Deep Learning Network for Magnetic Anomaly Target Signal Denoising

Magnetic anomaly target detection techniques are used in many areas, such as detection of buried objects and exploration of seabed tectonics. Currently, the prevailing orthogonal basis functions detection method is not satisfactory at low signal-to-noise ratios. In order to obtain better detection results, this paper proposes a denoising method based on deep learning network. The input of the denoising network is the time-frequency spectra obtained by the continuous wavelet transform, which has richer features. The denoising network uses the U-Net network as the main framework, and some inception and dense-inception blocks are added to improve the denoising performance. Results using real experimental data verify that the proposed method can remove most magnetic noises and obtain clean target signals, which greatly reduces the false alarm in detection and facilitates the in-depth analysis of magnetic anomaly targets.