CN-121995362-A - Intelligent tracking method for high maneuvering target in complex clutter environment

CN121995362ACN 121995362 ACN121995362 ACN 121995362ACN-121995362-A

Abstract

The application belongs to the technical field of target tracking, and discloses an intelligent tracking method for a high maneuvering target in a complex clutter environment. The method comprises the steps of carrying out differential and normalization preprocessing on radar measurement, constructing and training a space-time feature extraction network, integrating a convolutional neural network with a gating circulation unit, extracting space-time features of target motion and predicting states, constructing and training a twin network based on VGG16, evaluating similarity of target and clutter distribution features, realizing preliminary screening before correlation, carrying out matched screening on new measurement in a wave gate by utilizing the space-time network predicting states and utilizing the twin network in a tracking stage, and finally completing optimal data correlation and track updating by a Hungary algorithm. According to the method, state prediction and association judgment are optimized through the deep learning network, so that tracking accuracy and robustness of the high maneuvering target in a complex clutter environment are improved effectively.

Inventors

WANG HAITAO
XIA TIAN
CHEN SHICHAO
FAN YIFEI
SU JIA
LI TAO
GUO ZIXUN
TAO MINGLIANG
WANG LING

Assignees

西北工业大学

Dates

Publication Date: 20260508
Application Date: 20260108

Claims (10)

1. The intelligent tracking method for the high maneuvering target in the complex clutter environment is characterized by comprising the following steps of: S1, acquiring a high maneuvering target true value state after radar scanning And corresponding measurement Performing differential processing on the measurement sequence to obtain a differential measurement sequence And for the differential measurement sequence Normalization processing is carried out, and the normalized differential measurement sequence and the corresponding differential target state are processed Composition of training samples Wherein The difference value between the normalized target true value state and the normalized measurement is obtained; s2, constructing a space-time feature extraction network and based on the training sample The spatial feature extraction module is a convolutional neural network and is used for extracting spatial features of a normalized differential measurement sequence, and the time sequence feature extraction module is a gating circulation unit network and is used for extracting time sequence features of the normalized differential measurement sequence and outputting a predicted differential target state; S3, acquiring a distribution feature matrix of a target and clutter, and preprocessing the distribution feature matrix to construct a distribution feature data set, constructing a twin network which is based on a VGG16 main network and shares weight with two branches, training the twin network based on the distribution feature data set and a binary cross entropy loss function, and enabling the twin network to output a similarity value representing the similarity of two input distribution features; S4, for the test time Will be of length Historical measurement sequence of (a) Performing differential processing and normalization processing, and inputting the trained space-time feature extraction network to obtain a predicted differential target state Obtaining the moment after inverse normalization and reduction Target state prediction value of (2) ; S5, at the moment For new measurements within associated wave gates Measurements associated with tracks from a previous time Calculating the similarity of the distribution characteristics measured at the moment on each track and the newly increased measurement subset with the similarity higher than a preset threshold is screened; S6, constructing an associated cost matrix based on the target state predicted value for the screened newly added measurement subset and the existing flight path, solving the associated cost matrix by applying a Hungary algorithm to obtain an optimal matching relation between the newly added measurement and the flight path, and extracting a network to update the flight path state according to the matching relation.
2. The method for intelligent tracking of high maneuver target in complex clutter environment according to claim 1, wherein the differential processing in step S1 comprises time period In-line measurement sequence Performing differential processing to extract dynamic characteristics of the target and obtain differential measurement sequences : ; The normalization processing adopts a min-max normalization method, and the numerical range of the differential measurement sequence is mapped to the [ -1,1] interval.
3. The intelligent tracking method for high maneuvering target in complex clutter environment according to claim 1, wherein the loss function adopted in step S2 is : Wherein, the Is a true differential target state that is to be determined, In order to predict the differential target state, Is the number of samples.
4. The intelligent tracking method for the high maneuver target in the complex clutter environment according to claim 1, wherein the preprocessing of the distribution feature matrix in step S3 includes normalizing the distribution feature matrix, and then adjusting the matrix to a preset size by a bicubic interpolation algorithm.
5. The intelligent tracking method for high mobility targets in complex clutter environment according to claim 1, wherein the loss function adopted in the training of the twin network in step S3 Is a binary cross entropy loss function.
6. The intelligent tracking method for high maneuvering targets in complex clutter environments according to claim 1, wherein the preset threshold in step S5 is 0.9.
7. The intelligent tracking method for high maneuvering target in complex clutter environment according to claim 1, wherein the correlation cost matrix in step S6 Elements of (a) The method comprises the following steps: Wherein, the Is the first after screening The newly added measurement, H is the observation matrix, Is the first Predicted state of the strip track.
8. The method of claim 1, wherein the motion model of the high maneuver target comprises a constant velocity model, a uniform acceleration model, a coordinated turning model, and any combination thereof.
9. An electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor implements the intelligent tracking method of high maneuver targets in complex clutter environments of any one of claims 1-8 when executing the program.
10. A computer readable storage medium having stored thereon a computer program, wherein the program when executed by a processor implements the intelligent tracking method of high maneuver targets in a complex clutter environment as claimed in any one of claims 1 to 8.

