Risk assessment method and device, electronic equipment and storage medium

Abstract

The present application relates to the technical field of risk management and control, and provides a risk assessment method and device, electronic equipment and storage medium, target features in an image to be evaluated are identified and the image to be evaluated is corrected according to the target features, ground images of each corrected image to be evaluated are obtained based on an image region segmentation model, a plurality of features are obtained by performing feature recognition on the obtained plurality of ground images, the plurality of features are classified and it is judged whether the plurality of ground images are located on the same ground, a risk score and a risk assessment result are determined according to the judgment result, by identifying the features of the fixed ground in the plurality of vehicle images taken from different directions, it is determined whether the plurality of vehicle images are on the same ground, so as to avoid the risk of fraud by the insurance applicant, and reduce the probability of loss of fraud by the insurance company.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of risk control, and particularly relates to a risk assessment method and device, electronic equipment and a storage medium. BACKGROUND

[0002] In recent years, the underwriting process of vehicle insurance has gradually realized online and customer self-service. The insurance company generally requires the customer to upload multiple images of the vehicle taken from different directions to the system. When identifying multiple images, the risk control and identification measures of the traditional system are prone to misjudgment, and cannot accurately determine whether the multiple images are at the same vehicle accident scene, which brings fraud losses to the insurance company. SUMMARY

[0003] In view of the above, it is necessary to provide a risk assessment method and device, electronic equipment and a storage medium, which can determine whether multiple vehicle images are at the same ground by recognizing the features of the fixed ground in the multiple vehicle images taken from different directions, so as to avoid the risk of fraud by the insurance applicant, and reduce the probability of fraud losses of the insurance company.

[0004] The first aspect of the present application provides a risk assessment method, the method comprising: acquiring multiple images to be evaluated, identifying a target feature in each image to be evaluated, and correcting each image to be evaluated based on the target feature;

[0005] Segmenting each corrected image to be evaluated using a pre-trained image region segmentation model to obtain a ground image of each corrected image to be evaluated;

[0006] Identifying features of the obtained multiple ground images to obtain multiple features;

[0007] Classifying the multiple features, and determining whether the multiple ground images are located at the same ground based on the classification result;

[0008] Determining a risk score according to the determination result and obtaining a risk assessment result of the multiple images to be evaluated.

[0009] According to an optional embodiment of the present application, the image to be evaluated includes a vehicle image, and the identifying a target feature in each image to be evaluated and correcting each image to be evaluated based on the target feature comprises:

[0010] Identifying and positioning the target feature in each image to be evaluated based on optical character recognition (OCR) technology;

[0011] Determining the center point of the target feature according to the positioning result;

[0012] The origin correction is performed based on the center point with the target feature as a reference, wherein the target feature includes a license plate of the vehicle, and the origin correction includes rotation correction.

[0013] According to an optional embodiment of the present application, the training method of the image region segmentation model comprises:

[0014] A sample set is acquired, and the sample set is divided into a training set and a test set, wherein the sample set includes a preset number of vehicle images;

[0015] The DeepLab V3 architecture is trained by using the training set, and an initial image region segmentation model is obtained;

[0016] The initial image region segmentation model is tested by using the test set, and the parameters of the initial image segmentation model are adjusted according to the test result until the image region segmentation model meeting the preset requirement is obtained.

[0017] According to an optional embodiment of the present application, the region segmentation of each corrected to-be-evaluated image by using the pre-trained image region segmentation model to obtain the ground image of each corrected to-be-evaluated image comprises:

[0018] Each corrected to-be-evaluated image is input into the image region segmentation model to obtain the foreground region and the background region of each corrected to-be-evaluated image, wherein the to-be-evaluated image includes a vehicle image, the foreground region includes a vehicle body region, and the background region includes a ground region and a surrounding environment region outside the vehicle body;

[0019] The vehicle body region is extended outward according to a preset proportion to obtain an extended region, the extended region does not include the vehicle body region, and the extended region belongs to the ground region;

[0020] The region outside the extended region is masked to obtain an image of the extended region, and the image of the extended region is taken as the ground image, wherein each ground image corresponds to one corrected to-be-evaluated image.

[0021] According to an optional embodiment of the present application, the method for feature recognition of the obtained multiple ground images comprises a scale-invariant feature transform (SIFT) algorithm.

[0022] The multiple features include n kinds of features, and each ground image contains k features, each feature corresponding to one ground image, wherein n and k are both integers greater than or equal to 1.

[0023] According to an optional embodiment of the present application, the method for classifying the multiple features comprises a cosine similarity algorithm.

[0024] the judging whether the multiple ground images are located at the same ground based on the classification result comprises:

[0025] when the classification result indicates that the k features belong to the same category, determining that the k features are the same kind of feature;

[0026] determining the ground images containing the same kind of feature as being located at the same ground.

[0027] According to an optional embodiment of the present application, the determining the risk score and obtaining the risk assessment result of the multiple to-be-evaluated images according to the judgment result comprises:

[0028] calculating the ratio of the number of the ground images located at the same ground to the total number of the ground images, and determining the risk score according to the ratio and a preset score formula;

[0029] when the risk score is higher than a preset risk threshold, determining that the multiple to-be-evaluated images have a fraud risk.

[0030] The second aspect of the present application provides a risk assessment device, which comprises a correction module, a segmentation module, an identification module, a classification module and an evaluation module.

[0031] The correction module is used to acquire multiple to-be-evaluated images, identify target features in each to-be-evaluated image, and correct each to-be-evaluated image based on the target features.

[0032] The segmentation module is used to perform region segmentation on each corrected to-be-evaluated image by using a pre-trained image region segmentation model, and obtain ground images of each corrected to-be-evaluated image.

[0033] The identification module is used to perform feature identification on the obtained multiple ground images, and obtain multiple features.

[0034] The classification module is used to classify the multiple features, and judge whether the multiple ground images are located at the same ground based on the classification result.

[0035] The evaluation module is used to determine a risk score and obtain a risk assessment result of the multiple to-be-evaluated images according to the judgment result.

[0036] The third aspect of the present application provides an electronic device, which comprises a processor and a memory, and the processor is used to implement the risk assessment method when executing a computer program stored in the memory.

[0037] A fourth aspect of the present application provides a computer readable storage medium, and the computer readable storage medium stores a computer program, and the computer program is executed by a processor to implement the risk assessment method.

