A passenger anomaly identification method and device based on DBSCAN

Abstract

The application discloses a passenger abnormality identification method and device based on DBSCAN, and the method comprises the following steps: establishing a data sample, performing unsupervised abnormality classification according to the data sample, and automatically calculating the similarity of the abnormal data itself; obtaining a passenger abnormality judgment result according to the similarity; wherein the unsupervised abnormality classification according to the data sample is specifically classified according to the abnormality type, the feature data is obtained, the feature data of different abnormality types is formed, the DBSCAN algorithm is used for clustering analysis on the feature data of different abnormality types, and clustering cluster data is obtained, wherein the clustering cluster data comprises an abnormal cluster center and a conventional cluster center. According to the application, the characteristics are calculated according to various personnel attributes, and the accuracy of the DBSCAN algorithm classification is improved; in the data classification process of the DBSCAN algorithm, outlier points can be distinguished, the data is not only beneficial to the analysis of new abnormal types, but also avoids that a small amount of data is forced to be divided into a certain clustering cluster, thereby affecting the classification accuracy.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of electronic information technology, and particularly relates to a passenger abnormality identification method based on a DBSCAN algorithm. BACKGROUND

[0002] Currently, there are two methods for border inspection staff to perform document screening and risk identification on passengers, one is a manual judgment method based on personal experience, and the other is an automatic judgment method realized by computer-aided establishment of an expert experience database.

[0003] The manual judgment method has the following disadvantages: (1) the risk identification of passengers by staff is subjective and cannot achieve unified standard risk judgment; (2) the risk identification experience of staff is uneven and cannot guarantee the risk passenger identification rate; and (3) the manual judgment efficiency is insufficient, which makes the customs inspection unable to maintain high efficiency.

[0004] In the automatic judgment method, experts' experience database is established to identify risk passengers according to expert rules, but the expert experience method of the expert experience database has the following defects: (1) the corresponding rules generated by the expert experience method may have some differences from the real distribution of data, resulting in missed detection or false detection; (2) when there is no corresponding identification rule for a risk passenger in the expert experience database, the risk passenger cannot be identified; and (3) when the data volume of a certain type of risk passenger is extremely small or does not exist, the machine learning method cannot learn and capture the relationship between the data. For such extreme cases, the single use of the machine learning method has a certain missed detection rate. SUMMARY

[0005] In the following, a brief summary of embodiments of the present application is given in order to provide a basic understanding of some aspects of the present application. It should be understood that the following summary is not an exhaustive overview of the present application. It is not intended to identify key or important parts of the present application nor is it intended to limit the scope of the present application. Its purpose is merely to present some concepts in a simplified form as a prelude to a more detailed description to be discussed later.

[0006] According to one aspect of the present application, a passenger abnormality identification method based on DBSCAN is provided, which comprises: establishing a data sample, performing unsupervised anomaly classification according to the data sample, and automatically calculating the similarity of the abnormal data itself; obtaining a passenger abnormality judgment result according to the similarity; wherein the unsupervised anomaly classification according to the data sample is specifically classifying the data sample according to the abnormal type and obtaining the feature data thereof, forming feature data of different abnormal types, and respectively performing clustering analysis on the feature data of different abnormal types by using the DBSCAN algorithm to obtain clustering cluster data, wherein the clustering cluster data comprises an abnormal cluster center and a normal cluster center.

[0007] Further, the data sample includes passenger portrait data.

[0008] Further, unsupervised anomaly classification is performed according to the data sample, and specifically includes:

[0009] Step 1: classifying the data sample according to the anomaly type to form passenger information of different anomaly types, the anomaly type can be a first type of abnormal personnel, a second type of abnormal personnel, or a third type of abnormal personnel;

[0010] Step 2: defining attributes of passenger information of various anomaly types respectively; calculating features according to the attribute definition of passenger information of various anomaly types to form various anomaly type feature data (each type of passenger information corresponds to one anomaly type feature data);

