Abnormal gang dynamic identification method and device based on image information, equipment and medium

Abstract

The present application relates to an abnormal group dynamic identification method, device, equipment and medium based on image information, which comprises the following steps: S1. obtaining user information of uploaded images; S2. obtaining image category information of uploaded images; S3. constructing a user portrait and updating a user database; S4. detecting abnormal users, including using a preset abnormal detection method to detect the uploaded images of the users; S5. calculating and determining the correlation degree of the users of the uploaded images and other users; S6. determining the group relationship according to the correlation degree and detecting abnormal users in the group relationship, so as to identify abnormal groups. The present application makes full use of the uploaded images of the users, and timely and accurately identifies and excavates abnormal groups, so as to achieve the purpose of early warning and effectively avoid losses.

Description

Technical Field

[0001] The present invention belongs to the technical field of trademark anti-counterfeiting, and specifically relates to a method, device and medium for dynamically identifying abnormal gangs based on image information. Background Art

[0002] With the widespread adoption and rapid development of smartphones and the internet, a number of mobile apps have emerged, allowing users to conveniently upload images for category identification or authenticity verification. For example, online banking uses image apps to verify identity through ID cards, while the fast-moving consumer goods industry uses image apps to verify the authenticity of product images.

[0003] However, the applicant found that: since the recognition accuracy of the image APP system based on the machine learning method is not absolutely correct and driven by profit, some researchers or counterfeiting gangs will attack by uploading different types or carefully forged images to the image APP in order to obtain forged images that can pass the recognition system in the image APP, thereby making the image APP unable to correctly identify forged images or fake images, and losing its due function.

[0004] Therefore, it is very necessary to timely discover and identify gangs based on the types and behaviors of pictures uploaded by users, and to carry out targeted protection, timely monitoring and early warning of the gangs' research behaviors.

[0005] In the existing group mining scheme, the use of information uploaded by users in the image is not taken into account, and dynamic identification is not performed, resulting in low accuracy and slow speed of mining groups. Summary of the Invention

[0006] In order to overcome the shortcomings of the existing technology, the purpose of the present invention is to provide a dynamic identification method for abnormal groups based on image information, which constructs user portraits by utilizing users' uploaded images, detects abnormal users, calculates and determines the degree of user association, and timely and accurately identifies and mines abnormal groups, as well as devices, equipment and media based on the dynamic identification method for abnormal groups based on image information.

[0007] In order to solve the above problems, the technical solutions adopted by the present invention are as follows:

[0008] The present invention provides a method for dynamically identifying abnormal groups based on image information, comprising:

[0009] S1. Get the user information of the uploaded image;

[0010] S2. Get image category information of the uploaded image;

[0011] S3. Build user profiles and update the user database;

[0012] S4 detects abnormal users, including using a preset abnormality detection method to detect abnormalities in users who upload images, and updates the abnormal user database;

[0013] S5. Calculate and determine the degree of association between the user who uploaded the image and other users, including:

[0014] The association analysis method is used to calculate the similarity between the user who uploaded the image and other users in the user database. If the similarity is greater than the set threshold, it is considered to be associated, otherwise it is not associated;

[0015] S6. Determine the group relationship based on the degree of association, update the group database, and detect abnormal users in the group relationship, thereby identifying abnormal groups and updating the abnormal group database.

[0016] Furthermore, the step of “calculating the similarity between the user who uploaded the image and other users in the user database using an association analysis method” in S5 specifically includes:

[0017] The latitude and longitude coordinates, IP address, image category, and upload time of the geographic location extracted from the user portrait are substituted into the following formula to calculate the actual distance between the user who uploaded the image and other users. In the formula, a is the radian difference between the latitudes of user A and user B, b is the radian difference between the longitudes of user A and user B, lat1 and lat2 are the latitudes of user A and user B, respectively.

[0018]

[0019] The similarity of users who uploaded the images is calculated based on the extracted features.