[0038] To sum up, the risk assessment method, device, electronic equipment and storage medium provided by the present application first identify the target features in the to-be-evaluated image and correct the to-be-evaluated image according to the target features, then obtain the ground image of each corrected to-be-evaluated image based on the pre-trained image region segmentation model, and perform feature recognition on the obtained multiple ground images to obtain multiple features, then classify the multiple features and judge whether the multiple ground images are located on the same ground, and finally determine the risk score and risk assessment result according to the judgment result. It can be applied to risk assessment of vehicle insurance industry, etc. By identifying the features of the fixed ground in the multiple vehicle images taken from different directions, it is determined whether the multiple vehicle images are on the same ground, so as to determine whether the multiple vehicle images belong to the same vehicle accident case at the same location, so as to avoid the risk of fraudulent insurance clients, identify the fraud in the vehicle insurance process and reduce the probability of fraud loss of the insurance company. BRIEF DESCRIPTION OF DRAWINGS

[0039] Figure 1 is a flowchart of the risk assessment method provided by the first embodiment of the present application.

[0040] Figure 2 is a structural diagram of the risk assessment device provided by the second embodiment of the present application.

[0041] Figure 3 is a structural schematic diagram of the electronic equipment provided by the third embodiment of the present application. DETAILED DESCRIPTION

[0042] In order to more clearly understand the above-mentioned purposes, features and advantages of the present application, the present application will be described in detail below in combination with the drawings and specific embodiments. It should be noted that the embodiments of the present application and the features in the embodiments can be combined with each other without conflict.

[0043] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which the present application belongs. The terms used in the specification of the present application are only for the purpose of describing the embodiments in one optional embodiment, and are not intended to limit the present application.

[0044] The risk assessment method provided by the embodiments of the present application is executed by an electronic equipment, and correspondingly, the risk assessment device runs in the electronic equipment.

[0045] Embodiment one

[0046] Figure 1 is a flowchart of the risk assessment method provided by Embodiment One of the present application. The risk assessment method specifically comprises the following steps, and the order of the steps in the flowchart can be changed according to different requirements, and some steps can be omitted.

[0047] S11, a plurality of images to be evaluated are obtained, target features in each image to be evaluated are identified, and each image to be evaluated is corrected based on the target features.

[0048] In an optional embodiment, the scheme provided by the embodiments of the present application can be applied to risk assessment of the vehicle insurance industry. By identifying the features of the ground in a plurality of vehicle images taken from different directions, it is determined whether the plurality of vehicle images are on the same ground, thereby determining whether the plurality of vehicle images belong to the same vehicle accident case at the same location, so as to avoid the risk of fraudulent insurance clients, identify fraudulent behaviors in the vehicle insurance process, and reduce the probability of fraudulent losses suffered by insurance companies.

[0049] In an optional embodiment, the image to be evaluated includes a vehicle image. The electronic device can receive a set of vehicle images taken from different directions uploaded by a user through a preset insurance platform, thereby obtaining the plurality of images to be evaluated, wherein each image to be evaluated generally contains only one vehicle, and the image to be evaluated is a rectangular image.

[0050] In an optional embodiment, the target feature includes a license plate of a vehicle, or a vehicle license plate or a vehicle number plate.

[0051] Before identifying the target feature in each image to be evaluated, the method further comprises pre-processing the plurality of images to be evaluated, and the pre-processing includes image color space conversion, gray correction, image smoothing processing, etc.

[0052] In an optional embodiment, the color of the license plate is mostly blue background with white characters, yellow background with black characters, white background with red characters, green background with white characters, etc. Based on the image color space conversion technology, the area of the license plate can be distinguished from the image to be evaluated by using different color channels, so as to facilitate subsequent processing and positioning of the license plate.

[0053] For example, when the license plate A is blue background with white characters, the blue B (Blue) channel can be used to convert the color space of the image to be evaluated, so that the area of the license plate A can be obviously presented in the image to be evaluated. Moreover, this method will not affect the text area (such as Chinese characters, numbers, English letters, etc.) of the license plate A, because blue (255, 0, 0) and white (255, 255, 255) are not distinguished in the B channel.

[0054] In an optional embodiment, the gray scale correction can enhance the change range of the gray scale of the image to be evaluated, enrich the gray scale level of the image to be evaluated, enhance the contrast and resolution of the image to be evaluated, and facilitate subsequent processing and positioning of the license plate.

[0055] In an optional embodiment, the image smoothing processing uses a method of averaging neighboring pixels to filter out noise in the image to be evaluated by a filter, thereby improving the definition of the image to be evaluated and facilitating subsequent processing and positioning of the license plate. The average neighboring pixel value represents the average value of the pixels of all neighboring points of the center point, and the neighboring point represents a pixel point within a predetermined adjacent range of the center point.

[0056] Specifically, when performing the image smoothing processing, a difference threshold value can also be set for the difference between the pixel value of the center point of the image to be evaluated and the average neighboring pixel value. When the difference between the center point and any neighboring point is greater than the difference threshold value, the pixel value of the any neighboring point is updated to the average neighboring pixel value, and the pixel values of the remaining neighboring points are not updated, thereby reducing the blurring caused by the image smoothing processing.

[0057] In an optional embodiment, the identifying the target feature in each image to be evaluated and correcting each image to be evaluated based on the target feature comprises:

[0058] identifying and positioning the target feature in each image to be evaluated based on an optical character recognition (OCR) technology;

[0059] determining the center point of the target feature according to the positioning result;

[0060] performing origin correction based on the center point with the target feature as a reference, wherein the origin correction includes rotation correction.

[0061] Specifically, the target feature includes a license plate of a vehicle. First, a text region containing text in each of the to-be-evaluated images is identified based on an OCR technology, and each text in the text region is identified and detected, so as to determine a text region with a specific text format, for example, the specific text format can be "provincial name abbreviation + representative letter of city + 5 random letters or numbers". Then, the license plate containing the text region can be located according to the text region, and the license plate can also be located based on the method in the preprocessing. After the license plate is located, a rectangular coordinate system can be established in the to-be-evaluated image and the coordinates of four vertices of the license plate are determined, and the center point of the coordinates of the four vertices is taken as the center point of the license plate. Finally, based on the center point, rotation correction is performed with the four straight lines of the frame of the license plate as a reference, so that the four straight lines of the frame of the license plate are respectively parallel to the four edges of the rectangular image of the to-be-evaluated image, for example, two long edges of the four straight lines of the frame of the license plate are rotationally corrected to be parallel to two long edges of the rectangular image of the to-be-evaluated image, and two wide edges of the four straight lines of the frame of the license plate are rotationally corrected to be parallel to two wide edges of the rectangular image of the to-be-evaluated image.