[0011] The calculation of features according to the attribute definition of passenger information of various anomaly types includes first type of abnormal personnel feature calculation, second type of abnormal personnel feature calculation and third type of abnormal personnel feature calculation, wherein the first type of abnormal personnel feature includes passport type, age range, overseas travel trajectory, overseas stay time, entry and exit frequency, illegal record and sensitive time period travel times, and the first type of abnormal personnel feature calculation includes the step of data normalization of the first type of abnormal personnel feature; the second type of abnormal personnel feature includes passenger nationality, age range, certificate type, place of origin and destination, entry and exit times, entry and exit interval, accommodation place and illegal record, and the second type of abnormal personnel feature calculation includes the step of data normalization of the second type of abnormal personnel feature; the third type of abnormal personnel feature includes passenger nationality, gender, visa stay period, personnel category, stay area in China, visa category, illegal record and whether applying for a family reunion visa, and the third type of abnormal personnel feature calculation includes the step of data normalization of the third type of abnormal personnel feature;

[0012] Step 3: separately performing DBSCAN clustering calculation on each anomaly type feature data;

[0013] Step 31: determining the neighborhood radius Eps and the number threshold MinPts of data objects in the neighborhood in the DBSCAN algorithm:

[0014] ① The neighborhood radius Eps is obtained by drawing a k-distance curve, and the parameter corresponding to the obvious inflection point position of the k-distance curve is selected; in the k-distance curve, given the K neighborhood parameter k, for each point in the data, the corresponding kth nearest neighbor distance is calculated, and all points in the data set are sorted in descending order according to the nearest neighbor distance, which is called a sorted k-distance graph, and the k-distance value corresponding to the first valley point position in the k-distance graph is selected as Eps. Generally, the value of k is set to 4. The k-distance curve is prior art, which will not be described here;

[0015] 2. The number of data objects in the neighborhood threshold MinPts, which is selected to satisfy: MinPts ≥ dim + 1, where dim represents the dimension of the data to be clustered; generally, MinPts is selected to be greater than or equal to 3;

[0016] Step 32: DBSCAN clustering calculation process:

[0017] Input: data set composed of abnormal type feature data, neighborhood radius Eps, and the number of data objects in the neighborhood threshold MinPts;

[0018] Output: density-connected cluster;

[0019] The processing flow is as follows:

[0020] 1. Select an arbitrary data object point p from the data set;

[0021] 2. If the selected data object point p is a core point for the parameters Eps and MinPts, find all the density-reachable data object points from p to form a cluster;

[0022] 3. If the selected data object point p is an edge point, select another data object point;

[0023] 4. Repeat steps 2 and 3 until all points in the data set are processed;

[0024] 5. Obtain the clustering cluster c i and the object points {p i1 , p i2 ,..., p ik} in the cluster, outliers {o1, o2,..., o j}, i is the number of clustering clusters, k is the number of object points in different clustering clusters, and j is the number of outliers.

[0025] Step 4: According to step 3, obtain the clustering clusters of different abnormal type feature data, and obtain the clustering cluster centers through the clustering clusters, denoted as clustering cluster c1, clustering cluster c2, and clustering cluster c3. The object points in the clustering cluster c1 are {p 11 , p 12 ,..., p 1m}, the object points in the clustering cluster c2 are {p 21 , p 22 ,..., p 2n}, the object points in the clustering cluster c3 are {p 31 , p 32 ,..., p 3s}, and the outliers are {o1, o2,..., o j}. Let the cluster {p 11 , p 12 ,..., p1m} the gravity point p1 as the cluster center of the cluster c1, the cluster {p 21 , p 22 ,..., p 2n} the gravity point p2 as the cluster center of the cluster c2, the cluster {p 31 , p 32 ,..., p 3s} the gravity point p3 as the cluster center of the cluster c3; the definition of outliers is that all object points not divided into any cluster are defined as outliers, so the outliers are related to the selection of Eps and MinPts and the data distribution, and different combinations of Eps and MinPts get different numbers of outliers, so outliers may exist or may not exist. Outliers are outliers for the entire data set, not outliers for a certain cluster.

