Research On Data Processing Method Of Radar Targets Based On Long And Short-term Memory

Radar target data processing is a reprocessing process established after radar signal processing.The input is the points reported by the signal detectors,and the output is the track information of the target.The target and clutter parameter estimation contained in the point information is used to find the target and predict target's future kinematics information are its main tasks which corresponds to the three processes: track initiation,target position prediction,and track maintenance.This article focuses on the use of machine learning methods to achieve target data processing when the spatial state of motion equation is nonlinear.Firstly,the problem of track initiation is transformed into a binary classification problem of whether or not to initiate.Based on the kinematic information of the real target,the distinguishing features are constructed,and the random forest network is trained by using these information,then the trained random forest network is used to judge whether the combination of points meets the track initiation conditions.In view of the possible missing points,polynomial fitting method is used to fill in the missing points.Secondly,aiming at the problem of target motion information prediction,the long-term and short-term memory network is proposed to predict the state parameters of the target.Bypassing the establishment of non-linear motion model,through the time trend of the data itself,the radar target state is predicted by the artificial intelligence time series prediction method of long and short-term memory network.In order to ensure the scale invariance of the test samples to the network,the scale transformation of the training samples is needed.Finally,for track maintenance,an improved fuzzy C-means clustering method is used.The relevant gate is determined by the predicted value of the target state.Assuming that all the measurements in the gate may be derived from the target,a clustering algorithm is used to calculate the probability that each measurement originates from the radar target and the cluster center is obtained.The cluster center is considered to be the most likely location of the radar target.The results of simulation experiments show that the method studied in this paper can effectively deal with spatial target data processing problems using artificial intelligence without establishing a motion model.