[0062] In an optional implementation, a plurality of to-be-evaluated images of the same vehicle have the same stable feature, for example, a license plate of the vehicle, regardless of the angle or direction in which the to-be-evaluated images are taken. Based on the license plate as the origin, rotation correction is performed on the to-be-evaluated images, so that the influence of the diversity of the shooting angles of the to-be-evaluated images can be eliminated, thereby improving the robustness of subsequent ground feature identification.

[0063] S12, performing region segmentation on each of the corrected to-be-evaluated images by using a pre-trained image region segmentation model, to obtain a ground image of each of the corrected to-be-evaluated images.

[0064] In an optional implementation, the training method of the image region segmentation model includes:

[0065] obtaining a sample set, and dividing the sample set into a training set and a test set, wherein the sample set includes a preset number of vehicle images;

[0066] training a DeepLab V3 architecture by using the training set, to obtain an initial image region segmentation model;

[0067] testing the initial image region segmentation model by using the test set, adjusting parameters of the initial image segmentation model according to a test result, and stopping until the image region segmentation model meeting a preset requirement is obtained.

[0068] In an optional embodiment, a public dataset PASCAL VOC 2012 can be acquired, and the preset number (e.g., 50,000) of images (i.e., the preset number of vehicle images) identified as cars or vehicles are extracted therefrom as the sample set, and the sample set is divided into a training set and a test set in a ratio of, for example, 7:3.

[0069] In an optional embodiment, the DeepLab architecture is a network architecture based on a deep convolutional neural network (DCNN) and a probabilistic graphical model (PGM), and an atrous (hole) algorithm is used to expand the receptive field of the foreground region and the background region in the vehicle image, and a fully connected conditional random field (CRF) is used to improve the ability of the model to capture the details of the foreground region and the background region, wherein the foreground region includes the body region of the vehicle, and the background region includes the ground region outside the body and the surrounding environment region.

[0070] The DeepLabv3 architecture is an improved architecture based on the DeepLab architecture, and the DeepLabv3 architecture uses a hollow convolution to prevent the resolution of the image region segmentation from being too low, and uses a multi-proportion hole convolution cascade or a parallel hollow convolution with different expansion rates to capture the multi-scale background region of the vehicle image, thereby improving the accuracy of image region segmentation and obtaining more accurate foreground region and background region.

[0071] In an optional embodiment, the parameters of the initial image segmentation model include the expansion rate of the hollow convolution of the initial image segmentation model, the size of the convolution kernel, the number of channels, and the like.

[0072] The preset requirement includes that the loss function of the initial image segmentation model converges to a preset convergence threshold. The loss function represents the difference between the accuracy of the initial image segmentation model and 1, and the smaller the value to which the loss function converges, the higher the accuracy of the initial image segmentation model.

[0073] In an optional embodiment, the region segmentation of each corrected to-be-evaluated image using the pre-trained image region segmentation model to obtain the ground image of each corrected to-be-evaluated image includes:

[0074] input each corrected to-be-evaluated image into the image region segmentation model to obtain a foreground region and a background region of each corrected to-be-evaluated image, wherein the to-be-evaluated image includes a vehicle image, the foreground region includes a vehicle body region of the vehicle, and the background region includes a ground region outside the vehicle body and a surrounding environment region;

[0075] extend the vehicle body region outward according to a preset proportion to obtain an extended region, the extended region does not include the vehicle body region, and the extended region belongs to the ground region;

[0076] mask regions outside the extended region, obtain an image of the extended region, and use the image of the extended region as the ground image, wherein each ground image corresponds to one corrected to-be-evaluated image.

[0077] Specifically, after the vehicle body region is determined, the edge of the vehicle body region, for example, the frame edge of the vehicle body, can be determined based on an image edge detection algorithm; the frame edge of the vehicle body is extended outward by a preset proportion (for example, 5%) to obtain the extended region, wherein the preset proportion can be adjusted according to the diversity of the parking environment. The extended region close to the frame edge of the vehicle body is extracted, the image of the extended region is used as the ground region, and all image information except the extended region is masked by using an image masking (mask) technology to obtain the image of the extended region as the ground image.

[0078] In an optional embodiment, the method provided by the embodiment of the present application can improve the accuracy of extracting the ground image, reduce the influence of the surrounding environment image outside the ground in each corrected to-be-evaluated image, and improve the accuracy of subsequent feature recognition of the ground image.

[0079] S13, performing feature recognition on the obtained multiple ground images to obtain multiple features.

[0080] In an optional embodiment, the method for performing feature recognition on the obtained multiple ground images includes a scale-invariant feature transform (SIFT) algorithm. The SIFT algorithm is a local feature descriptor used in the field of image processing, has scale invariance and direction invariance, and can be used to detect key points of an image.

[0081] Specifically, the feature recognition of the ground image based on the SIFT algorithm includes:

[0082] Scale-space extrema detection is performed on the ground image to obtain candidate key points, including: using a Difference of Gaussian function to identify the candidate key points of interest in the ground image;

[0083] Target key points are located from the candidate key points, including: determining the stability of the positions and scales of the candidate key points, and taking points with stability greater than a preset stability threshold as the target key points, wherein the stability includes scale invariance and direction invariance;

[0084] The direction of the target key point is determined, including: counting all gradient directions of the target key point in a local region of the ground image, and performing 1.5σ Gaussian filtering to obtain the direction of the target key point, wherein all gradient directions of the local region include: taking the target key point as the center and 1.5 times the scale of the Gaussian image where the target key point is located as the radius as the local region, and taking all gradient directions and gradient amplitudes of pixels within the circle of the local region as all gradient directions of the local region;

[0085] A descriptor of the target key point is constructed, including: determining an image region required for calculating the descriptor of the target key point, rotating the local region of the target key point to the direction of the target key point in the image region, and generating the descriptor of the target key point based on the rotation, wherein the descriptor defines the direction of the target key point with direction invariance;

[0086] The target key points and the descriptor are used to determine feature points (hereinafter referred to as SIFT feature points) of the ground image, and features of the ground image are obtained based on the feature points.

