Multi-Region Target Track Association Technique Based On Machine Learning

Hydroacoustic target trajectory correlation has always been an important part of sonar technology and one of the research hotspots.As the main body of marine activities,ship target tracking and situational estimation can effectively understand the ship navigation dynamics and movement trend.To improve the ship target detection and tracking capability is the top priority for the construction of a strong maritime nation.In recent years,with the trend of increasingly complex marine battlefield background,target tracking in a single area has gradually failed to meet the demand.This paper forms a multi-region target track association system by combining passive sonar detection nodes in multiple regions,and by processing the measurement and feature information obtained from detection in each region,the target tracks appearing in each region are successively associated,and finally the complete target situational information between multiple regions is obtained,realizing large-scene monitoring,multi-target tracking in the region,and joint vigilance between regions under a unified situation.The article firstly studies the model of passive sonar detection target,sets up multiple motion models of the target,establishes the physical model of passive sonar detection in the complex ocean background,and obtains the measurement information and feature information of passive sonar detection on the target in each region under the unified motion posture of the target.Secondly,in order to improve the accuracy of target tracking in underwater high interference density environment and lack of relevant a priori knowledge background of target and interference in each region,based on joint probability density correlation algorithm,the automatic tracking of multi-target trajectory applicable to the complex situation where the number of targets changes and the target state is unknown is realized.For the phenomenon of strong target occlusion in time and space,the target classification results appearing in each region are obtained based on the machine learning algorithm,and the complete track information of the target is further correlated with the multi-region tracks.Finally,the method of interrupted trajectory succession association based on generative adversarial network at the edge of the region is studied.To address the problems of large measurement error,incomplete information and discontinuous observation in the edge region,the generator of the generative adversarial network is trained with the noisy interrupted and continuous trajectory image datasets,and then the discriminator of the generative adversarial network is used to judge the authenticity of the generated continuous trajectory to achieve the interrupted trajectory succession association of the target at the edge of the region.The article constructs a physical model of multi-area passive sonar array detection,and conducts a study on the interrupted trajectory association between multi-areas,dealing with the trajectory association under the strong occlusion between areas and the missing trajectory of the target at the edge of the area.