[0020] Furthermore, the “using a preset anomaly detection method to perform anomaly detection on the user who uploaded the image” in S4 specifically includes:

[0021] Select features and extract features from user portraits;

[0022] User anomaly determination: The anomaly detection model in the preset anomaly detection method determines whether the user who uploaded the image is abnormal based on the extracted features.

[0023] Furthermore, before S4, an anomaly detection model is first established, specifically by obtaining a training set, and then obtaining an anomaly detection model based on the training set through a fitting method.

[0024] Furthermore, S2 specifically includes: obtaining image category information using a preset image classification model and an abnormal image detection model; the image classification model and the abnormal image detection model are both classification algorithms based on deep learning, which are established by collecting corresponding category images as training sets and test sets, and using neural network training to establish two models; wherein the image classification model is used to predict the classification of image objects, and the abnormal image detection model is used to predict different types of abnormal image classifications.

[0025] Furthermore, the user information described in S1 includes at least a user name, a MAC address, a geographical location, an IP address, and a time when the image was uploaded, wherein the MAC address serves as a unique identifier of the user;

[0026] When S3 is executed, the user unique identification code obtained in S1 is used to search in the current user database; if the user is not found, it is determined to be a new user and is given a designated unique user ID. At the same time, a user profile of the user is constructed based on the user information obtained by S1 and the image category information obtained by S2, and stored in the user database; if the user is found to be an existing user, the user information obtained by S1 and the image category information obtained by S2 are added to the user profile of the user, and the user database is updated.

[0027] Furthermore, S6 specifically includes:

[0028] Determine whether the similarity calculated in S5 is greater than a set threshold. If so, determine whether there is a gang relationship between the user who uploaded the image and the selected user in the user database, and update the gang database;

[0029] Check whether there are any abnormal users among the users in the above gang relationships. If so, they are identified as abnormal gangs and the abnormal gang database is updated.

[0030] On the other hand, the present invention also provides a device for dynamically identifying abnormal groups based on image information, comprising:

[0031] A first acquiring unit, configured to acquire information of a user who uploaded an image;

[0032] A second acquiring unit, configured to acquire image category information of an uploaded image;

[0033] User portrait building unit, used to build user portraits and update the user database;

[0034] An abnormal user detection unit, used to detect abnormalities of users who upload images using a preset abnormality detection method and update an abnormal user database;

[0035] The association program calculation and determination unit is used to calculate and determine the degree of association between the user who uploaded the image and other users, including: using an association analysis method to calculate the similarity between the user who uploaded the image and other users in the user database, and when the similarity is greater than a set threshold, it is considered to be associated, otherwise it is not associated;

[0036] The abnormal group determination unit is used to determine the group relationship according to the degree of association, update the group database, and detect abnormal users in the group relationship, thereby identifying abnormal groups and updating the abnormal group database.

[0037] In another aspect, the present invention further provides an electronic device, comprising:

[0038] Memory for storing computer programs;

[0039] The processor is configured to implement the above-mentioned steps of the method for dynamically identifying abnormal groups based on image information when executing the computer program stored in the memory.

[0040] On yet another aspect, the present invention further provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the above-mentioned method for dynamically identifying abnormal groups based on image information are implemented.

[0041] Compared with the prior art, the present invention has the following beneficial effects:

[0042] The present invention constructs user portraits by utilizing users' uploaded images, and detects abnormal users, calculates and determines the degree of user association, so as to timely and accurately identify and unearth abnormal groups such as counterfeiting groups, achieve early warning purposes, and effectively avoid losses. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 1 is a flow chart of the method for dynamically identifying abnormal groups based on image information according to the present invention;

[0044] Figure 2 This is a schematic diagram of using an undirected graph with a data structure to represent a group of users without abnormalities in the method for dynamic identification of abnormal groups based on image information of the present invention;

[0045] Figure 3 This is a schematic diagram of an abnormal group represented by an undirected graph data structure in the method for dynamic identification of abnormal groups based on image information of the present invention;

[0046] Figure 4 This is a schematic diagram of using an undirected graph with a data structure to represent another abnormal group in the method for dynamic identification of abnormal groups based on image information of the present invention;

[0047] Figure 5It is a structural diagram of the device for dynamic identification of abnormal groups based on image information according to the present invention;

[0048] Figure 6 It is a structural diagram of the abnormal detection model building module in the abnormal group dynamic identification device based on image information of the present invention;

[0049] Figure 7 This is a schematic diagram of the structure of the database in the device for dynamically identifying abnormal groups based on image information according to the present invention;

[0050] Figure 8 It is a structural diagram of the electronic device in the abnormal group dynamic identification device based on image information of the present invention;

[0051] Figure numerals: 10, first acquisition unit; 20, second acquisition unit; 30, user portrait construction unit; 40, abnormal user detection unit; 401, abnormal detection model construction module; 402, normal or abnormal user prediction subunit; 403, training sample acquisition subunit; 404, model training subunit; 50, association program calculation and judgment unit; 60, abnormal group judgment unit; 70, early warning module; 81, processor; 82, memory. DETAILED DESCRIPTION

[0052] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0053] See also Figure 1 The present invention provides a method for dynamically identifying abnormal groups based on image information, comprising:

[0054] Step S100. Obtain the user information of the uploaded image, which includes at least the user name, MAC address, geographic location (including the user's city, latitude and longitude coordinates), IP address, and time of uploading the image, wherein the MAC address serves as the user's unique identifier.

[0055] Step S200. Obtain image category information of the uploaded image; specifically, obtain image category information by using a preset image classification model and an abnormal image detection model; the obtaining process is to input the uploaded image of the user into the preset image classification model and the preset abnormal image detection model for prediction to obtain the category information of the image; for example, the image classification model can predict a certain type of commodity, and the abnormal image detection model can predict a screen flipping image, a non-real object image, and the like; the image classification model and the abnormal image detection model are both classification algorithms based on deep learning (the classification algorithm is an image classification method based on deep learning (such as ResNet-50, GoogleNet, etc.), and various image categories can be considered according to actual conditions; the abnormal image includes an original clear real image, a copied image, a black and white image, and a forged image such as a screen flipping image), two models are established by collecting corresponding category images as a training set and a test set and using a neural network; the image classification model is used to predict the real object classification of the image, the abnormal image detection model is used to predict the classification of different types of abnormal images, and the deep learning can preferably be a deep learning residual network RESNET-50, or other deep learning methods such as VGG, GoogLeNet, etc.

[0056] Step S300. Construct a user portrait and update a user database; specifically, find the user unique identification code obtained in step S100 in the current user database;

[0057] If not found, it is determined that the user is new, a specified unique user ID is given, and the user portrait of the user is constructed according to the user information obtained in step S100 and the image category information obtained in step S200, and stored in the user database; wherein the construction of the user portrait of the user is specifically: constructing the portrait of the user from the user attribute dimension, the user behavior dimension, etc. by defining statistical labels (such as night owl users: users who often upload images at midnight), rule labels (such as the gender of the user), learning labels (such as abnormal users), etc.

[0058] If found, the user information obtained in step S100 and the image category information obtained in step S200 are added to the user portrait of the user, and the user database is updated; for example, data is added to the corresponding database by using a data addition command in Hbase in big data.

[0059] Step S400. Detecting an abnormal user, including detecting an abnormal user of the uploaded image by using a preset abnormality detection method, and updating an abnormal user database; wherein the "detecting an abnormal user of the uploaded image by using a preset abnormality detection method" specifically includes:

[0060] Selecting a feature, extracting a feature from a user portrait, that is, extracting longitude and latitude coordinates, an IP address, an image category and the like of a user corresponding to the user portrait from a user database;

[0061] User abnormality determination, the abnormality detection model in the preset abnormality detection method determines whether the user of the uploaded image is abnormal according to the extracted feature, specifically, the abnormality detection model calculates an abnormality score value of the user (the abnormality score value ranges from 0 to 1) according to the extracted feature, when the abnormality score value is close to or equal to 1, it is determined that the user is abnormal, otherwise normal.

[0062] In addition, before this step, an abnormality detection model needs to be established first, specifically: obtaining a training set (that is, selecting some features of a user as a training set, that is, a feature set of the user), and then obtaining an abnormality detection model by a fitting method according to the training set; wherein the fitting method belongs to the prior art, such as: OneClassSVM and Local Outlier Factor and the like; the present application is aimed at the feature that the number of abnormal users is much less than that of normal users, and preferably adopts an isolation forest algorithm, that is: different feature combinations are selected from the feature set of the user to train and test by using the isolation forest algorithm, a feature combination with a better test result is selected to obtain an isolation tree, that is, an abnormality detection model, when determining the abnormality of the user, the abnormality detection model calculates an abnormality score value of the user through the isolation tree, the closer to 1, the more abnormal, and the ensemble.IsolationForest module provided in sklearn is used to realize the Isolation Forest algorithm to calculate the abnormality score value of the user.

[0063] For example:

[0064] from sklearn.ensemble import IsolationForest

[0065] X=[[-1.1],[0.3],[0.5],

[100] ]

[0066] clf=IsolationForest(random_state=0).fit(X)

[0067] clf.predict([[0.1],[0],

[90] )

[0068] array([1,1,0])

[0069] In the example above, X represents the feature set of four users. Using the IsolationForest(random_state=0).fit(X) function, we obtain an anomaly detection model. Then, using the predict function on the features of the three users, we obtain the scores of 1 (normal user), 1 (normal user), and 0 (abnormal user). In the implementation, we can also set a threshold of 0.9. If the anomaly score exceeds 0.9, the user is considered to be in the abnormal user class 0; otherwise, it is considered to be in the normal user class 1.

[0070] Step S500: Calculate and determine the degree of association between the user who uploaded the image and other users, including:

[0071] The association analysis method is used to calculate the similarity between the user who uploaded the image and other users in the user database. When the similarity is greater than the set threshold (the range of the set threshold is [0,1]), it is considered to be associated, otherwise it is not associated;

[0072] The “using an association analysis method to calculate the similarity between the user who uploaded the image and other users in the user database” specifically includes:

[0073] The latitude and longitude coordinates, IP address, image category, and upload time of the geographic location extracted from the user portrait (i.e., the latitude and longitude coordinates, IP address, image category, and upload time of the user corresponding to the user portrait are extracted from the user database). The latitude and longitude coordinates are substituted into the following formula to calculate the actual distance between the user who uploaded the image and other users, where a is the radian difference between the latitudes of user A and user B, b is the radian difference between the longitudes of user A and user B, lat1 and lat2 are their latitudes, respectively.

[0074]

[0075] The similarity of users who uploaded the images is calculated based on the extracted features. Specifically, the extracted features (qualitative features) are converted into quantitative features using feature engineering methods. Then, the similarity of users who uploaded the images is calculated based on the quantitative features and methods such as the Euclidean distance formula, cosine similarity, and Hamming distance.

[0076] Step S600. Determine the group relationship based on the degree of association, update the group database, and detect abnormal users in the group relationship, thereby identifying abnormal groups and updating the abnormal group database; specifically including:

[0077] Determine whether the similarity calculated in step S500 is greater than a set threshold. If so, determine whether there is a group relationship between the user who uploaded the image and the selected user in the user database, and update the group database;

[0078] Check whether there are any abnormal users among the users in the above gang relationships. If so, they are identified as abnormal gangs and the abnormal gang database is updated.

[0079] In the group database and abnormal group database, the group relationship can be stored in the graph data structure undirected graph. In the data structure undirected graph, each user is a node in the graph. If the user is an abnormal user, the node is marked as an abnormal node. The edge between two nodes indicates that there is a relationship between the two users, such as Figure 2 Therefore, in the graph, a connected branch represents a group. If there is an abnormal node in the connected branch, the branch represents an abnormal group. An isolated node in an undirected graph indicates that the user does not belong to any group, such as Figure 2 In this way, as the information of user-uploaded images is continuously updated and changed, the groups and abnormal groups also show dynamic changes. The original normal user node can become an abnormal node, the original two different groups can become the same group, and the original group can become an abnormal group.

[0080] Specifically, see Figure 2 , user 1 and user 3, user 2 and user 4 are considered to be different groups. After the user to be tested is introduced, these two groups plus the user to be tested are considered to be the same group. Figure 3 , since user 1 in the group is an abnormal user, the entire group is an abnormal group. Figure 4 In the example, users 1, 2, and 3 are a group. After associating with a new abnormal user, the group is identified as an abnormal group. This shows that groups and abnormal groups can be dynamically identified based on user-uploaded images. Implementing the method of the present invention enables abnormal group identification based on image information.

[0081] Based on the same inventive concept, the embodiment of the present application also provides a dynamic identification device for abnormal groups based on image information, such as Figure 5 As shown, the apparatus for dynamic identification of abnormal groups based on image information of the present invention includes a first acquisition unit 10, a second acquisition unit 20, a user profile construction unit 30, an abnormal user detection unit 40, an associated program calculation and determination unit 50, and an abnormal group determination unit 60. Specifically:

[0082] The first acquisition unit 10 is used to obtain user information of the uploaded image; for example: after the user uploads an image by taking a photo with a mobile phone, the first acquisition unit 10 obtains user information such as the user name, mobile phone MAC address, geographic location (including the user's city, latitude and longitude coordinates), IP address, and time of uploading the image from the uploaded image information of the user's mobile phone.

[0083] The second acquisition unit 20 includes an image classification model and an abnormal image detection model based on a deep learning classification algorithm, and acquires image category information of the uploaded image by using a preset image classification model and abnormal image detection model.

[0084] The user portrait construction unit 30 is used to construct a user portrait and update the user database; the user portrait includes at least the user name, MAC address, city, latitude and longitude, uploaded image category, whether to upload abnormal images, number of image uploads, image upload time period, etc., and the user portrait is used to provide subsequent user feature extraction.

[0085] The abnormal user detection unit 40 is used to perform abnormality detection on the user who uploaded the image using a preset abnormality detection method and update the abnormal user database; specifically, it is used to extract features from the user portrait and determine whether the user who uploaded the image is abnormal based on the extracted features using the abnormality detection model in the preset abnormality detection method. When the abnormality score value of the user calculated by the abnormality detection model based on the extracted features is close to or equal to 1, the user is determined to be abnormal; otherwise, the user is normal.

[0086] The abnormal user detection unit 40 includes an abnormal detection model building module 401 and a normal or abnormal user prediction subunit 402. Figure 6 As shown, the anomaly detection model construction module 401 includes a training sample acquisition subunit 403 and a model training subunit 404. The training sample acquisition subunit 403 of the anomaly detection model construction module 401 obtains a training set, marks the user (the mark is specifically: the normal category is marked as 0 and the abnormal category is marked as 1), and the parameters of the fitting distribution (mean and variance) are obtained to obtain an anomaly detection model; the model training subunit 404 is a feature selection subunit, which is used to train the anomaly detection model according to different feature combinations on the training set in the training sample acquisition subunit 403 according to a selected anomaly detection algorithm (such as the isolation forest algorithm) to obtain a better anomaly detection model as the final anomaly detection model; the normal or abnormal user prediction subunit 404 is an abnormal user determination subunit, which uses the obtained anomaly detection model to perform anomaly prediction on the user to be tested. When the prediction result is 0, the user is determined to be an abnormal user, and when the prediction result is 1, the user is determined to be a normal user.

[0087] The association program calculation and determination unit 50 is used to calculate and determine the degree of association between the user who uploaded the image and other users, including: using an association analysis method to calculate the similarity between the user who uploaded the image and other users in the user database. When the similarity is greater than a set threshold, it is considered to be associated, otherwise it is not associated.

[0088] The abnormal group determination unit 60 is used to determine the group relationship according to the degree of association, update the group database, and detect abnormal users in the group relationship, thereby identifying abnormal groups and updating the abnormal group database.