[0026] Step 5: combining experience and cluster center feature distribution of the cluster c1, the cluster c2 and the cluster c3, manually determine whether each cluster is abnormal or not, and determine the label attribute of the object points in the cluster;

[0027] Step 6: analyze the outlier distribution to determine whether there is an anomaly, if there is an anomaly, take this outlier distribution as an anomaly class cluster center d i ;

[0028] Step 7: if there is expert experience for a certain type of anomaly, supplement the possible anomaly distribution with the expert experience, and divide the feature part as an expert identified anomaly class cluster center d j ;

[0029] Step 8: get the normal cluster center point set {p1, p2,..., p n} and the abnormal cluster center point set {d n+1 , d n+2 ,..., d n+m};

[0030] Step 9: according to the above steps, the normal cluster center point set {p1, p2,..., p n1} and the abnormal cluster center point set {d n1+1 , d n1+2 ,..., d n1+m1} of the first type of abnormal personnel, the normal cluster center point set {q1, q2,..., q n2} and the abnormal cluster center point set {d n2+1 , d n2+2 ,..., d n2+m2} of the second type of abnormal personnel, and the normal cluster center point set {r1, r2,..., r n3} and an abnormal clustering center point set {d n3+1 , d n3+2 ,..., d n3+m3}.

[0031] Specifically, in step 31, different neighborhood radius Eps and MinPts have a great influence on the clustering result, so when selecting Eps and MinPts, both the k-distance valley point and the consistency of the clustering result with the data distribution (whether it can accurately divide normal data and abnormal data) should be considered; the specific steps are as follows:

[0032] 1. Before clustering the samples, manually label a certain number of normal personnel data and abnormal personnel data to form marker points (for example, before clustering the first type of abnormal personnel, 3 first type of abnormal personnel and 3 normal personnel are determined in the current data according to expert experience);

[0033] 2. Perform feature calculation on different personnel data to obtain feature data;

[0034] 3. Perform k-distance calculation on the feature data and draw a k-distance curve to determine the k-distance value corresponding to the position of different valley points;

[0035] 4. Take the k-distance of different valley points as the Eps parameter of DBSCAN algorithm in turn to obtain the Eps parameter list [Eps1, Eps2, …, Epsn-1, Epsn];

[0036] 5. According to MinPts ≥ dim+1, determine the candidate list [MinPts1, MinPts2, …, MinPtsm-1, MinPtsm] of DBSCAN parameter MinPts;

[0037] 6. Perform grid calculation on [Eps1, Eps2, …, Epsn-1, Epsn] and [MinPts1, MinPts2, …, MinPtsm-1, MinPtsm] to obtain different clustering clusters;

[0038] 7. Cluster the marker points of the first step, if different personnel types belong to different clustering clusters (for example, the first type of abnormal personnel belongs to clustering clusters c1 and c2, and the normal personnel belongs to c3), it meets the requirements, and the corresponding Eps and MinPts are retained;

[0039] 8. If there are multiple parameter candidate items, first select the largest Eps candidate item Eps-i, if there are multiple MinPts corresponding to Eps-i in the candidate items, select the smallest MinPts candidate item MinPts-j, because the larger Eps and the smaller MinPts, the fewer DBSCAN outliers.

[0040] Further, the automatic calculation of the similarity of the abnormal data itself, according to the similarity of the passenger abnormal judgment result, specifically includes:

[0041] For the new passenger of the border inspection, the similarity of the first type of abnormal personnel feature distribution z, the second type of abnormal personnel feature distribution q, the third type of abnormal personnel r and the corresponding cluster center point set is calculated, and whether it is abnormal and the type of abnormality is judged according to the similarity:

[0042] (1) Calculate the similarity of z and {p1, p2,..., p n1}、{d n1+1 , d n1+2 ,..., d n1+m1}, the highest similarity point corresponds to the label, that is, the label attribute corresponding to whether the passenger is the first type of abnormal personnel, such as p n1+2 The highest similarity, and the label corresponding to p n1+2 is abnormal, then the passenger belongs to the first type of abnormal personnel;

[0043] (2) Calculate the similarity of q and {q1, q2,..., q n2}、{d n2+1 , d n2+2 ,..., d n2+m2}, the highest similarity point corresponds to the label, that is, the label attribute corresponding to whether the passenger is the first type of abnormal personnel, such as q n2+2 The highest similarity, and the label corresponding to q n2+2 is abnormal, then the passenger belongs to the second type of abnormal personnel;

[0044] (3) Calculate the similarity of r and {r1, r2,..., r n3}、{d n3+1 , d n3+2 ,..., d n3+m3}, the highest similarity point corresponds to the label, that is, the label attribute corresponding to whether the passenger is the first type of abnormal personnel, such as r n3+2 The highest similarity, and the label corresponding to r n3+2 is abnormal, then the passenger belongs to the third type of abnormal personnel.

