Application Research Of Machine Learning In Target Detection And Tracking

With the rapid development of modern electronic technology,the military application of radar is becoming more and more important.Radar target detection and tracking is a key factor affecting radar performance.Radar target detection extracts real targets from a large number of clutter and interference.The role of radar target tracking is to predict the position of the radar target at the next moment and determine the motion trajectory of the target.Target detection is the premise of target tracking.Only the accuracy of target detection can ensure the effectiveness of target tracking.Since the target of radar detection is in a complex electromagnetic environment,the radar echo will contain many unnecessary interference factors,which brings great difficulty to the radar target detection and tracking.Therefore,it is of great engineering significance to study how to accurately detect real targets and tracks.This paper uses machine learning to solve the problem of target detection and tracking in radar data processing,reducing false targets and false track rates.Mainly done the following works:1.A clutter recognition method based on machine learning is proposed.Based on the 6-dimensional features of radar target points,the K-nearest neighbor method,support vector machine,BP neural network and LSTM neural network are used to detect radar target points and clutter points.The clutter recognition rate is effectively improved,and the problem that the residual clutter points interfere with the target tracking after the target detection is greatly improved.2.A target track prediction method based on neural network is proposed.The neural network is used to predict the coordinate of the radar target to achieve accurate estimation of the radar target position.3.A data association method based on neural network is proposed.According to the predicted track,the possibility that the associated point is the target point on the track is evaluated,and the correct rate of the track correlation is effectively improved.4.An improved nonlinear filtering algorithm is proposed.The method uses the BP neural network to further correct the filtering error of the nonlinear filtering.By adding the filtered estimation value and the corrected filtering error value,a more accurate filtering value is obtained.It can effectively improve the filtering accuracy and improve the target tracking performance.On the one hand,this paper further distinguishes the target points and the clutter points after the target detection,which ensures the reliability of the target tracking;On the other hand,the proposed new data association algorithm and filtering algorithm effectively improve the performance of target tracking.The simulation experiments show that the proposed methods have many advantages compared with the traditional algorithms,and have certain reference value for the radar target tracking.