[0087] In an optional implementation, the SIFT algorithm can achieve reliable matching of object pictures or scene pictures under different perspectives, and the extracted features have invariance to the scale and rotation of the image, for example, the scale invariance and the direction invariance, and have good robustness.

[0088] The plurality of features include n types of features (for example, ground craters, vehicle body fragments, etc.), and each ground image can include k features, and each feature corresponds to a ground image, wherein n and k are integers greater than or equal to 1.

[0089] S14, classifying the plurality of features, and determining whether the plurality of ground images are located on the same ground based on a classification result.

[0090] In an optional embodiment, the method for classifying the plurality of features includes a cosine similarity algorithm, and a plurality of cosine similarity ranges can be set, and features with a cosine similarity falling into the same cosine similarity range are classified into the same class.

[0091] In an optional embodiment, the classifying the plurality of features further includes:

[0092] Matching the SIFT feature points of the plurality of ground images to generate SIFT feature vectors between the plurality of ground images, the SIFT feature vectors including feature vectors composed of mutually matched SIFT feature points;

[0093] Calculating the Euclidean distances between the SIFT feature vectors, and classifying features corresponding to the SIFT feature vectors falling into the same Euclidean distance range into the same class according to a plurality of pre-set Euclidean distance ranges.

[0094] In an optional embodiment, the determining whether the plurality of ground images are located on the same ground based on the classification result includes: when the classification result indicates that k features belong to the same class, determining that the k features are the same kind of feature; and determining that a ground image containing the same kind of feature is located on the same ground.

[0095] S15, determining a risk score and obtaining a risk assessment result of the plurality of images to be evaluated according to the determination result.

[0096] In an optional embodiment, the determining a risk score and obtaining a risk assessment result of the plurality of images to be evaluated according to the determination result includes:

[0097] Calculating a ratio of the number of ground images located on the same ground to the total number of ground images, and determining the risk score according to the ratio and a pre-set scoring formula, the pre-set scoring formula can be: risk score = 10 x (1 - the ratio), wherein 10 represents the highest score;

[0098] When the risk score is higher than a pre-set risk threshold, it is determined that the plurality of images to be evaluated have a fraud risk.

[0099] It should be noted that since the features include n kinds of features, there can be multiple groups of ground images located on the same ground, and different groups of images can include different kinds of features. At this time, the ratio of the number of ground images located on the same ground to the total number of ground images can be calculated to determine the group with the largest number of images, and the ratio of the number of images in the group with the largest number of images to the total number of images can be calculated.

[0100] In addition, a plurality of risk score ranges can be set, and a plurality of risk levels can be determined according to the plurality of risk score ranges. For example, the risk score ranges can be set as [0, 3], [3, 7], and [7, 10], the risk score in the range of [0, 3] is determined as low risk, the risk score in the range of [3, 7] is determined as low risk, and the risk score in the range of [7, 10] is determined as high risk.

[0101] In an optional embodiment, the method further comprises: displaying the classification result and the judgment result by using a preset visual interface. For example, each type of feature labeled by using the feature points is displayed in the classification result, and the risk score and / or the risk level is displayed in the judgment result.

[0102] In addition, the visual interface can also respond to the modification operation of the relevant risk control personnel on the classification result and the judgment result, update the classification result and the judgment result, thereby improving the accuracy of risk assessment; and the risk control personnel can be reminded by using a highlight color such as red to prioritize the verification of high-risk images, thereby avoiding fraud loss.

[0103] In an optional embodiment, the risk assessment method provided by the embodiment of the present application first identifies the target feature in the image to be evaluated and corrects the image to be evaluated according to the target feature, then obtains the ground image of each corrected image to be evaluated based on a pre-trained image region segmentation model, and performs feature recognition on the obtained multiple ground images to obtain multiple features, then classifies the multiple features and judges whether the multiple ground images are located on the same ground, and finally determines the risk score and the risk assessment result according to the judgment result, which can be applied to risk assessment of the vehicle insurance industry. By identifying the features of the fixed ground in the multiple vehicle images taken from different directions, it is determined whether the multiple vehicle images are on the same ground, thereby determining whether the multiple vehicle images belong to the same vehicle accident case at the same location, so as to avoid the risk of fraud by the insurance applicant, identify the fraud behavior in the vehicle insurance link, and reduce the probability of fraud loss of the insurance company.

[0104] Embodiment Two

[0105] Figure 2 FIG. 1 is a structural diagram of a risk assessment device provided by the embodiment two of the present application.

[0106] In some embodiments, the risk assessment device 20 can include a plurality of functional modules composed of computer program segments. The computer programs of each program segment in the risk assessment device 20 can be stored in the memory of an electronic device and executed by at least one processor to perform the functions of risk assessment (see Figure 1 Description) risk assessment.

[0107] In this embodiment, the risk assessment device 20 can be divided into multiple functional modules according to the functions performed by the device. The functional modules can include a correction module 201, a segmentation module 202, an identification module 203, a classification module 204, and an assessment module 205. The module referred to in the present application refers to a series of computer program segments that can be executed by at least one processor and can complete a fixed function, which are stored in a memory. In this embodiment, the functions of each module will be described in detail in subsequent embodiments.

[0108] The correction module 201 is configured to obtain multiple images to be evaluated, identify target features in each image to be evaluated, and correct each image to be evaluated based on the target features.

[0109] In an optional implementation, the scheme provided by the embodiment of the present application can be applied to risk assessment of the vehicle insurance industry. By identifying the features of the ground in multiple vehicle images taken from different directions, it is determined whether the multiple vehicle images are on the same ground, thereby determining whether the multiple vehicle images belong to the same vehicle accident case at the same location, so as to avoid the risk of fraud by the insurance applicant, identify fraud in the vehicle insurance process, and reduce the probability of fraud loss of the insurance company.

[0110] In an optional implementation, the image to be evaluated includes a vehicle image. The electronic device can receive a set of vehicle images taken from different directions uploaded by a user through a preset insurance platform, thereby obtaining the multiple images to be evaluated. Each image to be evaluated generally includes only one vehicle, and the image to be evaluated is a rectangular image.

[0111] In an optional implementation, the target feature includes a license plate of a vehicle, also known as a vehicle license plate or a vehicle number plate.

[0112] Before identifying the target feature in each image to be evaluated, the method further includes pre-processing the multiple images to be evaluated, and the pre-processing includes image color space conversion, gray correction, image smoothing processing, and the like.

