Disclosure of Invention
In view of this, the embodiments of the present application provide a suspicious item member identification method based on image processing, so as to solve the technical defects existing in the prior art. The embodiment of the application also provides a suspicious item member identification device based on image processing, a computing device and a computer readable storage medium.
The embodiment of the application discloses a suspicious item member identification method based on image processing, which comprises the following steps:
acquiring a case auditing image set composed of case auditing images submitted by project members participating in a project;
extracting feature vectors of the case auditing images in the case auditing image set;
determining the image similarity between the case auditing images according to the feature vector;
clustering the case review images based on the similarity weights determined by the image similarity;
and determining suspicious project members in the project members according to the clustering result of the clusters.
Optionally, the extracting the feature vector of the case review image in the case review image set includes:
and inputting the case auditing image into a pre-trained deep learning model for image vectorization processing, and outputting the feature vector of the case auditing image.
Optionally, the determining the image similarity between the case review images according to the feature vector includes:
and calculating the image similarity between the case auditing images by using a word frequency inverse text frequency index algorithm according to the feature vector.
Optionally, the determining the image similarity between the case review images according to the feature vector includes:
and determining the image similarity between the case auditing images based on the vector distance between the feature vectors of the case auditing images.
Optionally, the clustering the case review image based on the similarity weight determined by the image similarity includes:
constructing a communication graph, establishing a one-to-one correspondence between the case auditing images and nodes in the communication graph, and taking a logarithmic value of the image similarity between the case auditing images as an edge weight between the nodes corresponding to the case auditing images;
and inputting the connected graph into a clustering model for clustering, and outputting the clustering identification of the case auditing image.
Optionally, the determining suspicious item members in the item members according to the clustering result of the clustering includes:
Determining the case auditing images with the same cluster identifier as suspicious images;
determining that the submission of the suspicious image is a suspicious one of the item members.
Optionally, after the step of determining suspicious item members in the item members according to the clustering result of the clustering is performed, the method includes:
and determining the set of submitters of the case auditing images with the same cluster identifier in the suspicious image as suspicious group partners.
Optionally, after the step of determining suspicious item members in the item members according to the clustering result of the clustering is performed, the method further includes:
acquiring the factual relationship of the suspicious item members;
determining that the item member with the actual factual relationship is an untrusted item member from the suspicious item members based on the factual relationship;
and determining a member set submitted with the case audit image with the same cluster identifier in the untrusted project members as a suspicious group partner.
Optionally, before the step of acquiring the case auditing image set composed of the case auditing images submitted by the project members participating in the project is executed, the method further includes:
acquiring a factual relationship of the project members;
Determining item members having actual factual relationships based on the factual relationships;
correspondingly, in the case auditing image set step of acquiring case auditing images submitted by project members participating in a project, the case auditing images of the project members with actual facts are acquired to form the case auditing image set.
Optionally, the factual relationship of the item members includes at least one of:
and fund transfer relationship, communication relationship, address book relationship and equipment relationship among the project members.
Optionally, the deep learning model is constructed based on any one of the following neural networks:
convolutional neural network, deep neural network.
Optionally, the clustering model is based on any one of the following algorithms:
label propagation algorithm, maximum connectivity graph algorithm.
The application provides a suspicious item member identification device based on image processing, which comprises:
the case auditing image set acquisition module is configured to acquire a case auditing image set composed of case auditing images submitted by project members participating in a project;
the feature vector extraction module is configured to extract feature vectors of the case auditing images in the case auditing image set;
An image similarity determining module configured to determine image similarity between the case review images according to the feature vector;
a clustering module configured to cluster the case review images based on the similarity weights determined by the image similarities;
and the suspicious item member determining module is configured to determine suspicious item members in the item members according to the clustering result of the clustering.
The present application provides a computing device comprising:
a memory and a processor;
the memory is for storing computer-executable instructions, and the processor is for executing the computer-executable instructions:
acquiring a case auditing image set composed of case auditing images submitted by project members participating in a project;
extracting feature vectors of the case auditing images in the case auditing image set;
determining the image similarity between the case auditing images according to the feature vector;
clustering the case review images based on the similarity weights determined by the image similarity;
and determining suspicious project members in the project members according to the clustering result of the clusters.