[0045] Among them, as a feasible scheme, the similarity of the first type of abnormal personnel feature distribution z, the second type of abnormal personnel feature distribution q, the third type of abnormal personnel r and the corresponding cluster center point set is calculated, and the similarity calculation formula is:

[0046] In specific application, for the outbound or inbound passenger, the abnormal attribute can be calculated in advance and stored in the database, and the personnel abnormal warning prompt is given to the on-site staff, and the passenger is screened and questioned.

[0047] According to another aspect of the present application, a DBSCAN-based passenger anomaly identification device is provided, comprising:

[0048] A first module for establishing data samples;

[0049] A second module for unsupervised anomaly classification according to the data samples;

[0050] A third module for automatically calculating the similarity of the anomaly data itself;

[0051] A fourth module for obtaining a passenger anomaly judgment result according to the similarity;

[0052] In the second module, unsupervised anomaly classification is performed according to the data samples, specifically, the data samples are classified according to the anomaly types and the feature data thereof is obtained, the feature data of different anomaly types is formed, and the feature data of different anomaly types is respectively subjected to clustering analysis by using the DBSCAN algorithm to obtain clustering cluster data, wherein the clustering cluster data includes anomaly cluster centers and normal cluster centers.

[0053] The method of the present application can be directly applied in the field of border inspection and passenger data analysis; the method combines machine learning and manual definition to identify anomalies of border inspection and passenger data. Compared with the prior art, the present application has the following advantages:

[0054] 1. The feature is calculated according to the attributes of various personnel, which is beneficial to the accuracy of DBSCAN algorithm classification;

[0055] 2. DBSCAN algorithm can distinguish outlier points during data classification, which not only helps to analyze new anomaly types, but also avoids a small amount of data being forced to be classified into a certain cluster, thereby affecting the classification accuracy;

[0056] 3. For a newly opened port, the amount of historical data may not include all types of data, and this patent supports manual definition of the distribution of a certain type of anomaly feature, which can reduce the missed detection to a certain extent. BRIEF DESCRIPTION OF DRAWINGS

[0057] The present application can be better understood by referring to the description given below in conjunction with the accompanying drawings, in which the same or similar reference numerals are used to indicate the same or similar components throughout the drawings. The drawings, together with the following detailed description, are included in the specification and form a part of the specification, and are used to further illustrate the preferred embodiments of the present application and explain the principles and advantages of the present application. In the drawings,

[0058] In the drawings:

[0059] Figure 1 is a schematic diagram of the principle of the embodiment of the present application. Detailed Implementation

[0060] Embodiments of the present invention will now be described with reference to the accompanying drawings. Elements and features described in one drawing or embodiment of the invention may be combined with elements and features shown in one or more other drawings or embodiments. It should be noted that, for clarity, representations and descriptions of components and processes unrelated to the present invention and known to those skilled in the art have been omitted from the drawings and description.

[0061] Currently, port staff rely on their own experience to subjectively judge passenger anomalies, which may lead to missed or incorrect detections. This patent provides an objective calculation method for passenger anomaly judgment by performing unsupervised anomaly classification based on passenger profile data and automatically calculating the similarity of the anomaly data itself.

[0062] As a specific embodiment, the DBSCAN-based passenger anomaly identification method of the present invention is implemented as follows:

[0063] Step 1: Select n regular passengers, where n is greater than or equal to 1000;

[0064] Step 2: Calculate different anomaly types: Category 1 anomaly personnel, Category 2 anomaly personnel, and Category 3 anomaly personnel;

[0065] (1) Calculation of characteristics of the first type of abnormal persons: passport type, age group, overseas travel trajectory, length of stay in various overseas places, frequency of entry and exit, criminal record, number of trips during sensitive time periods, and data normalization;

[0066] (2) Calculation of characteristics of the second category of abnormal persons: nationality, age group, document type, place of origin and destination, number of entries and exits, entry and exit intervals, accommodation location, and records of violations and irregularities, and data normalization is performed;

[0067] (3) Calculation of characteristics of the third category of abnormal persons: nationality, gender, visa stay period, personnel category, area of ​​stay in China, visa category, number of violations, whether they have applied for family reunion visas, and data normalization.

[0068] Step 3: Perform DBSCAN clustering calculations on the feature data of each anomaly type separately;

[0069] DBSCAN (Density-Based Spatial Clustering of Application with Noise) algorithm is a typical density-based clustering method, which defines the cluster as the maximum set of density-connected points, and can divide the area with sufficient density into clusters, and can find clusters of arbitrary shape in noisy spatial data sets. DBSCAN algorithm has two important parameters: Eps and MinPts. Eps is the neighborhood radius when defining density, and MinPts is the threshold when defining core points.

[0070] (1) DBSCAN parameter selection process:

[0071] ① The value of neighborhood radius Eps can be obtained by drawing a k-distance graph method, and the obvious inflection point position of the k-distance graph corresponds to a better parameter. If the parameter is set too small, most of the data cannot be clustered; if the parameter is set too large, multiple clusters and most of the objects will be merged into the same cluster.

[0072] K-distance: The definition of K-distance is given in detail in the original text of DBSCAN algorithm. Given the K-neighborhood parameter k, for each point in the data, calculate the corresponding kth nearest neighbor distance, and sort all points in the data set according to the descending order of the nearest neighbor distance, called the sorted k-distance graph. The k-distance value corresponding to the first valley point position in the graph is set as Eps. Generally, the value of k is set to 4.

[0073] ② The selection of the number of data objects in the neighborhood threshold MinPts has a guiding principle (a rule of thumb), MinPts ≥ dim + 1, where dim represents the dimension of the data to be clustered. It is unreasonable to set MinPts to 1, because if it is set to 1, each independent point is a cluster, and if MinPts ≤ 2, the result of the nearest neighbor of the hierarchical distance is the same. Therefore, MinPts must be selected to be greater than or equal to 3. If the value is too small, the results of sparse clusters will be considered as boundary points due to the density less than MinPts, and will not be used for further expansion of the class; if the value is too large, two adjacent clusters with high density may be merged into the same cluster. Therefore, whether the value is set appropriately will have a great impact on the clustering results.

[0074] (2) DBSCAN clustering calculation process:

[0075] Input: data set, neighborhood radius Eps, number of data objects in neighborhood threshold MinPts;

[0076] Output: density-connected cluster.

[0077] The processing flow is as follows:

[0078] ① randomly select a data object point p from the data set;

[0079] ② if the selected data object point p is a core point for parameters Eps and MinPts, find all data object points density-reachable from p to form a cluster;

[0080] ③ if the selected data object point p is an edge point, select another data object point;

[0081] ④ repeat steps ② and ③ until all points are processed;

[0082] ⑤ obtain the clustering cluster c i and the object points {p i1 , p i2 ,... p ik} in the cluster, the outlier points {o1, o2,... o j}, i is equal to the number of clustering clusters, k is equal to the number of object points in different clustering clusters, and j is equal to the number of outlier points;

[0083] Step 4, according to step 3, the characteristics of the first type of abnormal personnel, the second type of abnormal personnel, and the clustering cluster of the third type of abnormal personnel can be obtained. The clustering cluster center can be obtained through the clustering clusters. For example, the first type of abnormal personnel clustering cluster c1, the second type of abnormal personnel clustering cluster c2, and the third type of abnormal personnel clustering cluster c3, the object points in the clustering cluster c1 are {p 11 , p 12 ,... p 1m}, the outlier points {o1, o2,... o j}, the object points in the clustering cluster c2 are {p 21 , p 22 ,... p 2n}, the object points in the clustering cluster c3 are {p 31 , p 32 ,... p 3s}, the barycenter point p1 of the cluster {p 11 , p 12 ,... p 1m} is taken as the clustering center of the clustering cluster c1, the barycenter point p2 of the cluster {p 21 , p 22 ,... p 2n} is taken as the clustering center of the clustering cluster c2, and the barycenter point p3 of the cluster {p 31 , p 32 ,... p 3s} is taken as the clustering center of the clustering cluster c3;

[0084] Step 5, combine the cluster center feature distribution of cluster c1, cluster c2 and cluster c3, and determine whether each cluster is abnormal by manual judgment, and determine the label attribute of the object points in the cluster;

[0085] Step 6, since the outlier does not belong to any cluster, and the distribution is discrete, there is a certain probability of abnormality, and the analysis of the outlier distribution can determine the type of irregular anomaly, such as the existence of an abnormal outlier distribution as an abnormal cluster center d i ;

