Confidential carrier trajectory tracking and anomaly detection method and system based on deep learning

Abstract

The present invention provides a method and system for tracking and detecting a trajectory of a confidential carrier based on deep learning, and relates to the technical field of trajectory tracking. The method comprises: obtaining multi-source heterogeneous data during the execution of a task by a target confidential carrier; combining an adaptive weighting strategy to assign corresponding data weights to the multi-source heterogeneous data, and fusing the data weights with the multi-source heterogeneous data to obtain carrier trajectory data; obtaining optimized carrier trajectory information based on a hybrid filtering algorithm combining an unscented Kalman filter and a particle filter and combining a hybrid weight allocation strategy; calculating the local reachable density and local anomaly factor of each trajectory data point in the optimized carrier trajectory information under different neighbor values; combining the local anomaly factor values ​​under multiple scales to obtain a multi-scale anomaly score of the trajectory data point; and judging, according to a preset anomaly judgment threshold, a trajectory data point whose multi-scale anomaly score exceeds the anomaly judgment threshold as an anomaly point.

Description

Technical Field

[0001] The present invention relates to the field of trajectory tracking technology, and in particular to a method and system for tracking and detecting secret-related carrier trajectories and anomalies based on deep learning. Background Art

[0002] Since confidential carriers carry important secret information, in order to prevent them from being illegally obtained or leaked, it is particularly important to conduct accurate trajectory tracking and sensitive anomaly detection on confidential carriers.

[0003] Traditional methods for tracking and detecting abnormal trajectories of confidential carriers mainly include rule-based methods and statistics-based methods. The rule-based method sets predefined abnormal behavior rules and uses pattern matching and other technologies to identify abnormal trajectories. However, this method is difficult to adapt to complex and changeable abnormal behavior patterns and is prone to high missed detection rates and false alarm rates. The statistical-based method constructs a statistical model of normal trajectories and uses anomaly detection algorithms to identify abnormal trajectories that deviate from the normal model. However, this method has limited modeling capabilities for abnormal behaviors and is difficult to effectively characterize complex abnormal patterns under high-dimensional spatiotemporal features. Therefore, there is an urgent need for a method for tracking and detecting confidential carrier trajectories that can achieve accurate tracking and sensitive detection. Summary of the invention

[0004] In view of this, an embodiment of the present invention provides a method and system for tracking and detecting secret carrier trajectories and anomalies based on deep learning to eliminate or improve the problems existing in the prior art.

[0005] One aspect of the present invention provides a method for tracking and detecting anomalies of a confidential carrier based on deep learning, the method comprising the following steps:

[0006] Acquire multi-source heterogeneous data of the target confidential carrier during the execution of the mission, wherein the multi-source heterogeneous data includes GPS data, inertial navigation data and visual tracking data corresponding to the target confidential carrier; assign corresponding data weights to the multi-source heterogeneous data according to the data quality and prediction performance of the multi-source heterogeneous data in combination with an adaptive weighting strategy, and fuse the data weights with the multi-source heterogeneous data to obtain carrier trajectory data;

[0007] Based on the hybrid filtering algorithm combining unscented Kalman filtering and particle filtering, the fused carrier trajectory data is initially smoothed by unscented Kalman filtering, and the initial filtering results are secondary filtered by particle filtering, and the optimized carrier trajectory information is obtained by combining the hybrid weight allocation strategy;

[0008] For each trajectory data point in the optimized carrier trajectory information, a dynamic time warping distance matrix between it and other trajectory data points is calculated, and according to the dynamic time warping distance matrix, the local reachable density and local anomaly factor of the trajectory data point under different neighbor values ​​are calculated; the local anomaly factor values ​​under multiple scales are integrated to obtain a multi-scale anomaly score of the trajectory data point; according to a preset anomaly judgment threshold, the trajectory data point whose multi-scale anomaly score exceeds the anomaly judgment threshold is determined as an anomaly point;

[0009] Wherein, for each trajectory data point in the optimized carrier trajectory information, a dynamic time warping distance matrix between it and other trajectory data points is calculated, including: converting each trajectory data point into a time series, calculating the sequence similarity between the two time series by optimal nonlinear alignment of the time series corresponding to the two trajectory data points, and aggregating multiple sequence similarities into a dynamic time warping distance matrix;

[0010] The calculating the local reachable density of the trajectory data point under different neighbor values ​​includes:

[0011] Measuring the reciprocal of the average reachable distance between the trajectory data point and its neighboring data points, where the reciprocal of the average reachable distance reflects the local reachability density of the area where the trajectory data point is located;

[0012] Adaptively determine the neighbor values ​​corresponding to the trajectory data points, and use the multi-scale local anomaly factor method to calculate the local anomaly factor values ​​under different neighbor values.

[0013] In some embodiments of the present invention, based on a hybrid filtering algorithm combining unscented Kalman filtering and particle filtering, the fused carrier trajectory data is initially smoothed by unscented Kalman filtering, and the initial filtering result is secondary filtered by particle filtering, and the optimized carrier trajectory information is obtained by combining a hybrid weight allocation strategy, including:

[0014] Perform unscented Kalman filtering on the fused carrier trajectory data, establish the state space model of the target confidential carrier, determine the initial state variables and initial observation variables, construct the state transfer equation and observation equation according to the dynamic characteristics of the target confidential carrier, initialize the parameters of the unscented Kalman filter, recursively execute the prediction and update steps of the unscented Kalman filter, determine the target state variables of the target confidential carrier based on the minimum mean square error criterion, and update the state covariance matrix to achieve preliminary smoothing of the carrier trajectory data and obtain the initial filtering result;

[0015] Initialize the particle set of the particle filter according to the preliminary filtering result, predict the next state and observation value of each particle according to the state transfer equation and the observation equation, calculate the particle weight of each particle by using the likelihood function or the probability density function in combination with the actual observation value of each particle, resample the particle set according to the particle weight, iteratively update the particle set, determine the state distribution of the target confidential carrier by the Monte Carlo simulation method, and obtain the particle filtering result;

[0016] The initial filtering result and the particle filtering result are subjected to a mixed weight allocation strategy, wherein a first mixed weight is allocated to the initial filtering result and a second mixed weight is allocated to the particle filtering result, and weighted fusion is performed to obtain optimized carrier trajectory information.

