2D CNN-based multi-feature fusion detection method for the magnetic anomaly generated by submarine wake

Abstract

A Two-Dimensional Convolutional Neural Network (2D-CNN)-based multi-feature fusion detection method is proposed to improve the detection performance of the submarine wake magnetic anomaly in view of its characteristics of low frequency and low signal-to-noise ratio (SNR). The method involves pre-processing the original signal by using the Savitzky-Golay (S–G) filter, followed by Residual Structure processing to extract the time-domain information, FFT to extract the frequency domain information and Minimum-Entropy Filter (MEF) for noise analysis. The 2D-CNN model with three processing branches is utilised for further feature extraction and signal judgement. To train the method, Simulated target signal dataset is obtained through the submarine wake magnetic anomaly simulation model, and the simulated original signal is acquired by stacking measured noise. The proposed method exhibits great detection performance for signals with different Signal-to-Noise Ratios (SNRs) and various types of noise, achieving a recognition accuracy of 90 % for signals with SNRs above -10dB. The theoretical detection range of the submarine has been increased to over 1 km, outperforming similar neural networks based on magnetic dipole models.