Research On Multi-model Maneuvering Target Tracking Method Combined With Machine Learning

In the field of target tracking,maneuvering target tracking is a difficult point,which has important research significance and application value.Many experts and scholars at home and abroad have carried out research on the problem of maneuvering target tracking and achieved rich results.In recent years,with the continuous improvement of the level of science and technology,the speed,maneuverability and stealth capabilities of the tracked target have been significantly improved,which has brought new challenges to the task of maneuvering target tracking.The traditional single-model tracking algorithm works well when tracking nonmaneuvering targets.When the target’s maneuvering mode changes over time,the single-model tracking algorithm will lead to a decrease in tracking accuracy due to model mismatch.The multi-model tracking technology describes the motion mode of the target through a set of motion models,which greatly improves the tracking accuracy of maneuvering targets.Nevertheless,there are still many problems in the traditional maneuvering target tracking technology in terms of fine estimation of model probability,refinement of effective model set and optimization of observation period.To this end,this paper conducts research on single-sensor maneuvering target tracking methods for the above three problems,introduces reinforcement learning and regression learning techniques on the basis of traditional maneuvering target tracking techniques,and constructs a maneuvering target tracking method combined with machine learning.The main contents of this paper are as follows:1)IMM algorithm is a widely used fixed structure multi model algorithm.In order to solve the problem of unnecessary competition between models caused by excessive prior model set in fixed structure multi model algorithm and the decrease of tracking accuracy,neural network is used to optimize the IMM algorithm.After defining the input vector of the neural network,the neural network is used to estimate the probability of each model in the model set,and the estimated probability of each model output by the neural network and the estimated probability of each model output by the IMM are fused to improve the tracking accuracy.Simulation results show that the overall tracking accuracy of the proposed algorithm is significantly improved compared with the traditional IMM algorithm.2)Aiming at the problem that the model set size is difficult to determine in the MSA mechanism of VSMM algorithm,a VSMM algorithm based on Monte Carlo learning was proposed.Firstly,the state space,decision space and revenue are defined for maneuvering target tracking problem,then the model set size is learned by Monte Carlo learning,and finally the adaptive adjustment of model set size is realized by using the learning results in real time.Simulation results show that the proposed algorithm can dynamically change the number of active effective models when the target moves in different motion modes,and improve the tracking accuracy of the algorithm for maneuvering target with only a low computational cost.3)In the traditional sampling period adaptive algorithm based on the prediction error covariance threshold method,heuristic search is usually used to determine the next sampling interval of the sensor.When the scale of the sampling interval set is relatively large,this method takes up a large amount of computing resources and reduces the real-time performance of the algorithm.To solve these problems,a Monte Carlo learning method is proposed instead of heuristic search to complete the task of sampling interval confirmation.After completing the definition of state space,decision space,revenue and learning the expected revenue of "State-Action",the learning results are applied in real time.The simulation results show that the proposed algorithm meets the sensor sampling cycle adaptation,reduces the computing resources and meets the expected conditions.