[0017] In some embodiments of the present invention, the method further comprises:

[0018] In the update phase of the unscented Kalman filter and the particle filter, a filter residual and a particle residual are calculated, wherein the filter residual is the difference between a first actual observation value and a first predicted observation value of the unscented Kalman filter, and the particle residual is the difference between a second actual observation value and a second predicted observation value of the particle filter;

[0019] The unbiased estimator of the observation noise covariance matrix of the unscented Kalman filter is determined by the maximum likelihood estimation method as the process noise; the posterior distribution parameters of the observation noise covariance matrix of the particle filter are determined by the Bayesian estimation method as the observation noise;

[0020] According to the process noise and the observation noise, the observation noise covariance matrix and the process noise covariance matrix of the next moment in the unscented Kalman filter update phase are recursively adjusted dynamically, and the calculation formula of the particle weight of the next moment in the particle filter update phase and the process noise sampling method of the particle prediction phase are dynamically adjusted.

[0021] In some embodiments of the present invention, the formula for determining the unbiased estimator of the observation noise covariance matrix of the unscented Kalman filter by the maximum likelihood estimation method is as follows:

[0022]

[0023] in, represents the unbiased estimator of the observation noise covariance matrix, N represents the number of samples of the observation data, z(t) represents the observation vector at time t, h() represents the nonlinear observation function, and x(t|t-1) represents the estimated value of the predicted state at time t;

[0024] The formula for determining the posterior distribution parameters of the particle filter's observation noise covariance matrix using the Bayesian estimation method is as follows:

[0025] p(R|Z1:t)∝p(Z1:t|R)p(R);

[0026] Where R represents the observation noise covariance matrix, Z1:t represents all observation data from time 1 to t, p(Z1:t|R) represents the likelihood function of the observation data, p(R|Z1:t) represents the posterior probability density function of the observation noise covariance matrix, and p(R) represents the prior probability density function of the observation noise covariance matrix.

[0027] In some embodiments of the present invention, the local reachable density of the trajectory data point under different neighbor values ​​is calculated using the following formula:

[0028]

[0029] Among them, LRD(p i ) represents the local reachable density corresponding to the trajectory data point, N k represents the set of k nearest neighbors, p i 、p j Represent the i-th and j-th trajectory data points respectively, d k (p i ,p j ) indicates p i To its kth nearest neighbor data point p j The distance, d k (p j ,p i ) indicates p j To its kth nearest neighbor data point p i The distance, N k (p i ) represents the k-nearest neighbor set of the i-th trajectory data point;

[0030] The following formula is used to calculate the local anomaly factor of the trajectory data point under different neighbor values:

[0031]

[0032] Among them, LOF(p i ) represents the corresponding local abnormal factor, LRD(p j ) represents the local reachable density corresponding to the j-th trajectory data point.

[0033] In some embodiments of the present invention, the multi-scale anomaly score is obtained by calculating local anomaly factor values ​​under different neighbor values ​​and integrating the anomaly factors under multiple scales.

[0034] Another aspect of the present invention provides a deep learning-based confidential carrier trajectory tracking and anomaly detection system to perform the above method, the system comprising:

[0035] The first unit is used to obtain multi-source heterogeneous data during the execution of a task by a target confidential carrier, wherein the multi-source heterogeneous data includes GPS data, inertial navigation data, and visual tracking data corresponding to the target confidential carrier; according to the data quality and prediction performance of the multi-source heterogeneous data, in combination with an adaptive weighting strategy, a corresponding data weight is assigned to the multi-source heterogeneous data, and the data weight is fused with the multi-source heterogeneous data to obtain carrier trajectory data;

[0036] The second unit is used for a hybrid filtering algorithm based on the combination of unscented Kalman filtering and particle filtering. The fused carrier trajectory data is initially smoothed by unscented Kalman filtering, and the initial filtering result is secondary filtered by particle filtering, and the optimized carrier trajectory information is obtained by combining the hybrid weight allocation strategy;

[0037] The third unit is used to calculate, for each trajectory data point in the optimized carrier trajectory information, a dynamic time warping distance matrix between it and other trajectory data points, and calculate, based on the dynamic time warping distance matrix, the local reachable density and local anomaly factor of the trajectory data point under different neighbor values; obtain a multi-scale anomaly score of the trajectory data point by integrating the local anomaly factor values ​​under multiple scales; and determine, based on a preset anomaly judgment threshold, a trajectory data point whose multi-scale anomaly score exceeds the anomaly judgment threshold as an anomaly point.

[0038] Another aspect of the present invention provides an electronic device, including a processor, a memory, and a computer program / instructions stored in the memory, wherein the processor is used to execute the computer program / instructions, and when the computer program / instructions are executed, the electronic device implements the steps of the method described in any of the above embodiments.

[0039] Another aspect of the present invention provides a computer-readable storage medium having a computer program / instruction stored thereon, which implements the steps of the method described in any of the above embodiments when the computer program / instruction is executed by a processor.

[0040] The present application adopts an adaptive weighting strategy to assign data weights to multi-source heterogeneous data, and fuses the data weights with the multi-source heterogeneous data to obtain carrier trajectory data. The method of the present application can adaptively adjust the data weights according to the data quality and prediction performance, effectively improving the accuracy and robustness of the trajectory data. Based on a hybrid filtering algorithm combining unscented Kalman filtering and particle filtering, the fused carrier trajectory data is subjected to preliminary smoothing and secondary filtering, and the optimized carrier trajectory information is obtained by combining the hybrid weight allocation strategy. The method of the present application makes full use of the linear optimization capability of the unscented Kalman filter and the nonlinear processing capability of the particle filter, effectively eliminating the noise and abnormal interference in the trajectory data, and improving the smoothness and continuity of the trajectory information.

[0041] Furthermore, for the optimized carrier trajectory information, the dynamic time-warped distance matrix between the trajectory data points is calculated, and the local reachable density and local anomaly factor are calculated under different neighbor values. The method of the present application can effectively measure the temporal similarity between trajectory data points, capture abnormal changes in local trajectory patterns, and improve the sensitivity and accuracy of anomaly detection. In addition, the local anomaly factor values ​​at multiple scales are integrated to obtain the multi-scale anomaly scores of the trajectory data points, and the anomaly points are identified according to the preset anomaly judgment threshold. By fusing multi-scale anomaly information, the method of the present application can comprehensively characterize the degree of abnormality of the trajectory data points, thereby improving the accuracy and reliability of anomaly detection.

[0042] Additional advantages, purposes, and features of the present invention will be described in part in the following description, and will become apparent to those skilled in the art after studying the following, or may be learned from the practice of the present invention. The purposes and other advantages of the present invention may be achieved and obtained by the structures specifically indicated in the specification and the accompanying drawings.

[0043] Those skilled in the art will appreciate that the objectives and advantages that can be achieved with the present invention are not limited to the above specific description, and the above and other objectives that can be achieved by the present invention will be more clearly understood from the following detailed description. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] The drawings described herein are used to provide a further understanding of the present invention, constitute a part of the present application, and do not constitute a limitation of the present invention. In the drawings:

[0045] Figure 1 Schematic diagram of the process of a method for tracking trajectory and detecting anomalies of confidential carriers based on deep learning in one embodiment of the present invention.