[0113] In an optional implementation, the color of the license plate is mostly blue background and white characters, yellow background and black characters, white background and red characters, green background and white characters, and the like. Based on the image color space conversion technology, the area of the license plate can be distinguished from the image to be evaluated using different color channels, so as to facilitate subsequent processing and positioning of the license plate.

[0114] For example, when the license plate A is blue with white characters, the color space conversion of the image to be evaluated can be performed by using the blue B channel, so that the license plate A can be clearly shown. This method does not affect the characters (e.g., Chinese characters, numbers, English letters, etc.) of the license plate A, because the blue color (255, 0, 0) and the white color (255, 255, 255) are not distinguished in the B channel.

[0115] In an optional embodiment, the gray scale correction can enhance the variation range of the gray scale of the image to be evaluated, enrich the gray scale levels of the image to be evaluated, enhance the contrast and resolution of the image to be evaluated, and facilitate subsequent processing and positioning of the license plate.

[0116] In an optional embodiment, the image smoothing processing uses the method of neighborhood average value to filter the noise of the image to be evaluated by using a filter, so as to improve the definition of the image to be evaluated and facilitate subsequent processing and positioning of the license plate. The neighborhood average pixel value represents the average value of the pixels of all neighborhood points of the center point, and the neighborhood point represents a pixel point within a preset adjacent range of the center point.

[0117] Specifically, when the image smoothing processing is performed, a difference threshold value can be set for the difference between the pixel value of the center point of the image to be evaluated and the neighborhood average pixel value. When the difference between the pixel of the center point and any neighborhood point is greater than the difference threshold value, the pixel value of the any neighborhood point is updated to the neighborhood average pixel value, and the pixel values of the remaining neighborhood points are not updated, so as to reduce the blur caused by the image smoothing processing.

[0118] In an optional embodiment, the identification of the target feature in each image to be evaluated and the correction of each image to be evaluated based on the target feature include:

[0119] identifying and positioning the target feature in each image to be evaluated based on the optical character recognition (OCR) technology;

[0120] determining the center point of the target feature according to the positioning result;

[0121] performing origin correction based on the center point by taking the target feature as a reference, wherein the origin correction includes rotation correction.

[0122] Specifically, the target feature includes a license plate of a vehicle. First, a text region containing text in each of the to-be-evaluated images is identified based on an OCR technology, each text in the text region is recognized and detected, so as to determine a text region with a specific text format, for example, the specific text format can be "provincial name abbreviation + representative letter of city + 5 random letters or numbers". Then, the license plate containing the text region can be located according to the text region, and the license plate can also be located based on the method in the preprocessing. After the license plate is located, a rectangular coordinate system can be established in the to-be-evaluated image and the coordinates of four vertices of the license plate are determined, and the center point of the coordinates of the four vertices is taken as the center point of the license plate. Finally, based on the center point, rotation correction is performed with the four straight lines of the frame of the license plate as a reference, so that the four straight lines of the frame of the license plate are respectively parallel to the four edges of the rectangular image of the to-be-evaluated image, for example, two long edges of the four straight lines of the frame of the license plate are rotationally corrected to be parallel to two long edges of the rectangular image of the to-be-evaluated image, and two wide edges of the four straight lines of the frame of the license plate are rotationally corrected to be parallel to two wide edges of the rectangular image of the to-be-evaluated image.

[0123] In an optional implementation, a plurality of to-be-evaluated images of the same vehicle have the same stable feature, for example, a license plate of the vehicle, regardless of the angle or position at which the to-be-evaluated images are captured. Rotation correction is performed on the to-be-evaluated images based on the license plate as the origin, which can eliminate the influence of the diversity of the capturing angle of the to-be-evaluated images, thereby improving the robustness of subsequent ground feature recognition.

[0124] The segmentation module 202 is configured to perform region segmentation on each of the corrected to-be-evaluated images by using a pre-trained image region segmentation model, to obtain a ground image of each of the corrected to-be-evaluated images.

[0125] In an optional implementation, the training method of the image region segmentation model includes:

[0126] A sample set is obtained, and the sample set is divided into a training set and a test set, wherein the sample set includes a preset number of vehicle images;

[0127] The training set is used to train a DeepLab V3 architecture, to obtain an initial image region segmentation model;

[0128] The test set is used to test the initial image region segmentation model, and parameters of the initial image segmentation model are adjusted according to a test result, until the image region segmentation model meeting a preset requirement is obtained.

[0129] In an optional embodiment, a public dataset PASCAL VOC 2012 can be acquired, and the preset number (e.g., 50,000) of images (i.e., the preset number of vehicle images) identified as cars or vehicles are extracted therefrom as the sample set, and the sample set is divided into a training set and a test set in a ratio of, for example, 7:3.

[0130] In an optional embodiment, the DeepLab architecture is a network architecture based on a deep convolutional neural network (DCNN) and a probabilistic graphical model (PGM), and an atrous (hole) algorithm is used to expand the receptive field of the foreground region and the background region in the vehicle image, and a fully connected conditional random field (CRF) is used to improve the ability of the model to capture the details of the foreground region and the background region, wherein the foreground region includes the body region of the vehicle, and the background region includes the ground region outside the body and the surrounding environment region.

[0131] The DeepLabv3 architecture is an improved architecture based on the DeepLab architecture, and the DeepLabv3 architecture uses a hollow convolution to prevent the resolution of the image region segmentation from being too low, and uses a multi-proportion hole convolution cascade or a parallel hollow convolution with different expansion rates to capture the multi-scale background region of the vehicle image, thereby improving the accuracy of image region segmentation and obtaining more accurate foreground region and background region.

[0132] In an optional embodiment, the parameters of the initial image segmentation model include the expansion rate of the hollow convolution of the initial image segmentation model, the size and the number of channels of the convolution kernel, and the like.

[0133] The preset requirement includes that the loss function of the initial image segmentation model converges to a preset convergence threshold. The loss function represents the difference between the accuracy of the initial image segmentation model and 1, and the smaller the value to which the loss function converges, the higher the accuracy of the initial image segmentation model.

