repeated vehicle images according to the vehicle similarity.

For the specific definition of the vehicle image classification device, please refer to the above definition of the vehicle image classification device method, which will not be repeated here. Each module in the above-mentioned vehicle image classification device can be implemented in whole or in part by software, hardware, and a combination thereof. The foregoing modules may be embedded in the form of hardware or independent of the processor in the computer device, or may be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the foregoing modules.

In one embodiment, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as shown in FIG. 10. The computer equipment includes a processor, a memory, a network interface and a database connected through a system bus. Among them, the processor of the computer device is used to provide calculation and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, computer readable instructions, and a database. The internal memory provides an environment for the operation of the operating system and computer-readable instructions in the non-volatile storage medium. The database of the computer equipment is used for the data used in the vehicle image classification method in the above embodiment. The network interface of the computer device is used to communicate with an external terminal through a network connection. The computer-readable instructions are executed by the processor to realize a vehicle image classification method.

In one embodiment, a computer device is provided, including a memory, a processor, and computer-readable instructions stored in the memory and capable of running on the processor. When the processor executes the computer-readable instructions, Vehicle image classification method.

In one embodiment, one or more readable storage media storing computer readable instructions are provided, and when the computer readable instructions are executed by one or more processors, the one or more processors execute The vehicle image classification method in the above embodiment. Wherein, the readable storage medium includes a non-volatile readable storage medium and a volatile readable storage medium.

A person of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be implemented by instructing relevant hardware through computer-readable instructions, which can be stored in a non-volatile computer. In a readable storage medium, when the computer-readable instructions are executed, they may include the processes of the above-mentioned method embodiments. Wherein, any reference to memory, storage, database or other media used in the embodiments provided in this application may include non-volatile and/or volatile memory. Non-volatile memory may include read only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not a limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous chain Channel (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

Those skilled in the art can clearly understand that for the convenience and conciseness of description, only the division of the above-mentioned functional units and modules is used as an example. In practical applications, the above-mentioned functions can be allocated to different functional units and modules as required. Module completion means dividing the internal structure of the device into different functional units or modules to complete all or part of the functions described above.

The above-mentioned embodiments are only used to illustrate the technical solutions of the present application, not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that it can still implement the foregoing The technical solutions recorded in the examples are modified, or some of the technical features are equivalently replaced; these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the application, and should be included in Within the scope of protection of this application.

Claims (20)