[0046] Figure 2 This is a structural diagram of a confidential carrier trajectory tracking and anomaly detection system based on deep learning in one embodiment of the present invention. DETAILED DESCRIPTION

[0047] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the embodiments and the accompanying drawings. Here, the illustrative embodiments of the present invention and their descriptions are used to explain the present invention, but are not intended to limit the present invention.

[0048] It should also be noted that, in order to avoid obscuring the present invention due to unnecessary details, only structures and / or processing steps closely related to the solutions according to the present invention are shown in the accompanying drawings, while other details that are not closely related to the present invention are omitted.

[0049] It should be emphasized that the term “include / comprises” when used herein refers to the presence of features, elements, steps or components, but does not exclude the presence or addition of one or more other features, elements, steps or components.

[0050] Figure 1 FIG. 1 is a flow chart of a method for tracking trajectory of a confidential carrier and detecting anomalies based on deep learning according to an embodiment of the present invention. Figure 1 As shown, the method includes:

[0051] Step S101: Acquire multi-source heterogeneous data during the execution of a task by a target confidential carrier, wherein the multi-source heterogeneous data includes GPS data, inertial navigation data, and visual tracking data corresponding to the target confidential carrier; assign corresponding data weights to the multi-source heterogeneous data based on the data quality and prediction performance of the multi-source heterogeneous data in combination with an adaptive weighting strategy, and fuse the data weights with the multi-source heterogeneous data to obtain carrier trajectory data.

[0052] For example, when fusing multi-source heterogeneous data to obtain carrier trajectory data, it is necessary to assign appropriate data weights through an adaptive weighting strategy based on the data quality and prediction performance of each data source. The following is an example of how to determine data quality and prediction performance, and assign weights based on quality and performance:

[0053] Assume that there are three data sources providing trajectory data of the target confidential carrier, namely GPS data, inertial navigation data and visual tracking data. The visual target tracking algorithm is a commonly used target tracking method, which can estimate the trajectory and motion state of the target object in the image sequence based on the initial state of the given target object. This application uses the visual target tracking algorithm to detect, extract, identify and track the image sequence captured during the execution of the target confidential carrier, and obtain the visual tracking data of the target confidential carrier.

[0054] 1. Determine data quality:

[0055] -GPS data: The quality of GPS data is evaluated by analyzing the number of visible satellites, signal strength, and positioning accuracy of GPS data. Assuming the average positioning error of GPS data is 5 meters, the data quality score is 0.8.

[0056] - Inertial navigation data: The quality of the inertial navigation data is evaluated by analyzing indicators such as the cumulative drift error and initial alignment error of the inertial navigation data. Assuming that the average position drift error of the inertial navigation data is 10 meters / hour, the data quality score is 0.7.

[0057] - Visual tracking data: The quality of visual tracking data is evaluated by analyzing indicators such as tracking success rate and tracking accuracy. Assuming that the average tracking error of visual tracking data is 3 pixels, the data quality score is 0.9.

[0058] 2. Determine the prediction performance:

[0059] -GPS data: Evaluate the prediction performance of GPS data by analyzing its historical performance in trajectory prediction of target confidential carriers. Assuming the average trajectory prediction error based on GPS data is 8 meters, the prediction performance score is 0.75.

[0060] -Inertial navigation data: Evaluate the prediction performance of inertial navigation data by analyzing its historical performance in trajectory prediction of target classified carriers. Assuming the average trajectory prediction error based on inertial navigation data is 12 meters, the prediction performance score is 0.7.

[0061] - Visual tracking data: Evaluate the prediction performance of visual tracking data by analyzing its historical performance in trajectory prediction of target confidential carriers. Assume that the average trajectory prediction error based on visual tracking data is 6 meters and the prediction performance score is 0.8.

[0062] 3. Assign data weights:

[0063] - Combining data quality and prediction performance scores, the weight of each data source is calculated through an adaptive weighting strategy.

[0064] -Assume that the weight coefficient of the data quality score is 0.6 and the weight coefficient of the prediction performance score is 0.4.

[0065] -GPS data weight = 0.6*0.8+0.4*0.75 = 0.78

[0066] -Inertial navigation data weight = 0.6*0.7+0.4*0.7 = 0.7

[0067] - Visual tracking data weight = 0.6*0.9+0.4*0.8 = 0.86

[0068] -Normalize the weights to get the final data weights:

[0069] -GPS data weight = 0.78 / (0.78+0.7+0.86) = 0.333

[0070] -Inertial navigation data weight = 0.7 / (0.78+0.7+0.86) = 0.299

[0071] - Visual tracking data weight = 0.86 / (0.78+0.7+0.86) = 0.368

[0072] 4. Fusion of multi-source heterogeneous data:

[0073] -Multiply the carrier trajectory data provided by each data source by the corresponding data weight to obtain the weighted trajectory data.

[0074] -Fusing the weighted trajectory data, such as taking a weighted average, to obtain the final carrier trajectory data.

[0075] The above examples illustrate how to use adaptive weighting strategies to assign data weights based on the data quality and prediction performance of multi-source heterogeneous data, and fuse the weights with the corresponding data to obtain carrier trajectory data. In practical applications, it is necessary to select appropriate quality assessment indicators and prediction performance evaluation methods based on the specific data source type, data characteristics, and application requirements, and design effective adaptive weighting strategies to achieve optimal fusion of multi-source heterogeneous data and improve the reliability and accuracy of carrier trajectory data.

[0076] Step S102: Based on a hybrid filtering algorithm combining unscented Kalman filtering and particle filtering, the fused carrier trajectory data is initially smoothed by unscented Kalman filtering, and the initial filtering result is secondary filtered by particle filtering, and the optimized carrier trajectory information is obtained by combining the hybrid weight allocation strategy.

[0077] Step S103: For each trajectory data point in the optimized carrier trajectory information, calculate the dynamic time warping distance matrix between it and other trajectory data points, and calculate the local reachable density and local anomaly factor of the trajectory data point under different neighbor values ​​according to the dynamic time warping distance matrix; combine the local anomaly factor values ​​under multiple scales to obtain the multi-scale anomaly score of the trajectory data point; and according to a preset anomaly judgment threshold, determine the trajectory data point whose multi-scale anomaly score exceeds the anomaly judgment threshold as an anomaly point. The multi-scale anomaly score is obtained by calculating the local anomaly factor values ​​under different neighbor values ​​and combining the anomaly factors under multiple scales.

