Research On Target Track Initiation And Recognition Based On Deep Learning

With the increasingly congested airspace environment,research on track recognition and initiation based on massive flight target track data will help to deeply understand the airspace situation and provide effective support for airspace supervision.As one of the main means to obtain flight target track data,radar plays an incomparable role in track recognition and initiation research.However,traditional radar target track recognition and initiation algorithms have been difficult to cope with the current diversified,complex,and big data radar detection environment.In this paper,we conduct research on the recognition and initiation of radar target track based on deep learning technology.The main contents of the thesis are as follows.(1)A target track initiation algorithm based on deep learning and temporal-spatial characteristics of radar measurement is proposed.Most of the existing track initiation studies only consider one of real-time or initiation rate,and it is difficult to complete a fast and accurate track initiation in a strong clutter environment.In response to the above problems,we convert the track initiation problem into the true and false track identification problem for the radar measurement combination,and comprehensively considers the influence of the time and space dimension information of the measurement combination on the true and false track identification.We propose a target track initiation algorithm DLTS based on Deep Learning and Temporal-Spatial characteristics of radar measurement.The algorithm first selects the candidate set from the measurement combination set,extracts the temporal change vector and spatial distribution vector of the candidate set measurement combination,and uses it as the input of the gated recurrent unit and one-dimensional convolutional neural network hybrid model to obtain the time dimension characteristic and the space dimension characteristic,then merge them to obtain the space-time feature.Next,the self-attention mechanism is used to assign different weights to different dimensions of temporal-spatial characteristics,then input them into the classifier to realize the identification of true and false tracks,thereby completing the track initiation.Experimental results show that the DLTS algorithm effectively improves the real track initiation rate,false track initiation rate and average initiation time,and can achieve fast and accurate track initiation in a strong clutter environment.(2)A radar target track recognition algorithm based on semi-supervised generative adversarial networks is proposed.Existing target track recognition research lacks the consideration of outliers which are common in radar measurement data,and does not consider the difficulty of labeling massive track data.In response to the above problems,we proposes a Semi-Supervised Target Track recognition algorithm SSTT based on a semi-supervised generative adversarial networks.The algorithm first uses the identification and elimination of outliers in the target track data,and fills in the missing points.Next,a strong recognition flight feature is selected from the basic flight features of the target track.At the same time,in order to improve the robustness of the algorithm,advanced flight feature is extracted from the basic flight features.Then,the strong recognition flight feature and the advanced flight feature are combined to obtain a strong recognition flight feature combination,and input to the constructed semi-supervised generative adversarial networks model to complete the classification and recognition of the track type.Simulation experiments show that the SSTT algorithm can accurately complete the classification and recognition of target track types with only a small amount of labeled data,and effectively improve the accuracy,precision and recall rate of target track recognition.