[0134] In an optional embodiment, the region segmentation of each corrected to-be-evaluated image by using the pre-trained image region segmentation model to obtain the ground image of each corrected to-be-evaluated image includes:

[0135] input each corrected to-be-evaluated image into the image region segmentation model to obtain a foreground region and a background region of each corrected to-be-evaluated image, wherein the to-be-evaluated image includes a vehicle image, the foreground region includes a vehicle body region of the vehicle, and the background region includes a ground region outside the vehicle body and a surrounding environment region;

[0136] extend the vehicle body region outward according to a preset proportion to obtain an extended region, the extended region does not include the vehicle body region, and the extended region belongs to the ground region;

[0137] mask regions outside the extended region, obtain an image of the extended region, and use the image of the extended region as the ground image, wherein each ground image corresponds to one corrected to-be-evaluated image.

[0138] Specifically, after the vehicle body region is determined, the edge of the vehicle body region, for example, the frame edge of the vehicle body, can be determined based on an image edge detection algorithm; the frame edge of the vehicle body is extended outward by a preset proportion (for example, 5%) to obtain the extended region, wherein the preset proportion can be adjusted according to the diversity of the parking environment. The extended region close to the frame edge of the vehicle body is extracted, the image of the extended region is used as the ground region, and all image information except the extended region is masked by using an image masking (mask) technology to obtain the image of the extended region as the ground image.

[0139] In an optional implementation, the method provided by the embodiment of the present application can improve the accuracy of extracting the ground image, reduce the influence of the surrounding environment image outside the ground in each corrected to-be-evaluated image, and improve the accuracy of subsequent feature recognition of the ground image.

[0140] The recognition module 203 is configured to perform feature recognition on the obtained multiple ground images to obtain multiple features.

[0141] In an optional implementation, the method for performing feature recognition on the obtained multiple ground images includes a scale-invariant feature transform (SIFT) algorithm. The SIFT algorithm is a local feature descriptor used in the field of image processing, has scale invariance and direction invariance, and can be used to detect key points of an image.

[0142] Specifically, the feature recognition on the ground image based on the SIFT algorithm includes:

[0143] Scale-space extrema detection is performed on the ground image to obtain candidate key points, including: using a Difference of Gaussian function to identify the candidate key points of interest in the ground image;

[0144] Target key points are located from the candidate key points, including: determining the stability of the positions and scales of the candidate key points, and taking points with stability greater than a preset stability threshold as the target key points, wherein the stability includes scale invariance and direction invariance;

[0145] The direction of the target key point is determined, including: counting all gradient directions of the target key point in a local region of the ground image, and performing 1.5σ Gaussian filtering to obtain the direction of the target key point, wherein all gradient directions of the local region include: taking the target key point as the center and 1.5 times the scale of the Gaussian image where the target key point is located as the radius as the local region, and taking all gradient directions and gradient amplitudes of pixels within the circle of the local region as all gradient directions of the local region;

[0146] A descriptor of the target key point is constructed, including: determining an image region required for calculating the descriptor of the target key point, rotating the local region of the target key point to the direction of the target key point in the image region, and generating the descriptor of the target key point based on the rotation, wherein the descriptor defines the direction of the target key point with direction invariance;

[0147] The target key points and the descriptor are used to determine feature points (hereinafter referred to as SIFT feature points) of the ground image, and features of the ground image are obtained based on the feature points.

[0148] In an optional implementation, the SIFT algorithm can realize reliable matching of object pictures or scene pictures under different perspectives, and the extracted features have invariance to the scale and rotation of the image, for example, the scale invariance and the direction invariance, and have good robustness.

[0149] The plurality of features include n types of features (for example, ground craters, vehicle body fragments, etc.), and each ground image can include k features, and each feature corresponds to a ground image, wherein n and k are integers greater than or equal to 1.

[0150] The classification module 204 is configured to classify the plurality of features, and determine whether the plurality of ground images are located on the same ground based on a classification result.

[0151] In an optional embodiment, the method for classifying the plurality of features includes a cosine similarity algorithm, and a plurality of cosine similarity ranges can be set, and features with a cosine similarity falling within the same cosine similarity range are classified into the same class.

[0152] In an optional embodiment, the classifying the plurality of features further includes:

[0153] Matching the SIFT feature points of the plurality of ground images to generate SIFT feature vectors between the plurality of ground images, the SIFT feature vectors including feature vectors composed of mutually matched SIFT feature points;

[0154] Calculating the Euclidean distances between the SIFT feature vectors, and according to a plurality of pre-set Euclidean distance ranges, features corresponding to the SIFT feature vectors falling within the same Euclidean distance range are classified into the same class.

[0155] In an optional embodiment, the determining whether the plurality of ground images are located on the same ground based on the classification result includes: when the classification result indicates that k features belong to the same class, determining that the k features are the same kind of feature; and determining that the ground images containing the same kind of feature are located on the same ground.

[0156] The evaluation module 205 is configured to determine a risk score according to the determination result and obtain a risk evaluation result of the plurality of images to be evaluated.

[0157] In an optional embodiment, the determining a risk score according to the determination result and obtaining a risk evaluation result of the plurality of images to be evaluated includes:

[0158] Calculating a ratio of the number of ground images located on the same ground to the total number of ground images, and determining the risk score according to the ratio and a preset scoring formula, the preset scoring formula can be: risk score = 10 x (1-the ratio), wherein 10 represents the highest score;

[0159] When the risk score is higher than a preset risk threshold, it is determined that the plurality of images to be evaluated have a fraud risk.

[0160] It should be noted that since the features include n kinds of features, there can be multiple groups of ground images located on the same ground, and different groups of images can include different kinds of features. At this time, the ratio of the number of ground images located on the same ground to the total number of ground images can be calculated to determine the group with the largest number of images, and the ratio of the number of images in the group with the largest number of images to the total number of images can be calculated.

[0161] In addition, a plurality of risk score ranges can be set, and a plurality of risk levels can be determined according to the plurality of risk score ranges. For example, the risk score ranges can be set as [0, 3], [3, 7], and [7, 10], the risk score in the range [0, 3] is determined as low risk, the risk score in the range [3, 7] is determined as low risk, and the risk score in the range [7, 10] is determined as high risk.

[0162] In an optional embodiment, the method further comprises: displaying the classification result and the judgment result by using a preset visual interface. For example, each type of feature labeled by using the feature points is displayed in the classification result, and the risk score and / or the risk level is displayed in the judgment result.

[0163] In addition, the visual interface can also update the classification result and the judgment result in response to the modification operation of the relevant risk control personnel on the classification result and the judgment result, so as to improve the accuracy of risk assessment; and the risk control personnel can be reminded to prioritize the verification of high-risk images by using a highlight color such as red, so as to avoid fraud loss.