[0078] This application adopts an adaptive weighting strategy to assign data weights to multi-source heterogeneous data, and fuses the data weights with the multi-source heterogeneous data to obtain carrier trajectory data. This method can adaptively adjust the data weights according to the data quality and prediction performance, effectively improving the accuracy and robustness of the trajectory data. In addition, based on a hybrid filtering algorithm combining unscented Kalman filtering and particle filtering, the fused carrier trajectory data is subjected to preliminary smoothing and secondary filtering, and the optimized carrier trajectory information is obtained by combining the hybrid weight allocation strategy. This method makes full use of the linear optimization capability of the unscented Kalman filter and the nonlinear processing capability of the particle filter, effectively eliminating the noise and abnormal interference in the trajectory data, and improving the smoothness and continuity of the trajectory information.

[0079] Furthermore, the present application calculates the dynamic time-warped distance matrix between trajectory data points for the optimized carrier trajectory information, and calculates the local reachable density and local anomaly factor under different neighbor values. This method can effectively measure the temporal similarity between trajectory data points, capture abnormal changes in local trajectory patterns, and improve the sensitivity and accuracy of anomaly detection. The present application combines the local anomaly factor values ​​at multiple scales to obtain the multi-scale anomaly score of the trajectory data point, and identifies the anomaly point according to the preset anomaly judgment threshold. By fusing multi-scale anomaly information, this method can comprehensively characterize the degree of abnormality of the trajectory data point, thereby improving the accuracy and reliability of anomaly detection.

[0080] In some embodiments of the present invention, based on a hybrid filtering algorithm combining unscented Kalman filtering and particle filtering, the fused carrier trajectory data is initially smoothed by unscented Kalman filtering, and the initial filtering result is secondary filtered by particle filtering, and the optimized carrier trajectory information is obtained by combining a hybrid weight allocation strategy, including:

[0081] Perform unscented Kalman filtering on the fused carrier trajectory data, establish the state space model of the target confidential carrier, determine the initial state variables and initial observation variables, construct the state transfer equation and observation equation according to the dynamic characteristics of the target confidential carrier, initialize the parameters of the unscented Kalman filter, recursively execute the prediction and update steps of the unscented Kalman filter, determine the target state variables of the target confidential carrier based on the minimum mean square error criterion, and update the state covariance matrix to achieve preliminary smoothing of the carrier trajectory data and obtain the initial filtering result;

[0082] Initialize the particle set of the particle filter according to the preliminary filtering result, predict the next state and observation value of each particle according to the state transfer equation and the observation equation, calculate the particle weight of each particle by using the likelihood function or the probability density function in combination with the actual observation value of each particle, resample the particle set according to the particle weight, iteratively update the particle set, determine the state distribution of the target confidential carrier by the Monte Carlo simulation method, and obtain the particle filtering result;

[0083] Through a hybrid weight allocation strategy, a first hybrid weight is allocated to the initial filtering result, a second hybrid weight is allocated to the particle filtering result, and the initial filtering result and the particle filtering result are weightedly fused to obtain optimized carrier trajectory information.

[0084] Exemplarily, in order to further improve the smoothing effect of carrier trajectory data, this application proposes a hybrid filtering algorithm based on the combination of unscented Kalman filtering and particle filtering. The algorithm makes full use of the linear optimization capability of the unscented Kalman filter and the nonlinear processing capability of the particle filter, and realizes the refined processing of the carrier trajectory data through the cascade and fusion of the two filters.

[0085] First, the fused carrier trajectory data is subjected to unscented Kalman filtering, and the initial state variables and initial observation variables are determined by establishing the state space model of the target confidential carrier. According to the dynamic characteristics of the target confidential carrier, the state transfer equation and the observation equation are constructed. In the unscented Kalman filtering process, the state transfer equation and the observation equation need to be constructed according to the dynamic characteristics of the target confidential carrier. The dynamic characteristics of the target confidential carrier describe the physical characteristics and constraints of the carrier during the motion process, and reflect the dynamic relationship between the carrier state variables. Examples are given of several common dynamic characteristics of target confidential carriers, including uniform linear motion, uniformly accelerated linear motion, circular motion, and ballistic motion. In practical applications, it is necessary to select a suitable dynamic model based on the specific type and motion characteristics of the target confidential carrier, and determine the state variables and state transfer equations based on the actual situation.

[0086] The state transfer equation is as follows:

[0087] x(t+1)=F(t)x(t)+w(t);

[0088] The observation equation is as follows:

[0089] z(t)=H(t)x(t)+v(t);

[0090] Among them, x(t) is the state variable at time t, F(t) is the state transfer matrix, w(t) is the process noise, z(t) is the observation variable at time t, H(t) is the observation matrix, and v(t) is the observation noise.

[0091] Initialize the parameters of the unscented Kalman filter, including the initial state estimate x(0|0), the initial state covariance matrix P(0|0), the process noise covariance matrix Q, and the observation noise covariance matrix R. Recursively perform the prediction and update steps of the unscented Kalman filter, determine the target state variable of the target confidential carrier based on the minimum mean square error criterion, and update the state covariance matrix. The prediction and update formulas of the unscented Kalman filter are as follows:

[0092] Prediction steps:

[0093] x(t|t-1)=F(t-1)x(t-1|t-1);

[0094] P(t|t-1)=F(t-1)P(t-1|t-1)F T (t-1)+Q;

[0095] Update steps:

[0096] K(t)=P(t|t-1)H T (t)[H(t)P(t|t-1)H T (t)] -1 ;

[0097] x(t|t)=x(t|t-1)+K(t)[z(t)-H(t)x(t|t-1)];

[0098] P(t|t)=[IK(t)H(t)]P(t|t-1);

[0099] Among them, x(t|t-1) and P(t|t-1) are the predicted state estimate and the predicted state covariance matrix, K(t) is the Kalman gain, and x(t|t) and P(t|t) are the updated state estimate and the updated state covariance matrix. The unscented Kalman filter can achieve preliminary smoothing of the carrier trajectory data and obtain the initial filtering result.

[0100] Then, the particle set {x i (0), i = 1, 2, ..., M}, where M is the number of particles. According to the state transfer equation and observation equation, the next state x of each particle is predicted. i (t|t-1) and the observed value z i (t|t-1). Combined with the actual observation value z(t) of each particle, the likelihood function p(z(t)|x i (t|t-1)) or probability density function to calculate the particle weight w of each particle i (t), as shown in the following formula:

[0101] w i (t) = p(z(t)|x i (t|t-1));

[0102] Resample the particle set according to the particle weight to obtain a new particle set {x i (t), i = 1, 2, ..., N}. Iteratively update the particle set, determine the state distribution p(x(t)|z(1:t)) of the target confidential carrier through the Monte Carlo simulation method, and obtain the particle filtering result.