[0089] In addition, the apparatus for dynamically identifying abnormal groups based on image information of the present invention further includes an early warning module 70 , which is used to issue an early warning and take corresponding measures when the current user is identified as an abnormal user or an abnormal group.

[0090] As attached Figure 7 As shown, the abnormal group dynamic identification device based on image information of the present invention also includes a user database, an abnormal user database, a group database and an abnormal group database. The abnormal group dynamic identification device based on image information dynamically creates and updates the above four databases during the identification process; wherein abnormal users are determined by using an abnormality detection method on users in the user database, and the abnormal user database is updated; groups are determined by calculating the correlation between users in the user database, and the group database is dynamically updated; the group database is searched, and the abnormal group database is updated according to whether there are abnormal users in the newly joined group; and users in abnormal groups are marked as abnormal users, and then the abnormal user database is updated.

[0091] Based on the same inventive concept, an electronic device is also provided in the embodiment of the present application. The electronic device can realize the function of the aforementioned abnormal group dynamic identification device based on image information. Figure 8 The electronic device includes: at least one processor 81, and a memory 82 connected to the at least one processor 81, and a specific connection medium between the processor 81 and the memory 82, the connection medium can be a bus 83 (for convenience of representation, Figure 8 (Indicated by only one bold line, this does not mean there is only one bus or only one type of bus.) The bus 83 can be divided into an address bus, a data bus, a control bus, etc. The processor 81 and the memory 82 are interconnected via the bus 83. The memory 82 stores the four databases described above (i.e., the user database, the abnormal user database, the group database, and the abnormal group database), as well as instructions executable by at least one processor 81. By executing the instructions stored in the memory 82, the at least one processor 81 can implement the abnormal group dynamic identification method based on image information discussed above.

[0092] Furthermore, the processor 81 can implement Figure 5 The functions of each module in the device shown. The processor 81 is the control center of the device, and can use various interfaces and lines to connect the various parts of the entire control device, by running or executing instructions stored in the memory 82 and calling data stored in the memory 82. The processor 81 can be a general-purpose processor, such as a central processing unit (CPU), a digital signal processor, an application-specific integrated circuit, a field programmable gate array or other programmable logic device, a discrete gate or transistor logic device, a discrete hardware component, and can implement or execute the various methods, steps and logic block diagrams disclosed in the embodiments of the present application. A general-purpose processor can be a microprocessor or any conventional processor, etc. In combination with the steps of the method for dynamic identification of abnormal groups based on image information disclosed in the embodiments of the present application, it can be directly embodied as being executed by a hardware processor, or executed by a combination of hardware and software modules in the processor.

[0093] The memory 82 is a non-volatile computer-readable storage medium that can be used to store non-volatile software programs, non-volatile computer executable programs and modules. The memory 82 may include at least one type of storage medium, such as a flash memory, a hard disk, a multimedia card, a card-type memory, a random access memory (RAM), a static random access memory (SRAM), a programmable read-only memory (PROM), a read-only memory (ROM), an electrically erasable programmable read-only memory (EEPROM), a magnetic memory, a magnetic disk, an optical disk, etc. The memory 82 is any other medium that can be used to carry or store desired program code in the form of instructions or data structures and can be accessed by a computer, but is not limited thereto. The memory 82 in the embodiment of the present application can also be a circuit or any other device that can realize a storage function, for storing program instructions and / or data.

[0094] By programming the processor 81, the code corresponding to the method for dynamic identification of abnormal groups based on image information described in the above embodiment can be fixed into the chip, so that the chip can execute the code when running. Figure 1 The steps of the method for dynamic identification of abnormal groups based on image information in the embodiment shown are as follows: How to design and program the processor 81 is well known to those skilled in the art and will not be described in detail here.

[0095] The above embodiments are only preferred embodiments of the present invention and cannot be used to limit the scope of protection of the present invention. Any non-substantial changes and replacements made by technicians in this field on the basis of the present invention fall within the scope of protection required by the present invention.

Claims

1. A method for dynamic identification of abnormal groups based on image information, characterized in that: include; S1 obtains user information for uploading images, wherein the user information includes at least user name, MAC address, geographic location, IP address, and time of uploading the image, wherein the MAC address serves as a unique user identifier; S2. Obtaining image category information for the uploaded image, specifically by using a preset image classification model and an abnormal image detection model to obtain image category information; both the image classification model and the abnormal image detection model are deep learning-based classification algorithms, which are established by collecting images of corresponding categories as training and test sets and using neural network training; wherein the image classification model is used to predict the classification of physical objects in the image, and the abnormal image detection model is used to predict different types of abnormal image classifications; S3. Build a user profile and update the user database. When S3 is executed, the user's unique identification code obtained in S1 is searched in the current user database; if it cannot be found, it is determined to be a new user and is given a unique user ID. At the same time, based on the user information obtained in S1 and the image category information obtained in S2, a user profile of the user is constructed and stored in the user database; If the user is found to be an existing user, the user information obtained in S1 and the image category information obtained in S2 are added to the user profile of the user, and the user database is updated; S4 detects abnormal users, including using a preset abnormality detection method to detect abnormalities in users who upload images, and updates the abnormal user database; S5. Calculate and determine the degree of association between the user who uploaded the image and other users, including: The association analysis method is used to calculate the similarity between the user who uploaded the image and other users in the user database. If the similarity is greater than the set threshold, it is considered to be associated, otherwise it is not associated; S6. Determine group relationships based on the degree of association, update the group database, and detect abnormal users in the group relationships, thereby identifying abnormal groups and updating the abnormal group database; specifically including: Determine whether the similarity calculated in S5 is greater than a set threshold. If so, determine whether there is a gang relationship between the user who uploaded the image and the selected user in the user database, and update the gang database; Check whether there are any abnormal users among the users in the above gang relationship. If there are, they will be identified as abnormal gangs and the abnormal gang database will be updated; The "using the association analysis method to calculate the similarity between the user who uploaded the image and other users in the user database" described in S5 specifically includes: The latitude and longitude coordinates, IP address, image category, and upload time of the geographic location extracted from the user portrait are substituted into the following formula to calculate the actual distance between the user who uploaded the image and other users. In the formula, a is the radian difference between the latitudes of user A and user B, b is the radian difference between the longitudes of user A and user B, lat1 and lat2 are the latitudes of user A and user B, respectively. The similarity of users who uploaded the images is calculated based on the extracted features.

2. The method according to claim 1, characterized in that The "using a preset anomaly detection method to detect anomalies on users who upload images" mentioned in S4 specifically includes: Select features and extract features from user portraits; User anomaly determination: The anomaly detection model in the preset anomaly detection method determines whether the user who uploaded the image is abnormal based on the extracted features.

3. The method according to claim 2, characterized in that Before S4, an anomaly detection model is first established, specifically: a training set is obtained, and then an anomaly detection model is obtained based on the training set through a fitting method.

4. A device based on the method for dynamic identification of abnormal groups based on image information according to any one of claims 1 to 3, characterized in that: include: A first acquiring unit, configured to acquire information of a user who uploaded an image; A second acquiring unit, configured to acquire image category information of an uploaded image; User portrait building unit, used to build user portraits and update the user database; An abnormal user detection unit, used to detect abnormalities of users who upload images using a preset abnormality detection method and update an abnormal user database; The association program calculation and determination unit is used to calculate and determine the degree of association between the user who uploaded the image and other users, including: using an association analysis method to calculate the similarity between the user who uploaded the image and other users in the user database, and when the similarity is greater than a set threshold, it is considered to be associated, otherwise it is not associated; The abnormal group determination unit is used to determine the group relationship according to the degree of association, update the group database, and detect abnormal users in the group relationship, thereby identifying abnormal groups and updating the abnormal group database.

5. An electronic device, characterized in that: include: Memory for storing computer programs; A processor, configured to implement the method steps of any one of claims 1 to 3 when executing the computer program stored in the memory.

6. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method steps according to any one of claims 1 to 3 are implemented.

Images