The present application provides a computer readable storage medium storing computer instructions which, when executed by a processor, implement the steps of the suspicious item membership identification method based on image processing.
Compared with the prior art, the application has the following advantages:
the application provides a suspicious item member identification method based on image processing, which comprises the following steps: acquiring a case auditing image set composed of case auditing images submitted by project members participating in a project; extracting feature vectors of the case auditing images in the case auditing image set; determining the image similarity between the case auditing images according to the feature vector; clustering the case review images based on the similarity weights determined by the image similarity; and determining suspicious project members in the project members according to the clustering result of the clusters.
According to the suspicious item member identification method based on image processing, image similarity calculation is conducted on the case auditing images in the case auditing image set, and image clustering operation is conducted on the case auditing images according to the image similarity calculation result, so that the case auditing images with similarity are marked with the same identification, suspicious item members with the possibility of cheating and guaranteeing are identified according to the identification, and the occurrence probability of cheating and guaranteeing is reduced.
Detailed Description
In the following description, numerous specific details are set forth in order to provide a thorough understanding of the present application. This application is, however, susceptible of embodiment in many other ways than those herein described and similar generalizations can be made by those skilled in the art without departing from the spirit of the application and the application is therefore not limited to the specific embodiments disclosed below.
The terminology used in the one or more embodiments of the specification is for the purpose of describing particular embodiments only and is not intended to be limiting of the one or more embodiments of the specification. As used in this specification, one or more embodiments and the appended claims, the singular forms "a," "an," and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. It should also be understood that the term "and/or" as used in one or more embodiments of the present specification refers to and encompasses any or all possible combinations of one or more of the associated listed items.
It should be understood that, although the terms first, second, etc. may be used in one or more embodiments of this specification to describe various information, these information should not be limited by these terms. These terms are only used to distinguish one type of information from another. For example, a first may also be referred to as a second, and similarly, a second may also be referred to as a first, without departing from the scope of one or more embodiments of the present description. The word "if" as used herein may be interpreted as "at … …" or "at … …" or "responsive to a determination", depending on the context.
The application provides a suspicious item member identification method based on image processing, and also provides a suspicious item member identification device based on image processing, a computing device and a computer readable storage medium. The following detailed description, together with the drawings of the embodiments provided herein, respectively, describes the steps of the method one by one.
The embodiment of the suspicious item member identification method based on image processing provided by the application is as follows:
referring to fig. 1, a flowchart of a suspicious item member identification method based on image processing according to the present embodiment is shown; referring to fig. 2, a schematic diagram of a suspicious item member identification procedure based on image processing according to the present embodiment is shown.
Step S102, a case auditing image set composed of case auditing images submitted by project members participating in a project is obtained.
The insurance claim link generally requires the user to upload claim materials, such as real figures, and takes car insurance as an example, the insurance company can require the user to submit the appearance figures of the damaged vehicles so as to judge whether the claim application submitted by the user accords with the insurance guarantee clause. The project members participating in the project are users participating in the insurance project, and the case auditing images are physical pictures in claim materials submitted by the users participating in the insurance project, such as appearance pictures of damaged vehicles, traffic accident scene pictures and the like. In practical application, the claim-settling pictures uploaded by the claim-settling partners have similarity to a certain extent, so that the claim-settling partners can be identified by analyzing the claim-settling pictures, and the probability of occurrence of the claim-settling conditions of the claim-settling partners in the insurance claim-settling links is reduced.
According to the embodiment, the case auditing images in the case auditing image set are subjected to image recognition, so that generalized similarity relations of all the case auditing images in the case auditing image set are obtained, the similarity relations are expressed as that similar case auditing images belong to the same community and are gathered into the same class, suspicious item members are recognized by tracing submitters of the case auditing images belonging to the same community, and further, suspicious group partners can be defined according to communities to which the suspicious item members belong on the basis of recognizing the suspicious item members, so that the occurrence of cheating protection conditions is prevented.
For the purpose of reducing the calculation amount of image recognition, the range of the image recognition object may be narrowed before the image recognition is performed, and in a preferred implementation provided in the embodiment of the present application, before the case review image set composed of the case review images submitted by the project members of the participating project is acquired, the range of the image recognition object is narrowed by the following implementation manner:
acquiring a factual relationship of the project members;
determining item members having actual factual relationships based on the factual relationships;
correspondingly, in the process of acquiring a case auditing image set composed of case auditing images submitted by project members participating in a project, acquiring the case auditing images of the project members with actual facts, and composing the case auditing image set.
