Research On Underwater Track Fusion Based On Neural Network

In the modern marine combat system,the single-station combat method lacks effective information interaction,and the amount of information that each single node in the combat system can obtain is small,the reliability is low,and it is difficult to perform effective position estimation.Therefore,in order to make full use of the multi-node positioning information It is of great significance to realize the effective interaction and fusion of the positioning information of each node under the system.Based on the distributed information fusion structure,this paper studies the process of track fusion processing of the underwater acoustic local track data sent from each node to the fusion center.This paper firstly studies the target motion model modeling and analyzes the main interference factors of the hydroacoustic positioning error to model and generate reasonable local track data,and then studies two kinds of time based on the least square criterion and based on the interpolation and extrapolation criterion.The registration algorithm,as well as the two spatial registration algorithms based on the geocentric coordinate system and the Gauss-Krüger projection,preprocess the local track data for spatio-temporal registration,which is convenient for the next step of track association.Then,in a strong interference environment,this paper studies the correlation performance of neural network applications in track correlation,and the effects of convolutional neural networks(Convolutional Neural Networks,CNN)and Long Short-Term Memory(LSTM)The track correlation model is researched.In order to give full play to the advantages of the local track in the use of information before and after,this paper studies the improved track correlation model using Bi-directional Long Short-Term Memory(BiLSTM),and The convolutional neural network is further combined,and the correlation model of each neural network is compared and analyzed through simulation.The improved track correlation accuracy rate of BiLSTM has been improved.Finally,in order to reduce the system track error,this paper analyzes the fusion ability of neural network in the track fusion algorithm,studies a variety of track fusion algorithms,and performs the fusion estimation of the local track based on the results of the track correlation,and finally obtains the global estimate.Then analyze and verify through simulation and lake test data.