Description

Intelligent tracking method for high maneuvering target in complex clutter environment Technical Field The application belongs to the technical field of target tracking, and particularly relates to an intelligent tracking method for a high maneuvering target in a complex clutter environment. Background The radar target tracking technology is a core link in a modern radar system, maintains stable and accurate estimation of single or multiple target motion states in continuous observation, and provides support for situation awareness and decision. Conventional tracking methods are usually built on a framework of "filtering-prediction-correlation", and the performance depends largely on a priori modeling of the target motion law and accurate assumptions on observed noise statistics. In practice, this technology faces two increasingly serious challenges. First, the mobility of the target is significantly enhanced. The targets of modern high-speed aircrafts, unmanned planes, offshore quick boats and the like can execute high-mobility actions such as rapid steering, acceleration and deceleration, roundabout maneuver and the like, and the movement modes of the targets show strong non-linearity, non-stability and abrupt change characteristics. Traditional state estimation algorithms (such as Kalman filtering and expansion forms thereof) based on fixed models (such as uniform speed, uniform acceleration and coordinated turning models) are difficult to effectively match with the complex dynamics, prediction deviation accumulation is often caused by model mismatch, and finally tracking precision is reduced or even a target is lost. Secondly, the radar working environment is increasingly complex, especially in the sea detection scene of coasts, carrier-borne and the like. The background of sea surface, ground object and the like can generate strong clutter with time-varying, non-uniform and non-Gaussian characteristics. After radar detection, the clutter forms a large number of false alarm points similar to the real target in space and speed dimension. Traditional data correlation methods (e.g., nearest neighbor, joint probability data correlation, etc.) rely mostly on geometric distance or statistical likelihood between the target predicted position and the current measurement for matching. Under the condition of dense false alarm interference, the method has the limitations that firstly, the association calculated amount is increased in a combined explosive manner along with the number of the false alarms, the requirement on hardware calculation force is severe, the engineering realization cost is high, secondly, the false alarms with similar real targets and features are difficult to effectively distinguish by only relying on the association criterion of kinematic information, and error association and missed association are easy to generate, so that track confusion, false track breeding, real track breakage and tracking continuity are damaged. It is particularly critical that the two challenges described above do not exist in isolation, but rather are coupled to each other and worsen each other. The unpredictability of the motion trail of the high maneuvering target reduces the reliability of motion model prediction, so that the associated wave gate based on prediction is enlarged, more false alarms are further brought in, the associated ambiguity and the calculation burden are aggravated, and the increase of the false alarms in turn interferes with the accurate learning and estimation of the real motion mode of the high maneuvering target, so that a vicious circle is formed. Therefore, the prior art lacks an overall solution capable of cooperatively processing the high maneuvering characteristics and the complex clutter interference of the target, so that the tracking precision, the stability and the real-time performance of the radar system are difficult to guarantee in a complex actual combat environment. Disclosure of Invention The method aims to solve the problems of inaccurate prediction caused by mismatching of a motion model, and incorrect association, large calculated amount and track confusion caused by serious clutter interference when the traditional radar tracking method tracks a high maneuvering target in a complex clutter environment. By fusing the deep learning and optimization algorithm, the intelligent tracking method capable of improving the state prediction precision and the data association robustness simultaneously is provided, so that stable and accurate tracking of a high maneuvering target is realized. In order to achieve the technical purpose, the application adopts the following technical scheme: in one aspect of the present application, a method for intelligent tracking of a high maneuver target in a complex clutter environment is provided, comprising the steps of: S1, acquiring a high maneuvering target true value state after radar scanning And corresponding measurementPerforming differ