According to the implementation mode, firstly, screening is conducted on the project members based on whether the project members have actual fact relations, then, the case auditing image set formed by the case auditing images submitted by the project members with the actual fact relations is obtained, and image recognition is conducted on the basis of the case auditing image set formed, so that the image recognition range is reduced, and meanwhile, the possibility of recognizing suspicious project members is higher.
Wherein the factual relationship of the item members includes at least one of: and fund transfer relationship, communication relationship, address book relationship and equipment relationship among the project members.
The partner members participating in the cheating and guaranteeing the partner can carry out transaction exchange and intercommunication information through a certain medium in actual application, have funds transfer transaction records, communicate through communication equipment, the opposite sides are address book contacts, the multiple parties use the same login equipment, and the like, and all the parties belong to the fact relation, wherein the fact relation means that the project members are detected to have at least one relation in the fact relation, and the project members with the fact relation are judged to be suspicious project members under the condition that the project members are identified to submit case audit images belonging to the same community.
And step S104, extracting the characteristic vector of the case auditing image in the case auditing image set.
In practical application, the number of claim settlement pictures obtained by an insurance company in an insurance claim settlement link in a verification process is relatively large, so as to improve image processing efficiency, and in a preferred implementation manner provided in the embodiment of the application, the extracting feature vectors of the case audit images in the case audit image set includes: and inputting the case auditing image into a pre-trained deep learning model for image vectorization processing, and outputting the feature vector of the case auditing image.
In the process of calculating the similarity of the case auditing images, the images are required to be subjected to vectorization processing, the process of converting the images into vectors is realized based on a deep learning model, the deep learning model is a model which is trained by training samples in advance, the case auditing images are used as input quantities and are input into the deep learning model, and the output of the model is the feature vector of the case auditing images.
The deep learning model is preferably constructed based on a convolutional neural network (CNN, convolutional Neural Network) or a deep neural network (DNN, deep Neural Network), but other neural networks may be used, and the embodiment is not limited thereto.
And S106, determining the image similarity between the case auditing images according to the feature vectors.
In a preferred implementation manner provided in the embodiments of the present application, the determining, according to the feature vector, the image similarity between the case review images includes: and calculating the image similarity between the case auditing images by using a word Frequency inverse text Frequency index (TF-IDF, term Frequency-Inverse Document Frequency) algorithm according to the feature vector.
The word frequency inverse text frequency index algorithm is a statistical method, and an object is used for texts to evaluate the importance degree of a word to one document in a document set or a corpus.
In a preferred implementation manner provided in the embodiments of the present application, the determining, according to the feature vector, the image similarity between the case review images includes: and determining the image similarity between the case auditing images based on the vector distance between the feature vectors of the case auditing images.
The vector distance between the feature vectors of the case auditing image comprises at least one of the following: euclidean distance, cosine distance; the cosine distance is characterized in that the cosine value is close to 1, the included angle tends to be 0, and the image similarity is higher as the two vectors are more similar, and the Euclidean distance is characterized in that the smaller the Euclidean distance is, the more similar the two vectors are, and the higher the image similarity is.
And S108, clustering the case auditing images based on the similarity weight determined by the image similarity.
In a preferred implementation manner provided in the embodiments of the present application, the clustering the case review images based on the similarity weights determined by the image similarity includes:
constructing a communication graph, establishing a one-to-one correspondence between the case auditing images and nodes in the communication graph, and taking a logarithmic value of the image similarity between the case auditing images as an edge weight between the nodes corresponding to the case auditing images;
and inputting the connected graph into a clustering model for clustering, and outputting the clustering identification of the case auditing image.
Preferably, the clustering model is preferably implemented based on a label propagation algorithm (LPA, label Propagation Algorithm) or a maximum connectivity graph algorithm, but other algorithms may be used, which is not limited herein.
Taking the implementation of the clustering model based on a label propagation algorithm as an example, the clustering step is described:
the method comprises the steps that a clustering model is input into a communication graph constructed based on case auditing images and image similarity, in particular, in the process of constructing the communication graph, a one-to-one correspondence is established between nodes contained in the communication graph and the case auditing images, each node contains an image number corresponding to the case auditing image, the obtained image similarity between the case auditing images is taken as a logarithmic value to serve as edge weight between the case auditing image corresponding nodes, and the construction of the communication graph required by the clustering model is completed.
The clustering model is based on a label propagation algorithm, and the application scene of the label propagation algorithm is as follows: community discovery in the traditional sense refers to a group of nodes in a network with larger similarity, so that an internal connection is compact, an external sparse group structure is formed, a process of searching a community structure of a given network diagram is called community discovery, and a process of community discovery is a clustering process in general; the basic idea of the tag propagation algorithm is: the label with the largest number in the labels of the neighbor nodes of one node is used as the label of the node, specifically, each node is added with the label to represent the community to which the node belongs, and the community structure of the same label is formed through the propagation of the label.
The propagation process of the tag propagation algorithm can be summarized as:
1) Initially, each node is given a unique tag;
2) Each node uses the most tags in the tags of the neighbor nodes to update the own tag;
3) Step 2) is repeatedly performed until no more changes in the label of each node occur.
Inputting the connected graph into the clustering model to run a label propagation algorithm for clustering, dividing the case auditing images into communities, and outputting labels representing each community after division, wherein the labels are the clustering identifications of the case auditing images.
And step S110, determining suspicious ones of the item members according to the clustering result of the clusters.
In a preferred implementation manner provided in the embodiment of the present application, the determining a suspicious item member among the item members according to the clustering result of the clustering includes:
determining the case auditing images with the same cluster identifier as suspicious images;
determining that the submission of the suspicious image is a suspicious one of the item members.
In the step of constructing the connected graph, the image similarity between the case auditing images is taken as a logarithmic value to be used as the edge weight between the nodes corresponding to the case auditing images, so that the case auditing images with high similarity between the case auditing images are also high in weight of the edges of the corresponding nodes in the connected graph, the nodes are more tightly connected, the same community is easier to form, the clustering identification of the case auditing images output by the clustering model can be used as a standard for dividing the community, the case auditing images with the same clustering identification belong to the same community, the case auditing images are determined to be suspicious images, and accordingly, the suspicious images are submitted to suspicious item members in the item members.
Preferably, after the step of determining suspicious item members among the item members according to the clustering result of the clustering is performed, the method includes: and determining the set of submitters of the case auditing images with the same cluster identifier in the suspicious image as suspicious group partners.
After the suspicious item members are determined, the case auditing images with the same clustering identification form the same community according to the clustering result, and a group consisting of submitters corresponding to the case auditing images in the same community is likely to be suspicious group partners.
In a preferred implementation manner provided in this embodiment of the present application, after determining suspicious item members among the item members according to the clustering result of the clustering, on the basis of the clustering result obtained by the clustering, whether the suspicious item members are from the same spoofing party may also be determined by analyzing the factual relationship, which is specifically implemented as follows:
1) Acquiring the factual relationship of the suspicious item members;
2) Determining that the item member with the actual factual relationship is an untrusted item member from the suspicious item members based on the factual relationship;
3) And determining a member set submitted with the case audit image with the same cluster identifier in the untrusted project members as a suspicious group partner.
After determining the suspicious item members, screening the item members with actual fact relationships as unreliable item members based on the obtained fact relationships of the suspicious item members, wherein the graph clustering operation firstly realizes the first round of screening, screens case auditing images divided into the same community, submitters of the case auditing images have suspicions of group cheating, and secondly judges whether the suspicious item members have actual relationships in real life by judging whether the suspicious item members have the actual relationships or not, and through the two rounds of screening, the screened item members simultaneously have two conditions that submitted case auditing images are the suspicious images and the fact relationships among the item members exist, and the accuracy of the cheating security group counseling formed by the set of the unreliable item members is determined to be higher on the basis.
The suspicious item member identification method based on image processing provided by the application is further described with reference to fig. 2, and the method is specifically implemented as follows:
step S202, N claim pictures uploaded by a user are acquired in an insurance claim link.
