测绘学报 ›› 2022, Vol. 51 ›› Issue (10): 2034-2045.doi: 10.11947/j.AGCS.2022.20220326

• • 上一篇    下一篇

顾及样本优化选择的多核支持向量机滑坡灾害易发性分析评价

刘纪平1,2,3, 梁恩婕1,2, 徐胜华1,2, 刘猛猛1,2, 王勇1, 张福浩1, 罗安1   

  1. 1. 中国测绘科学研究院,北京 100036;
    2. 辽宁工程技术大学测绘与地理科学学院,辽宁 阜新 123000;
    3. 河南省科学院地理研究所,河南 郑州 450052
  • 收稿日期:2022-05-16 修回日期:2022-07-12 发布日期:2022-11-05
  • 通讯作者: 梁恩婕 E-mail:281998589@qq.com
  • 作者简介:刘纪平(1967—),男,博士,研究员,研究方向为政府地理空间大数据、政府地理信息服务、应急地理信息服务等。E-mail:liujp@casm.ac.cn
  • 基金资助:
    国家重点研发计划(2020YFC1511704)

Multi-kernel support vector machine considering sample optimization selection for analysis and evaluation of landslide disaster susceptibility

LIU Jiping1,2,3, LIANG Enjie1,2, XU Shenghua1,2, LIU Mengmeng1,2, WANG Yong1, ZHANG Fuhao1, LUO An1   

  1. 1. Chinese Academy of Surveying & Mapping, Beijing 100036, China;
    2. School of Geomatics, Liaoning Technology University, Fuxin 123000, China;
    3. Institute of Geography, Henan Academy of Sciences, Zhengzhou 450052, China
  • Received:2022-05-16 Revised:2022-07-12 Published:2022-11-05
  • Supported by:
    The National Key Research and Development Program of China (No. 2020YFC1511704)

摘要: 滑坡灾害易发性分析评价对地质灾害的防治与管理具有重要意义。针对滑坡灾害样本选择策略,单核支持向量机多特征映射不合理的问题,本文提出顾及样本优化选择的多核支持向量机(multiple kernel support vector machine,MKSVM)滑坡灾害易发性分析评价方法。为了保证样本平衡性并提高负样本的合理性,采用相对频率比(relative frequency,RF)综合评价各状态对于滑坡灾害易发性影响的重要程度,实现各评价因子状态的合理划分;利用确定性系数法(certainty factor,CF)计算各评价因子各状态分级影响滑坡灾害的敏感性,并在此基础上进行加权求和得到各栅格单元的滑坡灾害易发性指数,在滑坡灾害易发性指数极低和低易发区内随机选择与滑坡灾害点数目一致的非滑坡灾害点作为负样本数据。利用MKSVM对各特征空间最优核函数进行线性组合,解决了单一核函数映射不合理的问题,提高了模型的分类准确率和预测精度。以湖南省湘西土家族苗族自治州为研究区,从滑坡灾害易发性分区图、分区统计及评价模型精度3个方面对CF样本策略的MKSVM模型、CF样本策略的单核SVM模型、随机样本策略的MKSVM模型、随机样本策略的单核SVM模型进行了对比分析。结果表明,4种模型的受试者工作特征曲线(receiver operating characteristic,ROC)下的面积(area under curve,AUC)分别为0.859、0.809、0.798、0.766,验证了CF样本策略的合理性、有效性及MKSVM模型的可靠性。

关键词: 滑坡, 易发性, 确定性系数, 多核, 支持向量机

Abstract: The analysis and evaluation of landslide disaster susceptibility is of great significance to the prevention and management of geological disasters. In view of the sample selection strategy and the unreasonable multi-feature mapping in single-kernel vector machine, this paper proposes the landslide susceptibility analysis and evaluation method of multiple kernel support vector machine (MKSVM) considering the sample optimization selection. To ensure sample balance and improve the plausibility of negative samples, using the relative frequency ratio (relative frequency, RF) comprehensively evaluate the importance degree of each state in the influence of landslide disaster susceptibility, the purpose is to realize the reasonable division of each evaluation factor state; Using the deterministic coefficient method (certainty factor, CF) calculates the sensitivity of each state of each evaluation factor, the weighted sum has obtained the landslide disaster susceptibility index of each grid cell, non-landslide disaster points consistent with the number of landslide disaster points were randomly selected in the very low and low landslide disaster prone index as the negative sample data. Then, multi-kernel learning is used to select the SVM optimal kernel function and to linear combine the optimal kernel functions in each feature space to avoid unreasonable mapping of a single kernel function, and it improve the classification accuracy and prediction accuracy of the model. Taking Xiangxi Tujia and Miao Autonomous Prefecture of Hunan province as the research area, MKSVM model of CF sample strategy, single-kernel SVM model of CF sample strategy, MKSVM model of random sample strategy and single-kernel SVM model of random sample strategy were compared analyzed from three aspects of landslide disaster prone zoning map, partition statistics and evaluation model accuracy. The results indicate that the subject operating characteristic curves of the four models (receiver operating characteristic, area under the ROC) (area under curve, AUC) were 0.859,0.809,0.798,0.766, the rationality and validity of the CF sample strategy and the reliability of the MKSVM model are verified.

Key words: landslide, susceptibility, certainty factor, multi-kernel, support vector machine

中图分类号: