基于ResNet-GRU网络的弱磁异常探测方法  

Weak Magnetic Anomaly Detection Method Using ResNet-GRU Neural Network

作　　者：樊黎明 雷波 秦梦辉[2] 魏凡程 FAN Liming;LEI Bo;QIN Menghui;WEI Fancheng(Qingdao Research Institute,Northwestern Polytechnical University,Qingdao 266200;School of Marine Science and Technology,Northwestern Polytechnical University,Xi’an 710072)

机构地区：[1]西北工业大学青岛研究院,青岛266200 [2]西北工业大学航海学院,西安710072

出　　处：《导航与控制》2024年第5期121-128,共8页Navigation and Control

基　　金：中央高校基本科研业务费专项资金(编号:D5000220158);陕西省自然科学基础研究计划(编号:2024JC-YBQN-0684)。

摘　　要：在低信噪比下,目标产生的磁异常通常被磁噪声掩埋,这导致传统的磁异常方法检测性能下降。为了提高低信噪比下弱磁异常检测的性能,提出了基于ResNet-GRU网络的弱磁异常探测方法。本方法采用基于ResNet的Conv1D模块和GRU模块提取磁异常信号特征信息的多维特征,通过多特征融合实现磁异常信号的探测。为了训练模型,构建了实测的磁异常数据集,含有正样本数量为8646,负样本数据量为8431,利用该数据集进行模型训练。实验结果表明,所提方法的探测模型在测试集的精确率、准确率和F1值分别为90.39%、91.33%和90.18%,优于全连接神经网络模型和一维卷积神经网络模型。所提出的方法在低信噪比情况下具有良好的弱磁异常探测性能。In low signal-to-noise ratio(SNR)situations,magnetic anomaly generated by magnetic target is usually buried in the magnetic noise,leading to a decline in the detection performance of traditional magnetic anomaly methods.To improve the detection performance of weak magnetic anomaly under low SNR,a weak magnetic anomaly detection method using ResNet-GRU network is presented in this paper.In this method,the Conv1D modules based on ResNet and the GRU modules are employed to extract multidimensional features from magnetic anomaly signals,enabling the detection of such signals through the fusion of multiple features.To train the model,a real-world magnetic anomaly dataset is constructed,consisting of 8646 positive samples and 8431 negative samples.Experimental results demonstrate that proposed method using ResNet-GRU has an accuracy of 90.39%,a precision of 91.33%,and an F1 score of 90.18%on the test set,outperforming the performance of fully connected neural network model and one-dimensional convolutional neural network model.The proposed method has good detection performance of weak magnetic anomaly under low SNR.

关 键 词：磁异常探测 深度学习 特征提取 门控递归单元 

分 类 号：P631.2[天文地球—地质矿产勘探]