The insurance claim link generally requires the user to upload claim materials, such as claim pictures, and the insurance company analyzes the claim materials to determine whether the user meets the claim conditions. In this embodiment, suspicious item members are identified based on an image processing technology, first, N claim pictures uploaded by a user need to be acquired, where the N claim pictures are respectively expressed as: image1, image2, image3, … …, imageN.
And S204, extracting the characteristic vectors of the N claim pictures by using a deep learning model.
Based on the problems of large number of the claim pictures and large occupied calculation space, N claim pictures are uniformly processed by adopting a model, so that the image processing efficiency is improved. The method is characterized in that Image clustering operation is carried out according to Image similarity among claim pictures, in the process of calculating the Image similarity, vectorization processing is needed to be carried out on the images, the process of converting the images into vectors is realized based on an open-source deep learning framework, specifically, under the deep learning framework, a deep learning model is built and trained in advance by utilizing a convolutional neural network, and feature vectors of each of Image1 to Image N are obtained by inputting the claim pictures of the Image1 to the Image N into the deep learning model.
Step S206, calculating the image similarity between the N claim pictures by using a word frequency inverse text frequency index algorithm.
And carrying out vectorization operation on the N claim pictures to obtain corresponding feature vectors, and then calculating the image similarity between the claim pictures by adopting a word frequency inverse text frequency index algorithm as the basis of the picture clustering operation.
And step S208, clustering the N claim pictures by using a label propagation algorithm.
Specifically, the clustering operation of the claim-settling pictures is performed by a clustering model, and an algorithm adopted by the clustering model in the clustering process is a label propagation algorithm, and the steps are as follows:
1) Constructing a connected graph, wherein the N claim pictures Image1 to Image N correspond to nodes in the connected graph one by one, the corresponding nodes can be marked as Image1, image2, image3, … … and Image N, and the Image similarity among the N claim pictures takes a logarithmic value as edge weight among the nodes corresponding to the claim pictures;
2) Inputting the connected graph into a clustering model, obtaining the labels of the nodes by using a label propagation algorithm, wherein the nodes corresponding to the claim pictures with higher similarity have the same labels and are gathered into the same type.
Step S210, determining suspicious item members.
And taking the identification of the claim-settling pictures output by the clustering model as a standard for dividing communities, dividing the claim-settling pictures with the same clustering identification into the same communities, determining the claim-settling pictures with community attributes as suspicious pictures, and determining that the submission of the suspicious pictures is a suspicious project member.
Step S212, obtaining the factual relationship of the suspicious item members.
After the suspicious item members are determined, whether the suspicious item members are suspicious or not can be further judged through the fact relation among the suspicious item members, wherein the fact relation comprises fund transfer relation, communication relation, address book relation, equipment relation and the like among the item members, and the suspicious item members with fund transfer transaction records, communication through the communication equipment, mutual address book contact of the other party, the behavior that multiple parties use the same login equipment and the like are recognized as the fact relation.
Step S214, determining that the item member with the actual factual relation is an untrusted item member in the suspicious item members.
After the suspicious item members with similarity of the submitted claim-settling pictures are screened out through clustering operation of the claim-settling pictures, if the suspicious item members have a fact relation at the same time, the suspicious item members are more likely to participate in cheating protection, so that the item members with the fact relation can be screened out of the suspicious item members based on the fact relation to serve as unreliable item members.
Step S216, determining the member set of the claim image submitted by the member of the untrustworthy project and provided with the same label as a suspicious group partner.
The determined unreliable project members in the project members have great suspicion of cheating, while the project members belonging to the same cheating party mostly have similarity in some aspects and are easy to identify the same label, so that the collection of the member components of the claim images with the same label submitted in the unreliable project members can be determined to be suspicious, according to the clustering result obtained before.
The embodiment of the suspicious item member identification device based on image processing provided by the application is as follows:
In the foregoing embodiments, a method for identifying suspicious item members based on image processing is provided, and accordingly, the present application also provides a suspicious item member identification device based on image processing, which is described below with reference to the accompanying drawings.
Referring to fig. 3, a schematic diagram of an embodiment of a suspicious item member identification device based on image processing is provided.
Since the apparatus embodiments are substantially similar to the method embodiments, the description is relatively simple, and reference should be made to the corresponding descriptions of the method embodiments provided above for relevant parts. The device embodiments described below are merely illustrative.
