Research On Magnetic Anomaly Target Tracking Algorithm Based On Reinforcement Learning

Under the influence of the magnetic field,ferromagnetic materials in the geomagnetic field will cause weak abnormal changes in the surrounding geomagnetic field.Airborne magnetic anomaly detection technology can retrieve the position information of magnetic objects by capturing such magnetic anomaly changes.It plays an important role in reconnaissance and target tracking.Among them,using the airborne magnetic exploration platform to track the dynamic magnetic anomaly target has become a research hotspot and difficulty.At present,the actual tracking strategies mainly include spiral,alfalfa leaf,survey line and other classical trajectories,which are decided by the commanders based on their own experience.The strategy is highly dependent on the commander and is greatly influenced by subjectivity.At the same time,the tracking decision is not flexible enough and there are many redundant paths,which is easy to make the target lose.In order to promote the development of aeromagnetic tracking intelligent decision-making and improve the efficiency of magnetic anomaly target tracking,this paper studies magnetic anomaly target tracking based on reinforcement learning algorithm.The main research results of this paper are as follows:Firstly,this paper analyzes the basic principle of airborne magnetic prospecting technology and the tracking characteristics of airborne magnetic prospecting platform,and realizes the modeling and simulation platform of magnetic anomaly target tracking task environment.The basic motion model and magnetic anomaly signal detection model are modeled respectively,and the actual loss in practical application is analyzed to improve the model design.Based on the model,the simulation platform is developed to establish an experimental environment for the research of magnetic anomaly target tracking algorithm.At the same time,the evaluation index of magnetic anomaly target tracking task is defined,which provides a reference for the performance evaluation of tracking algorithm.Then,this paper studies the reinforcement learning magnetic anomaly target tracking algorithm under the condition that the target motion state is known,and realizes the complete design of the reinforcement learning tracking algorithm.Based on the dynamic target distribution probability model,the target potential area prediction theory is proposed to solve the sparse reward problem in magnetic anomaly target tracking.The performance of the algorithm is verified by the simulation platform.Then,aiming at the unknown target motion state,this paper further optimizes the magnetic anomaly target tracking algorithm,proposes a two-stage tracking framework and a target motion state prediction method,and realizes the reinforcement learning magnetic anomaly target tracking algorithm in this task environment.A simulation experiment is designed to evaluate the performance of the algorithm by comparing it with the improved spiral tracking scheme.Finally,the conclusion is drawn that the magnetic anomaly tracking algorithm based on reinforcement learning can effectively improve the intelligence of aeromagnetic tracking decision-making and the tracking efficiency of dynamic magnetic anomaly targets.