[0086] Step 7, the historical data may not contain all possible abnormal distributions, at this time, according to the expert experience, the possible abnormal distribution is supplemented to meet the comprehensiveness of subsequent anomaly detection. If there is expert experience for a certain type of anomaly, the feature part can be manually divided as an expert identified anomaly cluster center d j ;

[0087] Step 8, obtain the normal cluster center point set {p1, p2,..., p n} and the abnormal cluster center point set {d n+1 , d n+2 ,..., d n+m};

[0088] Step 9, according to the above steps, the normal cluster center point set {p1, p2,..., p n1} and the abnormal cluster center point set {d n1+1 , d n1+2 ,..., d n1+m1} of the first type of abnormal personnel, the normal cluster center point set {q1, q2,..., q n2} and the abnormal cluster center point set {d n2+1 , d n2+2 ,..., d n2+m2} of the second type of abnormal personnel, the normal cluster center point set {r1, r2,..., r n3} and the abnormal cluster center point set {d n3+1 , d n3+2 ,..., d n3+m3} of the third type of abnormal personnel are obtained.

[0089] Step 9, for the new passengers at the border port, the similarity of the first type of abnormal personnel feature distribution z, the second type of abnormal personnel feature distribution q, and the third type of abnormal personnel r to the corresponding type of normal cluster center point set and abnormal cluster center point set is calculated, so as to determine whether it is an abnormal personnel and which type of abnormal personnel it belongs to:

[0090] (1) Calculate z and {p1, p2,..., p n1}, {dn1+1 , d n1+2 ,..., d n1+m1}, the point with the highest similarity corresponds to the label, and the label attribute corresponding to whether the passenger is the first type of abnormal personnel, such as p and p n1+2 with the highest similarity, and p ni+2 corresponding to the label is abnormal, then the passenger belongs to the first type of abnormal personnel;

[0091] (2) Calculate the similarity of q and {q1, q2,..., q n2}, {d n2+1 , d n2+2 ,..., d n2+m2}, the point with the highest similarity corresponds to the label, and the label attribute corresponding to whether the passenger is the second type of abnormal personnel, such as q and q n2+2 with the highest similarity, and q n2+2 corresponding to the label is abnormal, then the passenger belongs to the second type of abnormal personnel;

[0092] (3) Calculate the similarity of r and {r1, r2,..., r n3}, {d n3+1 , d n3+2 ,..., d n3+n3}, the point with the highest similarity corresponds to the label, and the label attribute corresponding to whether the passenger is the third type of abnormal personnel, such as r and r n3+2 with the highest similarity, and r n3+2 corresponding to the label is abnormal, then the passenger belongs to the third type of abnormal personnel.

[0093] (4) In the embodiment, the calculation formula for calculating the similarity is:

[0094] Step 10, for the predicted outbound or inbound passenger, the abnormal attribute is calculated in advance and stored in the database, and the personnel abnormal early warning prompt is given to the on-site staff, and the passenger is screened and questioned when passing through the customs.

[0095] The method of the application can be directly applied to the field of research and judgment of border inspection passenger data; the method combines machine learning method and artificial definition, and performs abnormal discrimination on the border inspection passenger.

[0096] In addition, the embodiment of the application also provides a passenger abnormality identification device based on DBSCAN, which comprises:

[0097] The first module is used to establish a data sample;

[0098] The second module is used for unsupervised anomaly classification according to the data sample;

[0099] The third module is used for automatically calculating the similarity of the abnormal data itself.

[0100] a fourth module configured to obtain a passenger abnormality judgment result according to the similarity;

[0101] In the second module, unsupervised abnormality classification is performed according to the data samples, specifically, the data samples are classified according to abnormality types and feature data thereof is obtained, forming feature data of different abnormality types, and the feature data of different abnormality types is respectively subjected to clustering analysis by using a DBSCAN algorithm to obtain clustering cluster data, the clustering cluster data including abnormal cluster centers and normal cluster centers.

[0102] The first module, the second module, the third module and the fourth module all execute corresponding steps of the passenger abnormality recognition method based on DBSCAN.

[0103] The method of the present application is not limited to being executed in the time sequence described in the specification, but can also be executed in other time sequences, in parallel or independently. Therefore, the execution sequence of the method described in the specification does not constitute a limitation on the technical scope of the present application.