A vehicle image classification method, characterized in that it includes: Acquiring a vehicle loss assessment data set, where the vehicle loss assessment data set includes at least two vehicle loss assessment images; Recognize each vehicle damage assessment image in the vehicle damage assessment data set, and obtain the attribute information and damage point information of each vehicle damage assessment image; Classify the vehicle loss-specific image according to the attribute information to obtain a global image set and a partial image set; Setting an initial cluster cluster according to the attribute information and damaged point information of each of the global images, and performing a cluster analysis on the vehicle damage data set according to the initial cluster cluster to obtain different vehicle classification sets; Perform image similarity calculation on the vehicle classification images in each vehicle classification set, and eliminate duplicate vehicle classification images based on the image similarity. 3. The vehicle image classification method according to claim 1, wherein the vehicle damage assessment image includes a damaged area; The identification of each vehicle damage measurement image in the vehicle damage measurement data set to obtain the attribute information and damage point information of each vehicle damage measurement image includes: Use the preset damaged point recognition model to identify the damaged area of each vehicle's damage assessment image, and obtain the damaged point information of each vehicle's damage assessment image; Use the semantic segmentation model to process each vehicle's fixed-loss image to obtain the image information of each vehicle's fixed-loss image; A corresponding target recognition model is acquired according to the image information of each vehicle's damage assessment image, and the target recognition model is used to perform recognition processing on the vehicle damage assessment image to obtain attribute information of the vehicle damage assessment image. 5. The vehicle image classification method according to claim 1, wherein the classification of the vehicle image with loss based on the attribute information to obtain a global image set and a partial image set comprises: Acquiring preset classification information, and matching the attribute information of each vehicle's fixed-loss image according to the classification information; If the attribute information and the classification information are successfully matched, the corresponding vehicle loss-specific image is put into the global image collection; If the attribute information and the classification information fail to match, the corresponding vehicle loss-specific image is put into the partial image set. The vehicle image classification method according to claim 1, wherein the initial cluster cluster is set according to the attribute information and damaged point information of each of the global images, and the partial image set is adjusted according to the initial cluster cluster. Perform cluster analysis to get different vehicle classification sets, including: Convert the attribute information and damaged point information of each vehicle's damage-oriented image into feature vector to obtain the feature vector of each vehicle's damage-oriented image; Set the loss-fixing feature vector of each of the global images as the initial clustering point, and use a clustering algorithm to perform cluster analysis on the set of vehicle loss-fixing images according to the initial clustering points to obtain different vehicle clusters cluster; Calculate the vector distance from the loss-fixing feature vector of the loss-fixing image of each vehicle in each vehicle cluster to the center of the cluster; Determining a vehicle loss-based image whose vector distance exceeds a preset vector threshold as an image to be determined; Sending the to-be-determined image to the client to obtain the classification information returned by the client; Allocating each of the to-be-determined images to a corresponding vehicle cluster according to the classification information; The vehicle damage assessment images in each vehicle cluster are composed of vehicle classification sets. A vehicle image classification method, characterized in that it includes: Acquiring a vehicle loss assessment data set, the vehicle loss assessment data set including a global image and a local image; Performing classification processing on the vehicle damage data set to obtain a global image set and a partial image set; Use the preset recognition model to identify each vehicle damage image in the vehicle damage data set, and obtain the attribute information and damage point information of each vehicle damage image; Setting an initial cluster cluster according to the attribute information and damaged point information of each of the global images, and performing cluster analysis on a local image set according to the initial cluster cluster to obtain different vehicle classification sets; Carry out vehicle similarity calculation on the vehicle images in each vehicle classification set, and eliminate duplicate vehicle images according to the vehicle similarity. A vehicle image classification device, characterized in that it comprises: The first data set acquisition module is configured to acquire a vehicle loss assessment data set, where the vehicle loss assessment data set includes at least two vehicle loss assessment images; The first image recognition module is used for recognizing each vehicle's loss-fixing image in the vehicle's loss-fixing data set to obtain attribute information and damaged point information of each vehicle's loss-fixing image; The first image classification module is configured to classify the vehicle loss-specific image according to the attribute information to obtain a global image set and a partial image set; The first classification set acquisition module is used to set initial clusters according to the attribute information and damaged point information of each of the global images, and perform cluster analysis on the vehicle damage data set according to the initial clusters to obtain different Vehicle classification set; The first image elimination module is used to calculate the image similarity of the vehicle classification images in each vehicle classification set, and eliminate the repeated vehicle classification images according to the image similarity. The vehicle image classification device according to claim 6, wherein the classification set acquisition module comprises: The vector conversion unit is used to transform the attribute information and damaged point information of each vehicle's damage-oriented image into feature vector to obtain the feature vector of each vehicle's damage-oriented image; The cluster cluster acquisition unit is configured to set the loss-specific feature vector of each global image as an initial clustering point, and cluster the vehicle loss-specific image set according to the initial clustering point using a clustering algorithm Analyze, get different vehicle clusters; The vector distance calculation unit is used to calculate the vector distance from the loss-fixing feature vector of the loss-fixing image of each vehicle in each vehicle cluster to the center of the cluster cluster; An image dividing unit, configured to determine a vehicle loss-based image whose vector distance exceeds a preset vector threshold as a to-be-determined image; The classification information obtaining unit is configured to send the to-be-determined image to the client, and obtain the classification information returned by the client; An image allocation unit, configured to allocate each of the to-be-determined images to a corresponding vehicle cluster according to the classification information; The classification set composing unit is used to compose the vehicle classification set of the vehicle damage-related images in each vehicle cluster. 