Underwater Target Detection Based on Magnetic Gradient Tensor

Guoqing Yang,Fanzong Zeng,Hao Li,Zhenlong Xu,Jiafeng Han,Yipeng Meng,Xueting Zhang,Zefei Zhu

DOI: https://doi.org/10.1109/OCEANS51537.2024.10706271

2024-04-15

Oceans

Abstract:The Magnetic Anomaly Detection technique can be applied to locate and identify underwater small-scale magnetic targets. Compared to total magnetic field intensity and magnetic vectors, the magnetic gradient tensor has stronger anti-interference capabilities and provides higher-dimensional information. Utilizing pseudo-color imaging of the magnetic gradient tensor components combined with deep learning can reduce manual intervention and improve identification efficiency and accuracy. However, issues such as sparse measured data and poor magnetic imaging effects lead to poor identification accuracy. To address this, this paper adopts the Kriging interpolation method to increase magnetic gradient tensor data density and proposes an enhanced magnetic anomaly imaging method based on the boundary features of the target magnetic source. Additionally, Coordinate Attention is added to the neural network structure for improvements in identification accuracy. The analysis of finite element simulation data for common underwater magnetic targets demonstrates that the Kriging interpolation method results in smaller errors and higher accuracy compared to traditional linear interpolation methods. The enhanced magnetic anomaly imaging method ensures significantly clearer boundaries while maintaining the intensity of the target magnetic field. Furthermore, the addition of a coordinate attention to the detection model resulted in a 2% improvement in the mapping accuracy. In summary, this work provides a novel solution for the detection of underwater small-scale magnetic targets with different positions, intensities, and shapes.

Environmental Science,Engineering