The application provides a suspicious item member identification device based on image processing, which comprises:
the case review image set acquisition module 302 is configured to acquire a case review image set composed of case review images submitted by project members participating in a project;
a feature vector extraction module 304 configured to extract feature vectors of the case review images in the case review image set;
an image similarity determination module 306 configured to determine image similarity between the case review images from the feature vectors;
A clustering module 308 configured to cluster the case review images based on the similarity weights determined by the image similarities;
a suspicious item member determination module 310 configured to determine suspicious item members from the item members based on the clustering results of the clustering.
Optionally, the feature vector extraction module 304 is specifically configured to input the case review image into a pre-trained deep learning model to perform image vectorization processing, and output a feature vector of the case review image.
Optionally, the image similarity determining module 306 is specifically configured to calculate, according to the feature vector, the image similarity between the case review images by using a word frequency inverse text frequency index algorithm.
Optionally, the image similarity determining module 306 is specifically configured to determine the image similarity between the case review images based on the vector distance between the feature vectors of the case review images.
Optionally, the clustering module 308 includes:
the communication diagram construction submodule is configured to construct a communication diagram, establish a one-to-one correspondence between the case auditing images and nodes in the communication diagram, and take the logarithmic value of the image similarity between the case auditing images as the edge weight between the nodes corresponding to the case auditing images;
And the cluster identifier output module is configured to input the connected graph into a cluster model for clustering and output the cluster identifier of the case auditing image.
Optionally, the suspicious item member determination module 310 includes:
a suspicious image determination submodule configured to determine that the case review images having the same cluster identity are suspicious images;
a suspicious item member determination submodule configured to determine that submission of the suspicious image is a suspicious item member of the item members.
Optionally, the suspicious item member identification device based on image processing further includes:
a first suspicious group partner determination module configured to determine that a set of submitters of the case review images having the same cluster identity in the suspicious image is a suspicious group partner.
Optionally, the suspicious item member identification device based on image processing further includes:
a second factual relationship acquisition module configured to acquire a factual relationship of the suspicious item members;
an untrusted item member determination module configured to determine, among the suspicious item members, that an item member having an actual factual relationship is an untrusted item member based on the factual relationship;
A second suspicious group partner determination module configured to determine that a set of members of the untrusted item members submitting the case audit image having the same cluster identity are suspicious group partners.
Optionally, the suspicious item member identification device based on image processing further includes:
a first factual relationship acquisition module configured to acquire a factual relationship of the item members;
a case review image acquisition object determination module configured to determine item members having actual factual relationships based on the factual relationships;
correspondingly, the case review image set obtaining module 302 is specifically configured to obtain case review images of the project members with actual facts, so as to form the case review image set.
Optionally, the factual relationship of the item members includes at least one of:
and fund transfer relationship, communication relationship, address book relationship and equipment relationship among the project members.
Optionally, the deep learning model is constructed based on any one of the following neural networks:
convolutional neural network, deep neural network.
Optionally, the clustering model is based on any one of the following algorithms:
Label propagation algorithm, maximum connectivity graph algorithm.
An embodiment of a computing device provided herein is as follows:
fig. 4 is a block diagram illustrating a configuration of a computing device 400 according to an embodiment of the present description. The components of the computing device 400 include, but are not limited to, a memory 410 and a processor 420. Processor 420 is coupled to memory 410 via bus 430 and database 450 is used to hold data.
Computing device 400 also includes access device 440, access device 440 enabling computing device 400 to communicate via one or more networks 460. Examples of such networks include the Public Switched Telephone Network (PSTN), a Local Area Network (LAN), a Wide Area Network (WAN), a Personal Area Network (PAN), or a combination of communication networks such as the internet. The access device 440 may include one or more of any type of network interface, wired or wireless (e.g., a Network Interface Card (NIC)), such as an IEEE802.11 Wireless Local Area Network (WLAN) wireless interface, a worldwide interoperability for microwave access (Wi-MAX) interface, an ethernet interface, a Universal Serial Bus (USB) interface, a cellular network interface, a bluetooth interface, a Near Field Communication (NFC) interface, and so forth.
