摘要
准确、高效地识别岩石岩性是大数据时代地球科学研究的必然趋势和发展方向。传统岩石岩性识别方法多依赖人工判别,对相关知识与判别经验的要求很高。针对这一问题,该文提出一种基于Xception网络的自动化岩石图像分类方法,将InceptionV3网络中的卷积操作替换为深度可分离卷积模块,同时引入残差连接机制以大量减少模型参数与计算量,然后结合迁移学习思想提高图像分类准确率。选取嵊州地质调研中人工采集的10类岩石样本图像构建岩石图像数据集进行验证,结果表明,Xception网络模型对岩石岩性识别的准确率达86%,比其他主流的岩石图像分类模型的识别精度更高。
Accurate and efficient identification of rock lithology is the inevitable trend and development direction of geoscience research in the era of big data.Traditional rock lithology identification methods mostly rely on manual identification,which have high requirements for relevant knowledge and identification experience.This paper proposes an automated rock image classification method based on Xception network,combining deep learning technology and transfer learning to achieve the purpose of accurately and efficiently identifying rock lithology for rock images.This method replaces the convolution operation in the InceptionV3 network with depth wise separable convolution.At the same time,the method introduces a mechanism of residual connection and transfer learning,which can reduce a large number of model parameters and calculations and improve the accuracy of image classification.10 types of rock sample images manually collected in the Shengzhou geological survey were selected to construct a rock image data set for experimental analysis by means of image flipping,translation,rotation and noise addition.The results show that the rock recognition accuracy of the Xception network model reaches 86%,which is higher than other mainstream rock image classification models.
作者
谭永健
田苗
徐德馨
盛冠群
马凯
邱芹军
潘声勇
TAN Yong-jian;TIAN Miao;XU De-xin;SHENG Guan-qun;MA Kai;QIU Qin-jun;PAN Sheng-yong(Hubei Key Laboratory of Intelligent Vision Based Monitoring for Hydroelectric/College of Computer and Information Technology,China Three Gorges University,Yichang 443002;Wuhan Geomatics Institute,Wuhan 430074;School of Computer Science,China University of Geosciences,Wuhan 430074;Wuhan Zondy Cyber Science & Technology Co.,Ltd.,Wuhan 430074,China)
出处
《地理与地理信息科学》
CSCD
北大核心
2022年第3期17-22,共6页
Geography and Geo-Information Science
基金
国家自然科学基金原创性探索项目“地球科学知识图谱表示模式与群智协同构建”(42050101)
中国博士后科学基金项目(2021M702991)
国家自然科学基金项目“基于多模态数据理解及融合的三维地质模型构建方法研究”(41871311)
武汉市多要素城市地质调查示范项目(WHDYS-2020-004)。
关键词
岩石图像
岩性识别
迁移学习
深度可分离卷积
rock image
lithology identification
transfer learning
depth wise separable convolution
作者简介
谭永健(1998-),男,硕士研究生,主要从事深度学习与地学知识图谱研究;通讯作者:马凯,E-mail:makai@ctgu.edu.cn。