[0103] w i(t) represents the weight of the i-th particle at time t. The particle weight reflects the closeness of the particle to the true state. The larger the weight, the more likely the particle is to be close to the true state.

[0104] p(z(t)|x i (t|t-1)) means at time t, given the predicted state x of the ith particle i The conditional probability density of the observed value z(t) under (t|t-1). This conditional probability density reflects the degree of agreement between the predicted state and the actual observed value.

[0105] z(t) represents the actual observation value at time t, that is, the measurement value of the target state obtained by the sensor.

[0106] x i (t|t-1) represents the estimated state value of the i-th particle at time t, which is predicted based on the particle state at time t-1 and the system model. This state estimate is the predicted state obtained by transferring the state of the particle at the previous moment.

[0107] p(x(t)|z(1:t)) represents the posterior probability density of the target state x(t) at time t, given all observations z(1:t) from time 1 to time t. This posterior probability density reflects the estimated distribution of the target state after considering all observations.

[0108] x(t) represents the target state at time t, which is the quantity we want to estimate and track. In the target tracking problem, the target state usually includes information such as the target's position, velocity, and acceleration.

[0109] z(1:t) represents all observations from time 1 to time t, that is, the sensor's measurement sequence of the target state. It includes all observation information from the initial moment to the current moment.

[0110] z(1): represents the observation value at time 1.

[0111] z(2): represents the observation value at time 2. ......

[0112] z(t): represents the observation value at time t.

[0113] In summary, p(x(t)|z(1:t)) represents the posterior probability density of the target state x(t) at time t, given all observations z(1:t) from time 1 to time t. This posterior probability density represents the optimal estimated distribution of the target state after considering all observation information. It combines prior knowledge and observation data and reflects the latest understanding of the target state.

[0114] Finally, the initial filtering result and the particle filtering result are fused through a hybrid weight allocation strategy. The first hybrid weight α is assigned to the initial filtering result, and the second hybrid weight β is assigned to the particle filtering result, satisfying α+β=1. The hybrid weight can be adaptively adjusted according to the confidence or estimation error of the initial filtering result and the particle filtering result. The two filtering results are weightedly fused to obtain the optimized carrier trajectory information, as shown in the following formula:

[0115] x(t)=αx UKF(t) +βx PF(t) ;

[0116] Among them, x UKF(t) is the state estimate of the unscented Kalman filter, x PF(t) is the state estimate of the particle filter.

[0117] Through the cascade fusion of unscented Kalman filtering and particle filtering, the advantages of the two filtering algorithms can be fully utilized to achieve refined processing of carrier trajectory data. Unscented Kalman filtering can effectively process linear systems and perform preliminary smoothing of trajectory data; particle filtering can process nonlinear non-Gaussian systems and capture subtle changes in trajectory data. The hybrid weight allocation strategy can adaptively adjust the contribution of the two filtering results to improve the optimization effect of trajectory information. The hybrid filtering algorithm of the combined use of unscented Kalman filtering and particle filtering can significantly improve the smoothness and continuity of carrier trajectory data, laying a good data foundation for subsequent trajectory anomaly detection.

[0118] In some embodiments of the present invention, the method further comprises:

[0119] In the update phase of the unscented Kalman filter and the particle filter, the filter residual and the particle residual are calculated, wherein the filter residual is the difference between the first actual observation value and the first predicted observation value of the unscented Kalman filter, and the particle residual is the difference between the second actual observation value and the second predicted observation value of the particle filter;

[0120] The unbiased estimator of the observation noise covariance matrix of the unscented Kalman filter is determined by the maximum likelihood estimation method as the process noise, and the posterior distribution parameters of the observation noise covariance matrix of the particle filter are determined by the Bayesian estimation method as the observation noise;

[0121] According to the process noise and the observation noise, the observation noise covariance matrix and the process noise covariance matrix of the next moment in the unscented Kalman filter update phase are recursively adjusted dynamically, and the calculation formula of the particle weight of the next moment in the particle filter update phase and the process noise sampling method of the particle prediction phase are dynamically adjusted.

[0122] Exemplarily, in the update phase of the unscented Kalman filter and the particle filter, the filter residual and the particle residual are calculated respectively. The filter residual of the unscented Kalman filter is defined as the difference between the first actual observation value and the first predicted observation value. The particle residual of the particle filter is defined as the difference between the second actual observation value and the second predicted observation value.

[0123] The unbiased estimator of the observation noise covariance matrix of the unscented Kalman filter is determined by the maximum likelihood estimation method as the process noise. Assuming that the observation noise follows a Gaussian distribution with a mean of zero, the unbiased estimator of the observation noise covariance matrix can be expressed as the covariance of the filter residual.

[0124] The posterior distribution parameters of the observation noise covariance matrix of the particle filter are determined by the Bayesian estimation method as the observation noise. Assuming that the observation noise obeys the inverse Wishart distribution, according to the conjugate prior principle, the posterior distribution of the observation noise covariance matrix also obeys the inverse Wishart distribution. The degree of freedom parameter and scale matrix parameter of the inverse Wishart distribution are updated by the sample covariance matrix of the particle residual.

[0125] According to the estimated process noise and observation noise, the relevant parameters of the unscented Kalman filter and the particle filter are dynamically adjusted.

[0126] For particle filtering, the calculation formula of the particle weight at the next moment and the sampling method of the process noise in the particle prediction stage are dynamically adjusted. Substitute the observed noise covariance matrix into the particle weight calculation formula to obtain the revised particle weight calculation formula. In the particle prediction stage, according to the estimated process noise, the process noise sample is sampled from the corresponding Gaussian distribution to predict the next state of the particle.

[0127] This application can achieve online adaptive optimization of unscented Kalman filtering and particle filtering through adaptive noise estimation and dynamic parameter adjustment. By estimating the observation noise and process noise based on the filter residuals and particle residuals, and dynamically adjusting the parameters of the filter, it can effectively adapt to the real-time changes of the carrier trajectory data and improve the adaptability and robustness of the filtering algorithm. The introduction of adaptive noise estimation and dynamic parameter adjustment mechanism can further improve the smoothing effect of the hybrid filtering algorithm on the carrier trajectory data, and provide more reliable and accurate data support for subsequent trajectory anomaly detection.

