基于CNN的空间数据自适应切分算法  

An Adaptive Segmentation Algorithm for Spatial Data based on CNN

作　　者：魏海涛[1] 杜云艳 张佳丽 孙璐瑶 WEI Haitao;DU Yunyan;ZHANG Jiali;SUN Luyao(College of Resources,Shandong University of Science and Technology,Qingdao 271019,China;State Key Laboratory of Resources&Environmental Information Systems,Institute of Geographic Sciences and Natural Resources Research,Chinese Academy of Sciences,Beijing 100101,China)

机构地区：[1]山东科技大学资源学院,青岛271019 [2]中国科学院地理科学与资源研究所资源与环境信息系统国家重点实验室,北京100101

出　　处：《地球信息科学学报》2022年第6期1099-1106,共8页Journal of Geo-information Science

基　　金：国家重点研发计划项目(2017YFB0503605);山东科技大学人才引进科研启动基金项目(2017RCJJ081)。

摘　　要：针对空间数据划分方法无法自适应选择问题,本文提出了基于CNN的数据自适应划分算法(Adaptive Partition Algorithm for Space Vector Data-Convolutional Neural Networks,SVDAP-CNN)。该算法首先基于应用场景和其他相关资源实现特征描述和表达;其次基于层次聚类设计特征矩阵表达算法,体现特征间的局部相关性以减少方法选择时间和提高选择精度;最后通过CNN模型实现空间数据自适应划分。本文选取南海区域真实数据和模拟数据进行验证,与已有的数据划分方法选择算法进行比较。结果显示:针对真实数据,SVDAP-CNN算法的精度和时间效率分别提高了24.18%和25.67%;针对特征和特征间关系表达欠完备的模拟数据,SVDAP-CNN算法的精度和时间效率分别提高了27.02%和26.80%;针对选择结果易出错的数据切分方法,SVDAP-CNN算法的精度提高了19.92%,证明了该算法有较好的普适性;另外,本文结合南海区的实际应用证明了该算法的应用可行性。SVDAP-CNN算法的提出可为大数据量、多变的自动化空间应用分析提供技术支撑。The adaptive data segmentation method is the key technology of data parallel computing automation.However,because data segmentation methods are mostly aimed at specific application scenarios,there is no clear boundary between methods,and the concept definition is relatively vague and general.The method selection results have the characteristics of one sidedness,pertinence,subjectivity,and uncertainty.Aiming at the problem that the spatial data segmentation methods cannot be selected adaptively and making full use of the advantage that CNN can establish end-to-end mapping without regular causality,this paper proposes an adaptive data partition algorithm based on CNN(Adaptive Partition Algorithm for Space Vector Data-Convolutional Neural Networks,SVDAP-CNN,).The algorithm comprehensively considers the factors affecting the selection accuracy and time efficiency of spatial data segmentation methods.Firstly,the feature and relationship between features are extracted through the description and expression of features,and the feature association directed graph is generated;Secondly,based on directed graph and clustering algorithm,the expression algorithm of characteristic matrix is designed to generate sample database.The expression of feature matrix reflects the local correlation between features,which reduces the method selection time and improves the method selection accuracy;Finally,through the combination of CNN model and classification function(softmax),the adaptive segmentation of spatial data is realized.This paper selects the real data of the South China Sea and the simulation data generated by the software for verification and compares it with the existing data segmentation method selection algorithm.Experimental results show that:for the real data with complete and accurate feature description and correlation,the accuracy of SVDAP-CNN algorithm is improved by 24.18%and the time efficiency is improved by 25.67%;for the simulation data with incomplete expression of features and relationship between features,th

关 键 词：空间数据切分方法 自适应 CNN 选择算法 特征 特征矩阵 局部相关 南海 

分 类 号：P208[天文地球—地图制图学与地理信息工程] TP183[天文地球—测绘科学与技术]