基于ResNet的面向对象矿区占地信息提取研究  

Research on Mining Area Land Occupation Information ExtractionBased on ResNet and Object-oriented Method

作　　者：甄娜 陈涛[2] 霍光杰 李小芳 ZHEN Na;CHEN Tao;HUO Guangjie;LI Xiaofang(Henan Key Laboratory of Geological Disaster Prevention and Control,Natural Resources Monitoring Institute of Henan Province,Zhengzhou 450000,China;Institute of Geophysics and Geomatics,China University of Geosciences,Wuhan 430074,China;Henan Banger Environmental Protection Technology Co.Ltd.,Zhengzhou 450000,China)

机构地区：[1]河南省自然资源监测院,河南省地质灾害防治重点实验室,郑州450000 [2]中国地质大学(武汉)地球物理与空间信息学院,武汉430074 [3]河南邦尔环保科技有限公司,郑州450000

出　　处：《河南科学》2023年第4期619-624,共6页Henan Science

基　　金：河南省财政项目(豫财预﹝2014﹞134号,豫财预﹝2015﹞128号,豫财环﹝2016﹞44号);国家自然科学基金(62071439)。

摘　　要：利用高分二号影像结合面向对象方法展开矿区占地信息提取研究,采用面积比均值法确定分割尺度进行多尺度分割获取对象,基于空间优化工具选取特征后标记样本,其中训练集、测试集和验证集比例为3∶1.将样本集在ResNet模型中训练,应用于全部对象,并与CNN模型进行对比.结果表明,面向对象方法结合ResNet模型进行矿区占地信息提取总体精度为91.41%,Kappa系数为0.89,优于CNN方法.该方法适用于以露天采场和矿堆为主的矿区环境,可以为后续的矿区环境治理工作提供有效的技术支持.In this paper,the mining area information extraction research is carried out by combining Gaofen-2 image with object-oriented method,and the area ratio mean method is used to determine the segmentation scale for multi-scale segmentation to obtain objects.The samples are labeled after selecting features based on the spatial optimization tool,and the ratio of training set,test set and verification set is 3∶1.The sample set is trained in the ResNet model,applied to all objects,and compared to the CNN model.The results show that the overall accuracy of mining land area information extraction by object-oriented method combined with ResNet model is 91.41%,and the Kappa coefficient is 0.89,which is better than that of CNN method.The proposed method is suitable for the mining environment dominated by open-pit and mine piles,and can provide effective technical support for subsequent environmental governance work in mining areas.

关 键 词：矿区占地信息 面向对象方法 ResNet 信息提取 遥感图像 

分 类 号：TP753[自动化与计算机技术—检测技术与自动化装置]