[0128] In some embodiments of the present invention, the formula for determining the unbiased estimator of the observation noise covariance matrix of the unscented Kalman filter by the maximum likelihood estimation method is as follows:

[0129]

[0130] in, represents the unbiased estimator of the observation noise covariance matrix, N represents the number of samples of the observation data, z(t) represents the observation vector at time t, h() represents the nonlinear observation function, and x(t|t-1) represents the estimated value of the predicted state at time t;

[0131] The formula for determining the posterior distribution parameters of the particle filter's observation noise covariance matrix using the Bayesian estimation method is as follows:

[0132] p(R|Z1:t)∝p(Z1:t|R)p(R);

[0133] Where R represents the observation noise covariance matrix, Z1:t represents all observation data from time 1 to t, p(Z1:t|R) represents the likelihood function of the observation data, p(R|Z1:t) represents the posterior probability density function of the observation noise covariance matrix, and p(R) represents the prior probability density function of the observation noise covariance matrix.

[0134] In some embodiments of the present invention, for each trajectory data point in the optimized carrier trajectory information, calculating the dynamic time warping distance matrix between it and other trajectory data points includes:

[0135] Convert each trajectory data point into a time series, calculate the sequence similarity between the two time series through the optimal nonlinear alignment of the time series corresponding to the two trajectory data points, and summarize multiple sequence similarities into a dynamic time warping distance matrix;

[0136] Calculating the local reachable density of the trajectory data point under different neighbor values ​​includes:

[0137] Measuring the reciprocal of the average reachable distance between the trajectory data point and its neighboring data points, where the reciprocal of the average reachable distance reflects the local reachability density of the area where the trajectory data point is located;

[0138] Adaptively determine the neighbor values ​​corresponding to the trajectory data points, and use the multi-scale local anomaly factor method to calculate the local anomaly factor values ​​under different neighbor values, that is, to obtain the local anomaly factor value by integrating multiple local reachable densities.

[0139] For example, after obtaining the optimized carrier trajectory information, it is necessary to further analyze the abnormality of each trajectory data point to achieve accurate detection of trajectory anomalies. This paper proposes a trajectory anomaly detection method based on dynamic time warping and multi-scale local anomaly factors, which effectively identifies trajectory anomalies by calculating the dynamic time warping distance matrix and local reachability density between trajectory data points.

[0140] For each trajectory data point in the optimized carrier trajectory information, it is first converted into the form of a time series. Then, the sequence similarity between the two time series is calculated by the optimal nonlinear alignment of the time series corresponding to the two trajectory data points. The sequence similarity reflects the degree of similarity between the two trajectory data points in the time dimension, taking into account the temporal dynamic characteristics of the trajectory data points. Moreover, the present application aggregates multiple sequence similarities into a dynamic time warping distance matrix by calculating the sequence similarity between each pair of trajectory data points. The dynamic time warping distance matrix characterizes the temporal dynamic similarity between trajectory data points, and provides an important distance metric for subsequent anomaly detection.

[0141] On the basis of calculating the dynamic time warping distance matrix, the local reachable density of each trajectory data point under different neighbor values ​​is further calculated. The local reachable density reflects the local density characteristics of the area where the trajectory data point is located, and can be used to measure the degree of abnormality of the trajectory data point. Specifically, by calculating the inverse of the average reachable distance between the trajectory data point and its neighboring data points, the local density estimate of the trajectory data point is obtained. The larger the inverse of the average reachable distance, the higher the local density of the area where the trajectory data point is located, and the less likely the data point is to be an abnormal point.

[0142] In order to improve the adaptability and robustness of the local reachable density, this application adopts a multi-scale local anomaly factor method. This method adaptively determines the neighbor value of each trajectory data point and calculates the local anomaly factor value under different neighbor values. The local anomaly factor value reflects the degree of abnormality of the trajectory data point relative to its neighborhood data points. By comprehensively considering the local anomaly factor values ​​under multiple neighbor values, the comprehensive local reachable density of the trajectory data point is obtained. The multi-scale local anomaly factor method can adaptively capture abnormal features at different scales and improve the accuracy and robustness of anomaly detection.

[0143] The comprehensive use of dynamic time warping distance matrix and multi-scale local anomaly factor method can achieve accurate anomaly detection of carrier trajectory information. First, the dynamic time warping distance matrix is ​​used to measure the temporal dynamic similarity between trajectory data points, revealing the abnormal characteristics of trajectory data points in the time dimension. Then, the local reachable density of trajectory data points is calculated by the multi-scale local anomaly factor method to capture the abnormal characteristics of trajectory data points in the spatial dimension. Finally, considering the abnormal characteristics of the time dimension and the spatial dimension, the trajectory abnormal data points are identified by setting a suitable threshold.

[0144] Compared with traditional trajectory anomaly detection methods, the method proposed in this application for obtaining comprehensive local reachable density based on dynamic time warping and multi-scale local anomaly factors has the following advantages: First, the temporal dynamic characteristics of trajectory data points are taken into account, and the temporal similarity between trajectory data points is measured by the dynamic time warping distance matrix, thereby improving the temporal sensitivity of anomaly detection; second, a multi-scale local anomaly factor method is adopted to adaptively capture anomaly features at different scales, thereby improving the adaptability and robustness of anomaly detection; third, the anomaly features of the time dimension and the space dimension are comprehensively considered, thereby achieving comprehensive and accurate identification of trajectory anomalies.

[0145] By introducing the dynamic time-warping distance matrix and multi-scale local anomaly factor method based on the optimization of carrier trajectory information, accurate detection and identification of carrier trajectory anomalies can be achieved. This application makes full use of the spatiotemporal characteristics of trajectory data, adaptively captures multi-scale anomaly features, and provides reliable data support for subsequent abnormal trajectory analysis and disposal, which is of great significance for ensuring the security control of confidential carriers.

[0146] In some embodiments of the present invention, the local reachable density of the trajectory data point under different neighbor values ​​is calculated using the following formula:

[0147]

[0148] Among them, LRD(p i ) represents the local reachability density corresponding to the i-th trajectory data point, N k represents the set of k nearest neighbors, p i 、p j denote the i-th and j-th trajectory data points respectively, d k (p i ,p j ) indicates p i To its kth nearest neighbor data point p j The distance, d k (p j ,p i ) indicates p j To its kth nearest neighbor data point p i The distance, N k (p i ) represents the k-nearest neighbor set of the i-th trajectory data point;

[0149] The local anomaly factor of the trajectory data point under different neighbor values ​​is calculated as shown in the following formula:

[0150]

[0151] Among them, LOF(p i ) represents the corresponding local abnormal factor, LRD(pj ) represents the local reachable density corresponding to the j-th trajectory data point.

[0152] Figure 2 FIG. 1 is a schematic diagram of the structure of a system for tracking and detecting secret-related carrier trajectories and abnormalities based on deep learning according to an embodiment of the present invention. Figure 2 As shown, the system comprises:

[0153] The first unit is used to obtain multi-source heterogeneous data during the execution of a task by a target confidential carrier, wherein the multi-source heterogeneous data includes GPS data, inertial navigation data, and visual tracking data corresponding to the target confidential carrier; according to the data quality and prediction performance of the multi-source heterogeneous data, in combination with an adaptive weighting strategy, a corresponding data weight is assigned to the multi-source heterogeneous data, and the data weight is fused with the multi-source heterogeneous data to obtain carrier trajectory data;

[0154] The second unit is used for a hybrid filtering algorithm based on the combination of unscented Kalman filtering and particle filtering. The fused carrier trajectory data is initially smoothed by unscented Kalman filtering, and the initial filtering result is secondary filtered by particle filtering, and the optimized carrier trajectory information is obtained by combining the hybrid weight allocation strategy;

[0155] The third unit is used to calculate, for each trajectory data point in the optimized carrier trajectory information, a dynamic time warping distance matrix between it and other trajectory data points, and calculate the local reachable density and local anomaly factor of the trajectory data point under different neighbor values ​​according to the dynamic time warping distance matrix; obtain the multi-scale anomaly score of the trajectory data point by integrating the local anomaly factor values ​​under multiple scales; and determine, according to a preset anomaly judgment threshold, the trajectory data point whose multi-scale anomaly score exceeds the anomaly judgment threshold as an anomaly point, wherein the multi-scale anomaly score is obtained by calculating the local anomaly factor values ​​under different neighbor values ​​and integrating the anomaly factors under multiple scales.

[0156] The step of calculating the dynamic time warping distance matrix between each trajectory data point in the optimized carrier trajectory information and other trajectory data points includes:

[0157] Convert each trajectory data point into a time series, calculate the sequence similarity between the two time series through the optimal nonlinear alignment of the time series corresponding to the two trajectory data points, and summarize multiple sequence similarities into a dynamic time warping distance matrix;

[0158] The calculating the local reachable density of the trajectory data point under different neighbor values ​​includes:

[0159] Measuring the reciprocal of the average reachable distance between the trajectory data point and its neighboring data points, where the reciprocal of the average reachable distance reflects the local reachability density of the area where the trajectory data point is located;

[0160] Adaptively determine the neighbor values ​​corresponding to the trajectory data points, and use the multi-scale local anomaly factor method to calculate the local anomaly factor values ​​under different neighbor values, that is, to obtain the local anomaly factor value by integrating multiple local reachable densities.

[0161] Corresponding to the above method, the present invention also provides an electronic device, which includes a computer device, the computer device includes a processor and a memory, the memory stores a computer program / instructions, the processor is used to execute the computer program / instructions stored in the memory, and when the computer program / instructions are executed by the processor, the electronic device implements the steps of the method described above.

[0162] The embodiment of the present invention also provides a computer-readable storage medium on which a computer program / instruction is stored, and when the computer program / instruction is executed by a processor, the steps of the aforementioned edge computing server deployment method are implemented. The computer-readable storage medium can be a tangible storage medium, such as a random access memory (RAM), a memory, a read-only memory (ROM), an electrically programmable ROM, an electrically erasable programmable ROM, a register, a floppy disk, a hard disk, a removable storage disk, a CD-ROM, or any other form of storage medium known in the technical field.

[0163] It should be understood by those skilled in the art that the exemplary components, systems and methods described in conjunction with the embodiments disclosed herein can be implemented in hardware, software or a combination of the two. Whether it is performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention. When implemented in hardware, it can be, for example, an electronic circuit, an application specific integrated circuit (ASIC), appropriate firmware, a plug-in, a function card, etc. When implemented in software, the elements of the present invention are programs or code segments used to perform the required tasks. The program or code segment can be stored in a machine-readable medium, or transmitted on a transmission medium or a communication link via a data signal carried in a carrier.

[0164] It should be clear that the present invention is not limited to the specific configuration and processing described above and shown in the figures. For the sake of simplicity, a detailed description of the known method is omitted here. In the above embodiments, several specific steps are described and shown as examples. However, the method process of the present invention is not limited to the specific steps described and shown, and those skilled in the art can make various changes, modifications and additions, or change the order between the steps after understanding the spirit of the present invention.

[0165] In the present invention, features described and / or illustrated for one embodiment may be used in the same or similar manner in one or more other embodiments, and / or combined with features of other embodiments or replace features of other embodiments.

[0166] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the embodiments of the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for tracking and detecting secret carrier trajectories based on deep learning, characterized in that: The method comprises the following steps: Acquire multi-source heterogeneous data of the target confidential carrier during the execution of the mission, wherein the multi-source heterogeneous data includes GPS data, inertial navigation data and visual tracking data corresponding to the target confidential carrier; assign corresponding data weights to the multi-source heterogeneous data according to the data quality and prediction performance of the multi-source heterogeneous data in combination with an adaptive weighting strategy, and fuse the data weights with the multi-source heterogeneous data to obtain carrier trajectory data; Based on the hybrid filtering algorithm combining unscented Kalman filtering and particle filtering, the fused carrier trajectory data is initially smoothed by unscented Kalman filtering, and the initial filtering results are secondary filtered by particle filtering, and the optimized carrier trajectory information is obtained by combining the hybrid weight allocation strategy; For each trajectory data point in the optimized carrier trajectory information, a dynamic time warping distance matrix between it and other trajectory data points is calculated, and according to the dynamic time warping distance matrix, the local reachable density and local anomaly factor of the trajectory data point under different neighbor values ​​are calculated; the local anomaly factor values ​​under multiple scales are integrated to obtain a multi-scale anomaly score of the trajectory data point; according to a preset anomaly judgment threshold, the trajectory data point whose multi-scale anomaly score exceeds the anomaly judgment threshold is determined as an anomaly point; The step of calculating the dynamic time warping distance matrix between each trajectory data point in the optimized carrier trajectory information and other trajectory data points includes: converting each trajectory data point into a time series, calculating the sequence similarity between the two time series by optimal nonlinear alignment of the time series corresponding to the two trajectory data points, and aggregating multiple sequence similarities into a dynamic time warping distance matrix; The calculating the local reachable density and the local anomaly factor of the trajectory data point under different neighbor values ​​includes: Measuring the reciprocal of the average reachable distance between the trajectory data point and its neighboring data points, where the reciprocal of the average reachable distance reflects the local reachability density of the area where the trajectory data point is located; Adaptively determine the neighbor values ​​corresponding to the trajectory data points, and use the multi-scale local anomaly factor method to calculate the local anomaly factor values ​​under different neighbor values; The method of combining the hybrid weight allocation strategy to obtain optimized carrier trajectory information includes: A first mixing weight is assigned to the initial filtering result, a second mixing weight is assigned to the particle filtering result, and the optimized carrier trajectory information is obtained by weighted fusion of the initial filtering result and the particle filtering result.