7. The vehicle image classification device according to claim 6, wherein the vehicle damage assessment image includes a damaged area, and the first image recognition module includes: The damaged area recognition unit is used to identify the damaged area of each vehicle's damage assessment image by using a preset damaged point recognition model to obtain the damaged point information of each vehicle's damage assessment image; The image information acquisition unit is used to process each vehicle's fixed-loss image by using a semantic segmentation model to obtain image information of each vehicle's fixed-loss image; The attribute information acquisition unit is configured to acquire the corresponding target recognition model according to the image information of each vehicle damage image, and use the target recognition model to identify the vehicle damage image to obtain the vehicle damage image Attribute information. The vehicle image classification device according to claim 6, wherein the first image classification module comprises: The classification information acquisition unit is configured to acquire preset classification information, and match the attribute information of each vehicle's fixed-loss image according to the classification information; The classification unit is configured to put the corresponding vehicle loss-specific image into the global image collection when the attribute information and the classification information match successfully; when the attribute information and the classification information fail to match, then the corresponding The vehicle's fixed-loss image is put into the partial image set. A vehicle image classification device, characterized in that it comprises: The second data set acquisition module is configured to acquire a vehicle loss assessment data set, the vehicle loss assessment data set including global images and partial images; The second image classification module is used to classify the vehicle loss assessment data set to obtain a global image set and a local image set; The second image recognition module is used to recognize each vehicle damage image in the vehicle damage data set by using a preset recognition model to obtain attribute information and damage point information of each vehicle damage image; The second classification set acquisition module is used to set initial clusters according to the attribute information and damaged point information of each of the global images, and perform cluster analysis on the local image sets according to the initial clusters to obtain different vehicles Classification set The second image elimination module is used to calculate the vehicle similarity of the vehicle images in each vehicle classification set, and eliminate the repeated vehicle images according to the vehicle similarity. A computer device comprising a memory, a processor, and computer-readable instructions stored in the memory and capable of running on the processor, wherein the processor executes the computer-readable instructions as follows step: Acquiring a vehicle loss assessment data set, where the vehicle loss assessment data set includes at least two vehicle loss assessment images; Recognize each vehicle damage assessment image in the vehicle damage assessment data set, and obtain the attribute information and damage point information of each vehicle damage assessment image; Classify the vehicle loss-specific image according to the attribute information to obtain a global image set and a partial image set; Setting an initial cluster cluster according to the attribute information and damaged point information of each of the global images, and performing a cluster analysis on the vehicle damage data set according to the initial cluster cluster to obtain different vehicle classification sets; Perform image similarity calculation on the vehicle classification images in each vehicle classification set, and eliminate duplicate vehicle classification images based on the image similarity. The computer device of claim 11, wherein the vehicle damage assessment image includes a damaged area; The identification of each vehicle damage measurement image in the vehicle damage measurement data set to obtain the attribute information and damage point information of each vehicle damage measurement image includes: Use the preset damaged point recognition model to identify the damaged area of each vehicle's damage assessment image, and obtain the damaged point information of each vehicle's damage assessment image; Use the semantic segmentation model to process each vehicle's fixed-loss image to obtain the image information of each vehicle's fixed-loss image; A corresponding target recognition model is acquired according to the image information of each vehicle's damage assessment image, and the target recognition model is used to perform recognition processing on the vehicle damage assessment image to obtain attribute information of the vehicle damage assessment image. The computer device according to claim 11, wherein said classifying said vehicle loss-specific image according to said attribute information to obtain a global image set and a partial image set comprises: Acquiring preset classification information, and matching the attribute information of each vehicle's fixed-loss image according to the classification information; If the attribute information and the classification information are successfully matched, the corresponding vehicle loss-specific image is put into the global image collection; If the attribute information and the classification information fail to match, the corresponding vehicle loss-specific image is put into the partial image set. The computer device according to claim 11, wherein the initial cluster cluster is set according to the attribute information and damaged point information of each of the global images, and the partial image sets are clustered according to the initial cluster clusters. Class analysis to obtain different vehicle classification sets, including: Convert the attribute information and damaged point information of each vehicle's damage-oriented image into feature vector to obtain the feature vector of each vehicle's damage-oriented image; Set the loss-fixing feature vector of each of the global images as the initial clustering point, and use a clustering algorithm to perform cluster analysis on the set of vehicle loss-fixing images according to the initial clustering points to obtain different