Magnetic Anomaly Target Motion State Detection Method Based on 3-D Convolutional Neural Network

Abstract

Magnetic anomaly detection (MAD) can be used to detect and track ferromagnetic targets in invisible environments. However, it is extremely challenging to calculate the target’s motion state information based on the passively detected magnetic field signals. To address this problem, we propose a magnetic anomaly target motion state detection method based on 3-D convolutional neural network (3D CNN). The method utilizes a magnetic field sensor array to collect and visualize magnetic signals based on a MAD model for moving targets. The processed magnetic field signals are imaged and then arranged in time sequence to generate a motion flow. After standardizing the images, they are input into the 3D CNN to detect and classify motion of interest. Experimental validation was performed using a semi-real dataset with 16 target movements of interest and a control group without target movements. The experimental results show that the accuracy of the proposed method can reach 0.9 on average, which can provide a theoretical basis and method reference for the motion state recognition of ferromagnetic targets.