[0164] In an optional embodiment, the risk assessment method provided by the embodiment of the present application first identifies the target feature in the image to be evaluated and corrects the image to be evaluated according to the target feature, then obtains the ground image of each corrected image to be evaluated based on the pre-trained image region segmentation model, and performs feature recognition on the obtained multiple ground images to obtain multiple features, then classifies the multiple features and judges whether the multiple ground images are located on the same ground, and finally determines the risk score and the risk assessment result according to the judgment result, which can be applied to risk assessment of the vehicle insurance industry. By identifying the features of the fixed ground in the multiple vehicle images taken from different directions, it is determined whether the multiple vehicle images are on the same ground, so as to determine whether the multiple vehicle images belong to the same vehicle accident case at the same location, so as to avoid the risk of fraud by the insured, identify the fraud in the vehicle insurance process, and reduce the probability of fraud loss of the insurance company.

[0165] Embodiment three

[0166] The embodiment provides a computer readable storage medium, and the computer readable storage medium stores a computer program. The computer program is executed by a processor to implement the steps in the risk assessment method embodiments described above, for example Figure 1 S11-S15 shown in the figure:

[0167] S11, obtaining a plurality of images to be evaluated, identifying a target feature in each image to be evaluated, and correcting each image to be evaluated based on the target feature;

[0168] S12, performing region segmentation on each of the corrected to-be-evaluated images by using a pre-trained image region segmentation model to obtain ground images of each of the corrected to-be-evaluated images;

[0169] S13, performing feature recognition on the obtained multiple ground images to obtain multiple features;

[0170] S14, classifying the multiple features, and determining whether the multiple ground images are located on the same ground based on a classification result;

[0171] S15, determining a risk score and obtaining a risk evaluation result of the multiple to-be-evaluated images according to the determination result.

[0172] Or, the computer program is executed by the processor to realize the functions of each module / unit in the above-mentioned device embodiments, for example Figure 2 Modules 201-205 in the above-mentioned device embodiments:

[0173] The correction module 201 is configured to obtain multiple to-be-evaluated images, recognize target features in each to-be-evaluated image, and correct each to-be-evaluated image based on the target features;

[0174] The segmentation module 202 is configured to perform region segmentation on each of the corrected to-be-evaluated images by using a pre-trained image region segmentation model to obtain ground images of each of the corrected to-be-evaluated images;

[0175] The recognition module 203 is configured to perform feature recognition on the obtained multiple ground images to obtain multiple features;

[0176] The classification module 204 is configured to classify the multiple features, and determine whether the multiple ground images are located on the same ground based on a classification result;

[0177] The evaluation module 205 is configured to determine a risk score and obtain a risk evaluation result of the multiple to-be-evaluated images according to the determination result.

[0178] Embodiment four

[0179] Referring to Figure 3 , which is a structural schematic diagram of an electronic device provided in Embodiment Three of the present application. In the preferred embodiment of the present application, the electronic device 3 includes a memory 31, at least one processor 32, at least one communication bus 33, and a transceiver 34.

[0180] Those skilled in the art should understand that Figure 3 The structure of the electronic device shown is not a limitation of the embodiments of the present application, and can be a bus structure or a star structure. The electronic device 3 can also include more or fewer other hardware or software, or different component arrangements than those shown.

[0181] In some embodiments, the electronic device 3 is a device capable of automatically performing numerical calculation and / or information processing according to pre-set or stored instructions, and the hardware thereof includes but is not limited to microprocessors, application-specific integrated circuits, programmable gate arrays, digital processors and embedded devices, etc. The electronic device 3 can also include a client device, which includes but is not limited to any electronic product capable of human-computer interaction with a client through a keyboard, a mouse, a remote controller, a touchpad or a voice control device, such as a personal computer, a tablet computer, a smart phone, a digital camera, etc.

[0182] It should be noted that the electronic device 3 is only an example, and other existing or future electronic products, such as those adaptable to the present application, should also be included in the protection scope of the present application and are hereby incorporated by reference.

[0183] In some embodiments, the memory 31 stores a computer program, and the computer program is executed by the at least one processor 32 to realize all or part of the steps of the risk assessment method as described. The memory 31 includes a Read-Only Memory (ROM), a Programmable Read-Only Memory (PROM), an Erasable Programmable Read-Only Memory (EPROM), a One-time Programmable Read-Only Memory (OTPROM), an Electrically-Erasable Programmable Read-Only Memory (EEPROM), a Compact Disc Read-Only Memory (CD-ROM) or other optical disk memories, magnetic disk memories, magnetic tape memories, or any other computer-readable medium capable of carrying or storing data.

[0184] Further, the computer-readable storage medium can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, at least one application required by a function, etc., and the data storage area can store data created according to the use of the blockchain node, etc.

[0185] The blockchain referred to in the present application is a new application mode of computer technologies such as distributed data storage, peer-to-peer transmission, consensus mechanism, and encryption algorithm. The blockchain is essentially a decentralized database and is a series of data blocks associated using cryptographic methods. Each data block contains information of a batch of network transactions and is used to verify the validity (anti-fake) of the information and generate the next block. The blockchain can include a blockchain underlying platform, a platform product service layer, and an application service layer.

[0186] In some embodiments, the at least one processor 32 is a control unit of the electronic device 3, which connects various components of the entire electronic device 3 through various interfaces and lines, and performs various functions of the electronic device 3 and processes data by running or executing programs or modules stored in the memory 31 and calling data stored in the memory 31. For example, the at least one processor 32 implements all or part of the steps of the risk assessment method described in the embodiments of the present application when executing the computer program stored in the memory, or implements all or part of the functions of the risk assessment device. The at least one processor 32 can be composed of integrated circuits, for example, can be composed of a single packaged integrated circuit, or can be composed of multiple packaged integrated circuits with the same function or different functions, including one or more central processing units (CPU), microprocessors, digital processing chips, graphics processors, and combinations of various control chips, etc.

[0187] In some embodiments, the at least one communication bus 33 is configured to realize the connection and communication between the memory 31, the at least one processor 32, and the like.

[0188] Although not shown, the electronic device 3 can also include a power supply (such as a battery) for powering various components. Preferably, the power supply can be logically connected to the at least one processor 32 through a power management device, so as to realize the functions of managing charging, discharging, and power consumption management, etc. through the power management device. The power supply can also include one or more direct or alternating current power supplies, recharging devices, power failure detection circuits, power converters or inverters, power status indicators, and any other components. The electronic device 3 can also include various sensors, Bluetooth modules, Wi-Fi modules, camera devices, etc., which are not described here.