In one embodiment of the present description, the above-described components of computing device 400, as well as other components not shown in FIG. 4, may also be connected to each other, such as by a bus. It should be understood that the block diagram of the computing device shown in FIG. 4 is for exemplary purposes only and is not intended to limit the scope of the present description. Those skilled in the art may add or replace other components as desired.
Computing device 400 may be any type of stationary or mobile computing device, including a mobile computer or mobile computing device (e.g., tablet, personal digital assistant, laptop, notebook, netbook, etc.), mobile phone (e.g., smart phone), wearable computing device (e.g., smart watch, smart glasses, etc.), or other type of mobile device, or a stationary computing device such as a desktop computer or PC. Computing device 400 may also be a mobile or stationary server.
The present application provides a computing device comprising a memory 410, a processor 420, and computer instructions stored on the memory and executable on the processor, the processor 420 for executing computer executable instructions to:
acquiring a case auditing image set composed of case auditing images submitted by project members participating in a project;
extracting feature vectors of the case auditing images in the case auditing image set;
determining the image similarity between the case auditing images according to the feature vector;
clustering the case review images based on the similarity weights determined by the image similarity;
And determining suspicious project members in the project members according to the clustering result of the clusters.
Optionally, the extracting the feature vector of the case review image in the case review image set includes:
and inputting the case auditing image into a pre-trained deep learning model for image vectorization processing, and outputting the feature vector of the case auditing image.
Optionally, the determining the image similarity between the case review images according to the feature vector includes:
and calculating the image similarity between the case auditing images by using a word frequency inverse text frequency index algorithm according to the feature vector.
Optionally, the determining the image similarity between the case review images according to the feature vector includes:
and determining the image similarity between the case auditing images based on the vector distance between the feature vectors of the case auditing images.
Optionally, the clustering the case review image based on the similarity weight determined by the image similarity includes:
constructing a communication graph, establishing a one-to-one correspondence between the case auditing images and nodes in the communication graph, and taking a logarithmic value of the image similarity between the case auditing images as an edge weight between the nodes corresponding to the case auditing images;
And inputting the connected graph into a clustering model for clustering, and outputting the clustering identification of the case auditing image.
Optionally, the determining suspicious item members in the item members according to the clustering result of the clustering includes:
determining the case auditing images with the same cluster identifier as suspicious images;
determining that the submission of the suspicious image is a suspicious one of the item members.
Optionally, after the step of determining suspicious item members in the item members according to the clustering result of the clustering is performed, the method includes:
and determining the set of submitters of the case auditing images with the same cluster identifier in the suspicious image as suspicious group partners.
Optionally, after the step of determining suspicious item members in the item members according to the clustering result of the clustering is performed, the method further includes:
acquiring the factual relationship of the suspicious item members;
determining that the item member with the actual factual relationship is an untrusted item member from the suspicious item members based on the factual relationship;
and determining a member set submitted with the case audit image with the same cluster identifier in the untrusted project members as a suspicious group partner.
Optionally, before the step of acquiring the case auditing image set composed of the case auditing images submitted by the project members participating in the project is executed, the method further includes:
acquiring a factual relationship of the project members;
determining item members having actual factual relationships based on the factual relationships;
correspondingly, in the case auditing image set step of acquiring case auditing images submitted by project members participating in a project, the case auditing images of the project members with actual facts are acquired to form the case auditing image set.
Optionally, the factual relationship of the item members includes at least one of:
and fund transfer relationship, communication relationship, address book relationship and equipment relationship among the project members.
Optionally, the deep learning model is constructed based on any one of the following neural networks:
convolutional neural network, deep neural network.
An embodiment of the present application also provides a computer-readable storage medium storing computer instructions that, when executed by a processor, implement:
acquiring a case auditing image set composed of case auditing images submitted by project members participating in a project;
Extracting feature vectors of the case auditing images in the case auditing image set;
determining the image similarity between the case auditing images according to the feature vector;
clustering the case review images based on the similarity weights determined by the image similarity;
and determining suspicious project members in the project members according to the clustering result of the clusters.
Optionally, the extracting the feature vector of the case review image in the case review image set includes:
and inputting the case auditing image into a pre-trained deep learning model for image vectorization processing, and outputting the feature vector of the case auditing image.