[0104] Although the present application has been disclosed above by describing specific embodiments of the present application, it should be understood that all the embodiments and examples described above are exemplary and not limiting. Those skilled in the art can design various modifications, improvements or equivalents of the present application within the spirit and scope of the appended claims. These modifications, improvements or equivalents should also be considered to be included in the protection scope of the present application.

Claims

1. A DBSCAN-based passenger anomaly identification method, characterized in that: Comprise: Establish a data sample, perform unsupervised anomaly classification according to the data sample, and automatically calculate the similarity of the abnormal data itself; According to the similarity, the passenger abnormal judgment result is obtained; Among them, according to the data sample, unsupervised anomaly classification is performed, specifically, the data sample is classified according to the abnormal type and the feature data thereof is obtained, forming feature data of different abnormal types, and the feature data of different abnormal types is respectively clustered and analyzed by using DBSCAN algorithm, to obtain cluster data, the cluster data includes abnormal cluster center and normal cluster center; According to the data sample, unsupervised anomaly classification is performed, specifically including: Step 1: classify the data sample according to the abnormal type to form passenger information of different abnormal types, Step 2: define the attributes of passenger information of various abnormal types respectively; calculate the features according to the attribute definition of passenger information of various abnormal types respectively to form various abnormal type feature data; Step 3: DBSCAN clustering calculation is performed on each abnormal type feature data separately: Step 31: determine the neighborhood radius Eps and the data object number threshold MinPts parameters in DBSCAN algorithm: Step 32: DBSCAN clustering calculation process: Input: data set composed of abnormal type feature data, neighborhood radius Eps, and data object number threshold MinPts; Output: density connected cluster; The processing procedure is as follows: Select an arbitrary data object point p from the data set; If the selected data object point p is a core point for parameters Eps and MinPts, find all data object points density-reachable from p, forming a cluster; If the selected data object point p is an edge point, another data object point is selected; repeat , steps until all points in the dataset have been processed; obtaining a cluster and object points within the cluster outliers i is the number of cluster, k is the number of object points in different cluster, j is the number of outliers Step 4: According to the different abnormal type feature data of step 3, the clustering cluster center is obtained through each type of clustering cluster, and is recorded as clustering cluster , clustering cluster , clustering cluster , clustering cluster , the object point in the clustering cluster , the object point in the clustering cluster , the object point in the clustering cluster , the object point in the clustering cluster , the outlier point , the gravity point of cluster is taken as the clustering center of the clustering cluster , the gravity point of cluster is taken as the clustering center of the clustering cluster , the gravity point of cluster is taken as the clustering center of the clustering cluster , the gravity point of cluster is taken as the clustering center of the clustering cluster , the gravity point of cluster is taken as the clustering center of the clustering cluster ; Step 5: Combining experience and clustering clusters , clustering clusters , clustering clusters The clustering center feature distribution of the clustering clusters, manually determine whether each clustering cluster is abnormal or not, and determine the label attribute of the object points in the cluster. Step 6: Analyze outlier distribution to determine if there are anomalies, if so, cluster the outlier distribution as an anomaly class cluster center ; Step 7: If there is expert experience for a certain anomaly type, manually partition the feature part as the expert identified anomaly cluster center ; Step 8: Obtain normal cluster center point set and abnormal cluster center point set ; Step 9: Obtain the normal clustering center point set of the first type of abnormal personnel according to the above steps and the abnormal clustering center point set , the normal clustering center point set of the second type of abnormal personnel and the abnormal clustering center point set , the normal clustering center point set of the third type of abnormal personnel and the abnormal clustering center point set . 2.The DBSCAN-based passenger anomaly identification method of claim 1, wherein: The data sample includes passenger portrait data.

