摘要
针对复合地形因子现有提取方法过程叠加繁琐、计算量大等问题,利用BP神经网络自适应性好、泛化能力强的优势创建了复合地形因子提取的BP神经网络法。以黄土高原DEM为数据源,选取复合地形复杂度指标为研究对象,采用控制变量法优选网络结构,结合Levenberg-Marquardt算法改进网络收敛速度与精度,优化样本库以克服过拟合问题,获取了综合性能最优泛化模型。实验结果表明,BP神经网络法较传统方法功效显著,拟合优度达到0.99,命中率达到99.98%,均方差为2.91×10^(-5),残差最大绝对值为0.16,平均绝对百分比误差为2.95%,模型提取10次结果标准差最大值为0.14;BP网络用于复合地形因子提取有效可行,为快速提取复合地形因子提供了新途径。
To address the issues with existing compound terrain factors extraction methods,such as cumbersome overlay process and large computation,a BP neural network method for compound terrain factors extraction was developed,taking the advantages of good self-adaptability and strong generalization ability based on BP neural network.The data source was digital elevation model of loess plateau,and the research object was the compound terrain complexity index(CTCI).Optimal network structure was acquired by control variables method,combined with Levenberg-Marquardt algorithm to improve the convergence speed and accuracy of the network.Optimizing sample base to overcome overfitting problems,the generalization model with comprehensive performance was obtained.According to the experiment results,the BP neural network method was significantly more effective than the traditional ones.The goodness of fit reached 0.99,the hit rate amounted to 99.98%,the mean square error was 2.91×10^(-5),the maximum absolute value of residuals was 0.16,the mean absolute percentage error was 2.95%,and the maximum standard deviation value of 10 times extraction was 0.14.BP network for compound terrain factors extraction was effective and feasible,which provided a new way for fast extraction of compound terrain factors.
作者
周访滨
马国伟
谢财昌
杨自强
钟绍平
ZHOU Fangbin;MA Guowei;XIE Caichang;YANG Ziqiang;ZHONG Shaoping(School of Traffic and Transportation Engineering,Changsha University of Science&Technology,Hunan 410114,China;Key Laboratory of Special Environment Road Engineering of Hunan Province,Changsha University of Science&Technology,Hunan 410114,China)
出处
《测绘科学》
CSCD
北大核心
2023年第1期227-235,共9页
Science of Surveying and Mapping
基金
国家自然科学基金项目(41671446)
湖南省自然科学基金项目(2021JJ30702)
特殊环境道路工程湖南省重点实验室开放基金资助重点项目(kfj140502)
作者简介
周访滨(1975—),男,甘肃灵台人,高级实验师,博士,主要研究方向为数字地形分析。E-mail:Arthur1975@126.com;通信作者:马国伟,硕士研究生,E-mail:pm0621@126.com