Optionally, the determining the image similarity between the case review images according to the feature vector includes:
and calculating the image similarity between the case auditing images by using a word frequency inverse text frequency index algorithm according to the feature vector.
Optionally, the determining the image similarity between the case review images according to the feature vector includes:
and determining the image similarity between the case auditing images based on the vector distance between the feature vectors of the case auditing images.
Optionally, the clustering the case review image based on the similarity weight determined by the image similarity includes:
constructing a communication graph, establishing a one-to-one correspondence between the case auditing images and nodes in the communication graph, and taking a logarithmic value of the image similarity between the case auditing images as an edge weight between the nodes corresponding to the case auditing images;
and inputting the connected graph into a clustering model for clustering, and outputting the clustering identification of the case auditing image.
Optionally, the determining suspicious item members in the item members according to the clustering result of the clustering includes:
determining the case auditing images with the same cluster identifier as suspicious images;
determining that the submission of the suspicious image is a suspicious one of the item members.
Optionally, after the step of determining suspicious item members in the item members according to the clustering result of the clustering is performed, the method includes:
and determining the set of submitters of the case auditing images with the same cluster identifier in the suspicious image as suspicious group partners.
Optionally, after the step of determining suspicious item members in the item members according to the clustering result of the clustering is performed, the method further includes:
Acquiring the factual relationship of the suspicious item members;
determining that the item member with the actual factual relationship is an untrusted item member from the suspicious item members based on the factual relationship;
and determining a member set submitted with the case audit image with the same cluster identifier in the untrusted project members as a suspicious group partner.
Optionally, before the step of acquiring the case auditing image set composed of the case auditing images submitted by the project members participating in the project is executed, the method further includes:
acquiring a factual relationship of the project members;
determining item members having actual factual relationships based on the factual relationships;
correspondingly, in the case auditing image set step of acquiring case auditing images submitted by project members participating in a project, the case auditing images of the project members with actual facts are acquired to form the case auditing image set.
Optionally, the factual relationship of the item members includes at least one of:
and fund transfer relationship, communication relationship, address book relationship and equipment relationship among the project members.
Optionally, the deep learning model is constructed based on any one of the following neural networks:
Convolutional neural network, deep neural network.
The above is an exemplary version of a computer-readable storage medium of the present embodiment. It should be noted that, the technical solution of the storage medium and the technical solution of the suspicious item member identification method based on image processing belong to the same concept, and details of the technical solution of the storage medium, which are not described in detail, can be referred to the description of the technical solution of the suspicious item member identification method based on image processing.
The computer instructions include computer program code that may be in source code form, object code form, executable file or some intermediate form, etc. The computer readable medium may include: any entity or device capable of carrying the computer program code, a recording medium, a U disk, a removable hard disk, a magnetic disk, an optical disk, a computer Memory, a Read-Only Memory (ROM), a random access Memory (RAM, random Access Memory), an electrical carrier signal, a telecommunications signal, a software distribution medium, and so forth. It should be noted that the computer readable medium contains content that can be appropriately scaled according to the requirements of jurisdictions in which such content is subject to legislation and patent practice, such as in certain jurisdictions in which such content is subject to legislation and patent practice, the computer readable medium does not include electrical carrier signals and telecommunication signals.
It should be noted that, for the sake of simplicity of description, the foregoing method embodiments are all expressed as a series of combinations of actions, but it should be understood by those skilled in the art that the present application is not limited by the order of actions described, as some steps may be performed in other order or simultaneously in accordance with the present application. Further, those skilled in the art will also appreciate that the embodiments described in the specification are all preferred embodiments, and that the acts and modules referred to are not necessarily all necessary for the present application.
In the foregoing embodiments, the descriptions of the embodiments are emphasized, and for parts of one embodiment that are not described in detail, reference may be made to the related descriptions of other embodiments.
The above-disclosed preferred embodiments of the present application are provided only as an aid to the elucidation of the present application. Alternative embodiments are not intended to be exhaustive or to limit the invention to the precise form disclosed. Obviously, many modifications and variations are possible in light of the above teaching. The embodiments were chosen and described in order to best explain the principles of the application and the practical application, to thereby enable others skilled in the art to best understand and utilize the application. This application is to be limited only by the claims and the full scope and equivalents thereof.