基于支持向量回归和贝叶斯优化的海岸线预测  

Shoreline prediction based on support vector regression and bayesian optimization

作　　者：李睿昕 王广军[1] 冯磊 梁四海[3] 高智雄 LI Ruixin;WANG Guangjun;FENG Lei;LIANG Sihai;GAO Zhixiong(School of Land Science and Technology,China University of Geosciences(Beijing),Beijing 100083,China;China Aero Geophysical Survey&Remote Sensing Center for Land and Resources,Beijing 100083,China;School of Water Resources and Environment,China University of Geosciences(Beijing),Beijing 100083,China)

机构地区：[1]中国地质大学(北京)土地科学技术学院,北京100083 [2]中国地质调查局自然资源航空物探遥感中心,北京100083 [3]中国地质大学(北京)水资源与环境学院,北京100083

出　　处：《海洋测绘》2022年第4期65-69,共5页Hydrographic Surveying and Charting

基　　金：中国地质调查局地调项目(DD20191006);中国科学院战略性先导科技专项(A类)子课题(XDA20100103)。

摘　　要：模拟预测海岸线的动态变化对于沿海地带的开发和科学管理有较重要意义。在获取围头湾研究区Landsat-8卫星时序影像和验潮站数据的基础上,利用支持向量回归模型,结合贝叶斯优化获取最佳超参数,进而预测得到未来年份的海岸线数据,并采用ROC匹配、平均偏移量和均方根误差3种方法验证预测得到的海岸数据。结果表明:基于贝叶斯优化的支持向量回归模型,可以较好地预测围头湾区域未来海岸线,对应的完整度、正确度和质量分别为87.06%、86.10%和80.05%,平均偏移量和均方根误差分别为0.64像素和0.96像素;对于自然岸线预测结果较好,而人工岸线预测部分结果较差。It is of great significance for the development and scientific management of coastal zone to simulate and predict the dynamic change of coastline.Based on the landsat-8 satellite sequence image and tide station data in the research area, this paper uses the support vector regression model, combined with Bayesian optimization to obtain the best hyperparameter, and then predicts the coastline data of the future years, and verifies the predicted coastline data by ROC curve matching, mean offset and root mean square error.The results show that the support vector regression model based on Bayesian optimization can predict the future coastline of Weitouwan area well, with the Complete, Correct and Quality of 87.06%,86.10% and 80.05%,and the mean offset and root mean square errors of 0.64 and 0.96,respectively.

关 键 词：海岸线预测 卫星时序影像 支持向量回归 贝叶斯优化 ROC匹配 

分 类 号：P229.4[天文地球—大地测量学与测量工程]