[0189] The integrated units in the form of software function modules can be stored in a computer readable storage medium. The software function modules are stored in a storage medium, and include a plurality of instructions for causing a computer device (which can be a personal computer, an electronic device, or a network device, etc.) or a processor to perform part of the method described in various embodiments of the present application.

[0190] In several embodiments provided in the present application, it should be understood that the disclosed apparatus and method can be implemented in other manners. For example, the described apparatus embodiments are merely illustrative. For example, the division of the modules is merely logical function division. In actual implementation, another division manner can be used.

[0191] The modules illustrated as separated components can or can not be physically separated, and the components illustrated as modules can or can not be physical units. They can be located in one position or distributed on multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the embodiments.

[0192] In addition, each function module in the various embodiments of the present application can be integrated in a processing unit, or each unit can exist physically, or two or more units can be integrated in one unit. The integrated unit can be implemented in the form of hardware, or in the form of hardware plus software function modules.

[0193] It is obvious for those skilled in the art that the present application is not limited to the details of the above exemplary embodiments, and the present application can be implemented in other specific forms without departing from the spirit or essential characteristics of the present application. Therefore, the embodiments should be considered as exemplary and non-limiting, and the scope of the present application is defined by the appended claims rather than the above description, and all changes falling within the meaning and range of equivalent elements of the claims are intended to be included in the present application. Any reference signs in the claims should not be considered as limiting the involved claims. In addition, it is obvious that the word "comprise" does not exclude other units or, the singular does not exclude the plural. The plurality of units or devices stated in the description can also be implemented by one unit or device through software or hardware. The words first, second, etc. are used to indicate names, and do not indicate any specific order.

[0194] Finally, it should be noted that the above embodiments are merely used to illustrate the technical solutions of the present application, rather than limit the present application. Although the present application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or replaced equivalently without departing from the spirit and scope of the present application.

Claims

1. A risk assessment method, characterized by, The method comprises: obtaining a plurality of to-be-evaluated images, identifying target features in each to-be-evaluated image, and correcting each to-be-evaluated image based on the target features; segmenting each corrected to-be-evaluated image using a pre-trained image region segmentation model to obtain a ground image of each corrected to-be-evaluated image, comprising: inputting each corrected to-be-evaluated image into the image region segmentation model to obtain a foreground region and a background region of each corrected to-be-evaluated image, wherein the to-be-evaluated image comprises a vehicle image, the foreground region comprises a vehicle body region, and the background region comprises a ground region outside the vehicle body and a surrounding environment region; extending the vehicle body region outward according to a preset ratio to obtain an extended region, the extended region does not include the vehicle body region, and the extended region belongs to the ground region; masking the region outside the extended region to obtain an image of the extended region and taking the image of the extended region as the ground image, wherein each ground image corresponds to a corrected to-be-evaluated image; performing feature recognition on the obtained plurality of ground images to obtain a plurality of features; classifying the plurality of features and determining whether the plurality of ground images are located on the same ground based on the classification result; determining a risk score and obtaining a risk evaluation result of the plurality of to-be-evaluated images according to the determination result.

2. The risk assessment method of claim 1, wherein, The to-be-evaluated image comprises a vehicle image, the method of identifying target features in each to-be-evaluated image and correcting each to-be-evaluated image based on the target features comprises: identifying and positioning the target features in each to-be-evaluated image based on optical character recognition (OCR) technology; determining the center point of the target features according to the positioning result; performing origin correction based on the center point with the target features as a reference, wherein the target features comprise a vehicle license plate, and the origin correction comprises rotation correction.

3. The risk assessment method of claim 1, wherein, The training method of the image region segmentation model comprises: obtaining a sample set, dividing the sample set into a training set and a test set, wherein the sample set comprises a preset number of vehicle images; training a DeepLab V3 architecture using the training set to obtain an initial image region segmentation model; testing the initial image region segmentation model using the test set, adjusting the parameters of the initial image segmentation model according to the test result until the image region segmentation model meeting the preset requirements is obtained.

4. The risk assessment method of claim 1, wherein, The method of performing feature recognition on the obtained plurality of ground images comprises a scale-invariant feature transform (SIFT) algorithm; The plurality of features comprises n types of features, and each ground image contains k features, each feature corresponding to a ground image, wherein n and k are both integers greater than or equal to 1.

5. The risk assessment method of claim 4, wherein, The method of classifying the plurality of features comprises a cosine similarity algorithm; The method of determining whether the plurality of ground images are located on the same ground based on the classification result comprises: when the classification result indicates that the k features belong to the same category, determining that the k features are the same type of feature; determining that the ground images containing the same type of feature are located on the same ground.

6. The risk assessment method of claim 1, wherein, The risk score is determined according to the judgment result, and a risk assessment result of the plurality of images to be evaluated is obtained, which comprises: calculating a ratio of the number of ground images located on the same ground to the total number of ground images, and determining the risk score according to the ratio and a preset scoring formula; when the risk score is higher than a preset risk threshold, it is determined that the plurality of images to be evaluated have fraud risks.

7. A risk assessment apparatus for implementing the method as claimed in claim 1, characterized in that, The device comprises a correction module, a segmentation module, an identification module, a classification module and an evaluation module: The correction module is used for acquiring a plurality of images to be evaluated, identifying target features in each image to be evaluated, and correcting each image to be evaluated based on the target features; The segmentation module is used for performing region segmentation on each corrected image to be evaluated by using a pre-trained image region segmentation model, and obtaining ground images of each corrected image to be evaluated; The identification module is used for performing feature identification on the obtained plurality of ground images, and obtaining a plurality of features; The classification module is used for classifying the plurality of features, and judging whether the plurality of ground images are located on the same ground based on the classification result; The evaluation module is used for determining a risk score according to the judgment result and obtaining a risk assessment result of the plurality of images to be evaluated.

8. An electronic device, comprising: The electronic device comprises a processor and a memory, and the processor is used for executing a computer program stored in the memory to realize the risk assessment method according to any one of claims 1 to 6.

9. A computer-readable storage medium having stored thereon a computer program, characterized in that The computer program is executed by the processor to realize the risk assessment method according to any one of claims 1 to 6.

Images