2. The method according to claim 1, characterized in that The hybrid filtering algorithm based on the combination of unscented Kalman filtering and particle filtering performs preliminary smoothing processing on the fused carrier trajectory data through unscented Kalman filtering, and performs secondary filtering on the obtained initial filtering results through particle filtering, and combines the hybrid weight allocation strategy to obtain optimized carrier trajectory information, including: Perform unscented Kalman filtering on the fused carrier trajectory data, establish a state space model of the target confidential carrier, determine the initial state variables and initial observation variables, construct the state transfer equation and observation equation according to the dynamic characteristics of the target confidential carrier, initialize the parameters of the unscented Kalman filter, recursively execute the prediction and update steps of the unscented Kalman filter, determine the target state variables of the target confidential carrier based on the minimum mean square error criterion, and update the state covariance matrix to achieve preliminary smoothing of the carrier trajectory data and obtain the initial filtering result; Initialize the particle set of the particle filter according to the preliminary filtering result, predict the next state and observation value of each particle according to the state transfer equation and the observation equation, calculate the particle weight of each particle by using the likelihood function or the probability density function in combination with the actual observation value of each particle, resample the particle set according to the particle weight, iteratively update the particle set, determine the state distribution of the target confidential carrier by the Monte Carlo simulation method, and obtain the particle filtering result; The initial filtering result and the particle filtering result are subjected to a mixed weight allocation strategy, wherein a first mixed weight is allocated to the initial filtering result and a second mixed weight is allocated to the particle filtering result, and weighted fusion is performed to obtain optimized carrier trajectory information.

3. The method according to claim 2, characterized in that The method further comprises: In the update phase of the unscented Kalman filter and the particle filter, a filter residual and a particle residual are calculated, wherein the filter residual is the difference between a first actual observation value and a first predicted observation value of the unscented Kalman filter, and the particle residual is the difference between a second actual observation value and a second predicted observation value of the particle filter; The unbiased estimator of the observation noise covariance matrix of the unscented Kalman filter is determined by the maximum likelihood estimation method as the process noise; the posterior distribution parameters of the observation noise covariance matrix of the particle filter are determined by the Bayesian estimation method as the observation noise; According to the process noise and the observation noise, the observation noise covariance matrix and the process noise covariance matrix of the next moment in the unscented Kalman filter update phase are recursively adjusted dynamically, and the calculation formula of the particle weight of the next moment in the particle filter update phase and the process noise sampling method of the particle prediction phase are dynamically adjusted.

4. The method according to claim 3, characterized in that The formula for determining the unbiased estimator of the observation noise covariance matrix of the unscented Kalman filter by the maximum likelihood estimation method is as follows: in, represents the unbiased estimator of the observation noise covariance matrix, N represents the number of samples of the observation data, z(t) represents the observation vector at time t, h() represents the nonlinear observation function, and x(t|t-1) represents the estimated value of the predicted state at time t; The formula for determining the posterior distribution parameters of the particle filter's observation noise covariance matrix using the Bayesian estimation method is as follows: p(R|Z1:t)∝p(Z1:t|R)p(R); Where R represents the observation noise covariance matrix, Z1:t represents all observation data from time 1 to t, p(Z1:t|R) represents the likelihood function of the observation data, p(R|Z1:t) represents the posterior probability density function of the observation noise covariance matrix, and p(R) represents the prior probability density function of the observation noise covariance matrix.

5. The method according to claim 1, characterized in that The local reachability density of the trajectory data point under different neighbor values ​​is calculated using the following formula: Among them, LRD(p i ) represents the local reachability density corresponding to the i-th trajectory data point, N k represents the set of k nearest neighbors, p i 、p j Represent the i-th and j-th trajectory data points respectively, d k (p i , p j ) indicates p i To its kth nearest neighbor data point p j The distance, d k (p j , p i ) indicates p j To its kth nearest neighbor data point p i The distance, N k (p i ) represents the k-nearest neighbor set of the i-th trajectory data point; The following formula is used to calculate the local anomaly factor of the trajectory data point under different neighbor values: Among them, LOF(p i ) represents the corresponding local abnormal factor, LRD(p j ) represents the local reachable density corresponding to the j-th trajectory data point.

6. The method according to claim 1, characterized in that: The multi-scale anomaly score is obtained by calculating the local anomaly factor values ​​under different neighbor values ​​and integrating the anomaly factors under multiple scales.

7. A deep learning-based confidential carrier trajectory tracking and anomaly detection system, used to implement the method described in any one of claims 1 to 6, characterized in that: The system includes: The first unit is used to obtain multi-source heterogeneous data during the execution of a task by a target confidential carrier, wherein the multi-source heterogeneous data includes GPS data, inertial navigation data, and visual tracking data corresponding to the target confidential carrier; according to the data quality and prediction performance of the multi-source heterogeneous data, in combination with an adaptive weighting strategy, a corresponding data weight is assigned to the multi-source heterogeneous data, and the data weight is fused with the multi-source heterogeneous data to obtain carrier trajectory data; The second unit is used for a hybrid filtering algorithm based on the combination of unscented Kalman filtering and particle filtering. The fused carrier trajectory data is initially smoothed by unscented Kalman filtering, and the initial filtering result is secondary filtered by particle filtering, and the optimized carrier trajectory information is obtained by combining the hybrid weight allocation strategy; The third unit is used to calculate, for each trajectory data point in the optimized carrier trajectory information, a dynamic time warping distance matrix between it and other trajectory data points, and calculate the local reachable density and local anomaly factor of the trajectory data point under different neighbor values ​​according to the dynamic time warping distance matrix; obtain a multi-scale anomaly score of the trajectory data point by integrating the local anomaly factor values ​​under multiple scales; and determine, according to a preset anomaly judgment threshold, a trajectory data point whose multi-scale anomaly score exceeds the anomaly judgment threshold as an anomaly point; The method of combining the hybrid weight allocation strategy to obtain optimized carrier trajectory information includes: A first mixing weight is assigned to the initial filtering result, a second mixing weight is assigned to the particle filtering result, and the optimized carrier trajectory information is obtained by weighted fusion of the initial filtering result and the particle filtering result.

8. An electronic device comprising a processor, a memory and a computer program / instruction stored in the memory, characterized in that: The processor is used to execute the computer program / instructions. When the computer program / instructions are executed, the electronic device implements the steps of the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program / instruction stored thereon, characterized in that: When the computer program / instructions are executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images