3. The DBSCAN-based passenger abnormality identification method according to claim 1, characterized in that: In step 31, when selecting Eps and MinPts, both the k-distance valley point and the consistency of clustering results and data distribution are considered; the specific steps are as follows: Before clustering the sample, manually label several normal personnel data and abnormal personnel data to form marker points; Calculate the features of different personnel data to obtain feature data; Calculate the k-distance of the feature data and draw a k-distance curve to determine the k-distance value corresponding to different valley points; In turn, use different valley point k-distance as DBSCAN algorithm Eps parameter to obtain Eps parameter list [Eps1, Eps2, …, Epsn-1, Epsn]; DBSCAN parameter MinPts is determined according to MinPts≥dim+1 to form candidate list [MinPts1, MinPts2, …, MinPtsm-1, MinPtsm]; Grid calculation is performed on [Eps1, Eps2, …, Epsn-1, Epsn] and [MinPts1, MinPts2, …, MinPtsm-1, MinPtsm] to obtain different clustering clusters; Cluster the marker points of the first step, if different personnel types belong to different clustering clusters, it meets the requirements, and the corresponding Eps and MinPts are retained; If there are multiple sets of parameter candidates, the Eps maximum candidate Eps-i is selected first, and if there are multiple sets of MinPts corresponding to Eps-i in the candidates, the MinPts minimum candidate MinPts-j is selected, because the larger the Eps and the smaller the MinPts, the fewer the DBSCAN outliers. 4.The DBSCAN-based passenger anomaly identification method of claim 3, wherein: In the step 1, the abnormal type is a first abnormal person, a second abnormal person, or a third abnormal person. 5.The DBSCAN-based passenger abnormality identification method of claim 4, wherein: In the step 2, the feature calculation according to the attribute definition of the passenger information of various abnormal types includes first abnormal person feature calculation, second abnormal person feature calculation, and third abnormal person feature calculation, wherein the first abnormal person features include passport type, age range, overseas travel trajectory, overseas stay time, entry and exit frequency, illegal record, and sensitive time period travel times, the first abnormal person feature calculation includes the step of data normalization of the first abnormal person features; the second abnormal person features include passenger nationality, age range, certificate type, place of origin and destination, entry and exit times, entry and exit interval, accommodation location, and illegal record, the second abnormal person feature calculation includes the step of data normalization of the second abnormal person features; the third abnormal person features include passenger nationality, gender, visa stay period, personnel category, stay area in China, visa category, illegal record, and whether applying for a family reunion visa, and the third abnormal person feature calculation includes the step of data normalization of the third abnormal person features. 6.The DBSCAN-based passenger abnormality identification method of claim 5, wherein: The automatic calculation of the similarity of the abnormal data itself, and the passenger abnormality judgment result obtained according to the similarity, specifically: for the new passenger of the border inspection, the similarity of the first abnormal person feature distribution z, the second abnormal person feature distribution q, the third abnormal person feature distribution r, and the corresponding cluster center point set is calculated, and whether it is abnormal is judged according to the calculated similarity: (1) calculate the similarity of p and , , the point with the highest similarity corresponds to the label, i.e. whether the passenger is the first type of abnormal personnel corresponding to the label attribute, such as p and The highest similarity, and The label corresponds to an abnormality, and the passenger belongs to the first type of abnormal personnel; (2) calculate the similarity of q and , , the point with the highest similarity corresponds to the label, i.e. whether the passenger is the first type of abnormal personnel corresponds to the label attribute, such as q and have the highest similarity, and corresponds to the label belongs to the abnormal, then the passenger belongs to the second type of abnormal personnel; (3) Calculate the similarity of r and , The point with the highest similarity corresponds to the label attribute of whether the passenger is the first type of abnormal personnel, such as r and The highest similarity, and The corresponding label belongs to the abnormal, and the passenger belongs to the third type of abnormal personnel. 7.The DBSCAN-based passenger anomaly identification method of claim 6, wherein: The similarity calculation of the first abnormal person feature distribution z, the second abnormal person feature distribution q, and the third abnormal person feature distribution r and the corresponding cluster center point set is calculated by using the following similarity calculation formula: 。 8.A DBSCAN-based passenger anomaly identification device, characterized by comprising: Including: The first module is used to establish a data sample; The second module is used to perform unsupervised anomaly classification according to the data sample; The third module is used to automatically calculate the similarity of the abnormal data itself; The fourth module is used to obtain the passenger abnormality judgment result according to the similarity; In the second module, the data sample is classified according to the abnormal type and the feature data is obtained, The feature data of different abnormal types is obtained, and the DBSCAN algorithm is used for cluster analysis on the feature data of different abnormal types to obtain cluster data, wherein the cluster data includes abnormal cluster centers and normal cluster centers; The passenger abnormality recognition device based on DBSCAN executes the steps of the passenger abnormality recognition method based on DBSCAN in any one of claims 1 to 7.

Images