Magnetic Anomaly Detection Network With Adaptive Time–Frequency Feature Expression
Yizhen Wang, Qi Han, Dechen Zhan, Qiong Li
Abstract
Magnetic anomaly detection (MAD) is a method that uses magnetometers to find hidden ferromagnetic objects based on variations in magnetic field signals. Many methods use time–frequency fusion features to detect target signals, but these features are extracted using unlearnable methods in the preprocessing. This article proposes a network with adaptive time–frequency feature expression to detect magnetic anomalies. The network adaptively learns the time–frequency features of magnetic anomaly signal through optimization, and in the feature fusion step, it adaptively selects key features for classification. In experiments, we compared our proposed method with the state-of-the-art (SOTA) time–frequency fusion detection method in simulation, semireal, and real datasets. Results show that our method has better detection performance, with high accuracy (ACC), detection rate (DR), and area under receiver operating characteristic curve (AUC). In real datasets, the proposed method outperforms other methods in low signal-to-noise ratio (SNR) cases and has a low false alarm rate (FAR), demonstrating its good comprehensive detection ability and robustness.