A Moving Ferromagnetic Target Localization Method Based on YOLO-V3

Moving ferromagnetic target localization and tracking offers a wide spectrum of uses in the military and medical areas. We present an automatic localization method based on YOLO- V3 to achieve fast recognition and precise localization of moving ferromagnetic targets. On the basis of ensuring efficiency and accuracy, this method has a strong anti-interference ability. To begin, the darknet-53 network extracts features from magnetic anomaly pictures taken at any time, and magnetic target feature maps are constructed at various scales to predict magnetic target positions. The boundary frame with varying sizes is then constructed automatically based on the real position of each magnetic target as the geometric center of the boundary frame. Finally, interactive datasets are created using COMSOL and MATLAB, and the proposed approach is compared to the Nara localization method in a variety of conditions, including single magnetic source, double magnetic source, noise, and noise free. The proposed method's recognition accuracy is over 95%, the average positioning error is around 1.5cm, and the positioning time is about 0.031s, according to experimental data. In terms of accuracy and robustness, the proposed method outperforms the Nara localization method, and it can also be used to monitor and localize ferromagnetic objects in real time, even when the data is noisy.