vehicle clusters cluster; Calculate the vector distance from the loss-fixing feature vector of the loss-fixing image of each vehicle in each vehicle cluster to the center of the cluster; Determining a vehicle loss-based image whose vector distance exceeds a preset vector threshold as an image to be determined; Sending the to-be-determined image to the client to obtain the classification information returned by the client; Allocating each of the to-be-determined images to a corresponding vehicle cluster according to the classification information; The vehicle damage assessment images in each vehicle cluster are composed of vehicle classification sets. A computer device comprising a memory, a processor, and computer-readable instructions stored in the memory and capable of running on the processor, wherein the processor executes the computer-readable instructions as follows step: Acquiring a vehicle loss assessment data set, the vehicle loss assessment data set including a global image and a local image; Performing classification processing on the vehicle damage data set to obtain a global image set and a partial image set; Use the preset recognition model to identify each vehicle damage image in the vehicle damage data set, and obtain the attribute information and damage point information of each vehicle damage image; Setting an initial cluster cluster according to the attribute information and damaged point information of each of the global images, and performing cluster analysis on a local image set according to the initial cluster cluster to obtain different vehicle classification sets; Carry out vehicle similarity calculation on the vehicle images in each vehicle classification set, and eliminate duplicate vehicle images according to the vehicle similarity. One or more readable storage media storing computer readable instructions, when the computer readable instructions are executed by one or more processors, the one or more processors execute the following steps: Acquiring a vehicle loss assessment data set, where the vehicle loss assessment data set includes at least two vehicle loss assessment images; Recognize each vehicle damage assessment image in the vehicle damage assessment data set, and obtain the attribute information and damage point information of each vehicle damage assessment image; Classify the vehicle loss-specific image according to the attribute information to obtain a global image set and a partial image set; Setting an initial cluster cluster according to the attribute information and damaged point information of each of the global images, and performing a cluster analysis on the vehicle damage data set according to the initial cluster cluster to obtain different vehicle classification sets; Perform image similarity calculation on the vehicle classification images in each vehicle classification set, and eliminate duplicate vehicle classification images based on the image similarity. 15. The readable storage medium of claim 16, wherein the vehicle damage-assessment image includes a damaged area; The identification of each vehicle damage measurement image in the vehicle damage measurement data set to obtain the attribute information and damage point information of each vehicle damage measurement image includes: Use the preset damaged point recognition model to identify the damaged area of each vehicle's damage assessment image, and obtain the damaged point information of each vehicle's damage assessment image; Use the semantic segmentation model to process each vehicle's fixed-loss image to obtain the image information of each vehicle's fixed-loss image; A corresponding target recognition model is acquired according to the image information of each vehicle's damage assessment image, and the target recognition model is used to perform recognition processing on the vehicle damage assessment image to obtain attribute information of the vehicle damage assessment image. The readable storage medium according to claim 16, wherein said classifying said vehicle loss-specific image according to said attribute information to obtain a global image set and a partial image set comprises: Acquiring preset classification information, and matching the attribute information of each vehicle's fixed-loss image according to the classification information; If the attribute information and the classification information are successfully matched, the corresponding vehicle loss-specific image is put into the global image collection; If the attribute information and the classification information fail to match, the corresponding vehicle loss-specific image is put into the partial image set. The readable storage medium according to claim 16, wherein the initial cluster cluster is set according to the attribute information and damaged point information of each of the global images, and the partial image set is adjusted according to the initial cluster cluster. Perform cluster analysis to get different vehicle classification sets, including: Convert the attribute information and damaged point information of each vehicle's damage-oriented image into feature vector to obtain the feature vector of each vehicle's damage-oriented image; Set the loss-fixing feature vector of each of the global images as the initial clustering point, and use a clustering algorithm to perform cluster analysis on the set of vehicle loss-fixing images according to the initial clustering points to obtain different vehicle clusters cluster; Calculate the vector distance from the loss-fixing feature vector of the loss-fixing image of each vehicle in each vehicle cluster to the center of the cluster; Determining a vehicle loss-based image whose vector distance exceeds a preset vector threshold as an image to be determined; Sending the to-be-determined image to the client to obtain the classification information returned by the client; Allocating each of the to-be-determined images to a corresponding vehicle cluster according to the classification information; The vehicle damage assessment images in each vehicle cluster are composed of vehicle classification sets. One or more readable storage media storing computer readable instructions, when the computer readable instructions are executed by one or more processors, the one or more processors execute the following steps: Acquiring a vehicle loss assessment data set, the vehicle loss assessment data set including a global image and a local image; Performing classification processing on the vehicle damage data set to obtain a global image set and a partial image set; Use the preset recognition model to identify each vehicle damage image in the vehicle damage data set, and obtain the attribute information and damage point information of each vehicle damage image; Setting an initial cluster cluster according to the attribute information and damaged point information of each of the global images, and performing cluster analysis on a local image set according to the initial cluster cluster to obtain different vehicle classification sets; Carry out vehicle similarity calculation on the vehicle images in each vehicle classification set, and eliminate duplicate vehicle images according to the vehicle similarity.

Images