Research On Moving Magnetic Target Localization And Identification Method Based On Magnetic Anomaly

Magnetic anomalies will develop in ferromagnetic items as a result of the geomagnetic field’s presence.Moving magnetic target location and recognition,a crucial subfield of magnetic anomaly detection,is well-suited for usage in the medical,transportation,military,and other sectors due to its high effectiveness,stability,and good camouflage.The real-time and precision demands of moving magnetic objects cannot be satisfied by the present positioning and identification techniques due to flaws including sluggish computing speed and weak resilience.This work examines the magnetic target localization approach and the magnetic target identification method,both of which are based on the magnetic dipole model.The following are the primary research findings:In terms of the hardware system design,this study builds a 3 by 3 magnetic sensor array using an RM3100 magnetic sensor chip and implements communication with the upper computer using a 485 bus as the foundation for location identification algorithm research.Also,an induction magnetometer based on the NSGA-II(Non-dominated Sorting Genetic Algorithms II)algorithm was created,which offered a hardware platform for the investigation of moving magnetic target detection from various angles.This work investigates two positioning algorithms: a two-dimensional visual localization approach based on YOLOV3(You Only Look Once V3)and a sixdimensional hybrid localization method based on enhanced PSO-LM(Particle Swarm Optimization,Levenberg-Marquardt).An effective,precise,and real-time deep learning object identification technique is YOLOV3.The neural network is trained on the distribution map of the magnetic anomaly field produced by the finite element analysis program COMSOL,and the neural network is then used to assess the performance of the localization technique by varying the signal-to-noise ratio.According to the simulation findings,the YOLOV3 positioning method operates in less than 0.1 seconds with a maximum positioning error of 0.011 m in a noise-free environment.It is vital to get information on the location and attitude of the magnetic source since only two-dimensional position data can satisfy the criteria of usage in a particular situation.The positioning issue is converted into a least squares problem when computing the position and pose information of a magnetic target.The upgraded PSO-LM algorithm is ultimately proven to be a six-dimensional positioning approach for moving magnetic targets by contrasting the solution capabilities of the LM and PSO algorithms for the magnetic source placement issue.The simulation experiment investigates the capacity to locate objects under various noise levels and object motions.The experimental findings indicate that,when there is no noise,the algorithm’s calculation time is less than 0.3 seconds,its positioning error is 0.003 meters,its orientation error is 0.006 radians,and its magnetic moment error is 0.31%.It is a quick and accurate placement technique.Also,in the measured experiment,the magnetic target was effectively found in both the static and moving states.This work examines shifting magnetic target identification from one and two dimensions in terms of recognition techniques.The link between moving magnetic objects and induced voltage is examined using the induction magnetometer’s finite element simulation model in terms of one-dimensional recognition,and the theoretical analysis is confirmed in a field experiment.Using an improved induction magnetometer,it is possible to determine whether moving objects are present and their speed,with a calculation error of less than 1 cm/s.Due to the powerful representation learning capabilities of deep neural networks,this research decides to adapt the issue of magnetic target identification into semantic segmentation and case segmentation in terms of twodimensional recognition.Via UNet and YOLACT(You Only Look At Coefficien Ts)networks,the magnetic target’s shape may be recognized,and this has been tested on hardware.Magnetic source models of various forms were constructed using the COMSOL program,and the resulting picture samples served as training data for two neural networks.The simulation results demonstrate that both models are capable of realizing magnetic objects of various forms.Yet,the YOLACT network has a greater identification accuracy when the magnetic field is hampered by the outside environment.Moreover,the YOLACT network has a faster processing speed,and the algorithm’s reasoning time is under 0.2 seconds.Furthermore,actual trials are used to confirm the suggested method’s viability.