基于多任务神经网络的水下震源定位方法研究
Research on Underwater Source Localization Method Based on Multi-task Neural Networks
杨丽燕 1王黎明 1韩星程 1武国强 2王鸿儒 3马文3
作者信息
- 1. 中北大学信息探测与处理山西省重点实验室 太原 030051
- 2. 太原重工股份有限公司 太原 030051
- 3. 山西太重数智科技股份有限公司 太原 030051
- 折叠
摘要
针对传统水下被动定位算法传感器阵列体积大、布设难、环境失配等问题,提出一种基于多任务神经网络的水下震源定位方法.该方法通过仿真浅海声场数据集,利用声信号到达各个测量传感器的相对时间差,结合深度学习方法,在多任务卷积神经网络MTL-CNN(Multi-task Convolutional Neural Network)和Attention-UNet结构基础上,设计MTL-Atten-tion-UNet神经网络模型,对水下震源的距离和深度进行联合估计.仿真结果表明:用MTL-Attention-UNet模型对水下震源进行定位的平均绝对误差比MTL-CNN网络模型小,定位性能更好.
Abstract
In order to solve the problems of large sensor array,difficult deployment and environmental mismatch of traditional underwater passive positioning algorithm,a method of underwater source localization based on multi-task neural network is pro-posed.By simulating the shallow sea acoustic field dataset,using the relative time difference of the acoustic signal to reach each measurement sensor,combined with the deep learning method,the MTL-Attention-UNet neural network model is designed on the basis of the multi-task convolutional neural network MTL-CNN(Multi-task Convolutional Neural Network)and Attention-UNet structure,and the distance and depth of the underwater seismic source are jointly estimated.The simulation results show that the av-erage absolute error of positioning the underwater source by MTL-Attention-UNet model is smaller than that of the MTL-CNN net-work model,and the positioning performance is better.
关键词
震源定位/神经网络/平均绝对误差Key words
source localization/neural networks/average absolute error引用本文复制引用
基金项目
国家自然科学基金青年基金(62203405)
山西省重点研发计划(2022ZDYF079)
山西省应用基础研究计划(20210302124545)
出版年
2024
国家科技期刊平台
NSTL
万方数据