基于通道注意力机制的智能地质填图  

Intelligent Geological Mapping Based on Channel Attention Mechanism

作　　者：陶从咏 郭晓宁[2] 于玉帅[3] 许笑玮 张世晖[5] 周文孝[6] 骆满生[4] 朱云海 徐亚东[4,6] Tao Congyong;Guo Xiaoning;Yu Yushuai;Xu Xiaowei;Zhang Shihui;Zhou Wenxiao;Luo Mansheng;Zhu Yunhai;Xu Yadong(School of Mathematics and Physics,China University of Geosciences,Wuhan Hubei 430074,China;Library,China University of Geosciences,Wuhan Hubei 430074,China;Wuhan Geological Survey Center,China Geological Survey,Wuhan Hubei 430205,China;State Key Laboratory of Biogeology and Environmental Geology,China University of Geosciences,Wuhan Hubei 430074,China;School of Geophysics and Geomatics,China University of Geosciences,Wuhan Hubei 430074,China;Institute of Geological Survey,China University of Geosciences,Wuhan Hubei 430074,China)

机构地区：[1]中国地质大学数学与物理学院,湖北武汉430074 [2]中国地质大学图书馆,湖北武汉430074 [3]中国地质调查局武汉地质调查中心,湖北武汉430205 [4]中国地质大学生物地质与环境地质国家重点实验室,湖北武汉430074 [5]中国地质大学地球物理与空间信息学院,湖北武汉430074 [6]中国地质大学地质调查研究院,湖北武汉430074

出　　处：《工程地球物理学报》2023年第3期427-436,共10页Chinese Journal of Engineering Geophysics

基　　金：中国地质调查局地质调查项目(编号:DD20221645,DD20190811);国家自然科学基金项目(编号:41302279)。

摘　　要：进入新世纪,科技的发展造就了大数据的爆发式增长,这为基于深度学习方法来研究地质学问题奠定了基础。卷积神经网络已被用于地质填图,但卷积操作关注的是数据空间维度的特征信息,无法建模不同通道维度之间的依赖关系。为了发掘不同通道的输入数据和特征图之间的关联性,提升智能地质填图的效果,本文在全卷积神经网络Unet中引入通道注意力模块--挤压-激励模块(Squeeze and Excitation Block,SE Block),提出了一种新网络SE-Unet,并将该网络应用于湖南省鲤鱼塘地区的1∶5万智能地质填图。实验结果表明,相比于Unet,SE-Unet智能地质填图的总体精确度由81.58%提高到了83.72%,可视化结果显示,两种原来难以识别的地质单元被大致识别出来。这验证了通道注意力机制能够提升网络的学习和表征能力,也说明了本方法对于提升智能地质填图效果的可行性与有效性。The development of technology has contributed to the explosive growth of geological big data,which has laid the foundation for deep learning methods to investigate geological problems.Convolutional neural networks have been used by many scholars to study geological mapping.However,the convolutional operation focuses on the feature information of data spatial dimensions and cannot model the dependencies between different channel dimensions.In order to explore the correlation between input data and feature maps of different channels and improve the effect of intelligent geological mapping,this paper introduces the squeeze and excitation block(SE Block),a channel attention module,into the fully convolutional neural network Unet,proposes a new network SE-Unet,which is applied to the 1∶50,000 intelligent geological mapping in the Liyutang area of Hunan province.The experimental results show that compared with Unet,the overall accuracy of SE-Unet for geological mapping is improved from 81.58%to 83.72%,and the visualization results demonstrate that two geological units,which were previously challenging to distinguish,have been roughly identified.This verifies that the channel attention mechanism can improve the learning and characterization ability of the network,and also illustrates the feasibility and effectiveness of this method for improving the effect of intelligent geological mapping.

关 键 词：通道注意力机制 智能地质填图 卷积神经网络 

分 类 号：P623[天文地球—地质矿产勘探]