摘要
为探究不同机器学习模型对水深反演精度的影响,本文根据遥感水深反演原理,针对极限梯度提升(XGBoost)、支持向量机(SVM)、核岭回归(KRR)和套索回归(LASSO)四种机器学习模型,利用WorldView-2多光谱卫星遥感影像和机载激光雷达(Lidar)测深数据,在训练样本和测试样本相同的情况下,结合网格搜索方法寻找机器学习算法的最优参数组合,对我国南海甘泉岛周边浅水水域进行水深反演实验。通过四类机器学习模型反演结果比较分析得出结果:在20 m以浅水深区域,XGBoost模型的学习能力较强,相关系数(R^(2))为0.97,均方根误差(RMSE)为0.85 m,平均绝对误差(MAE)为0.63 m,平均相对误差(RME)为19%,优于其他三种机器学习模型,总体效果最好,能够用于甘泉岛周边水域的预测,为后续开展水深反演研究提供借鉴意义。
According to the principle of remote sensing water depth inversion,this paper used the WorldView-2 multi-spectral satellite remote sensing image and airborne Lidar data for four machine learning models,including extreme gradient boosting(XGBoost),support vector machine(SVM),kernel ridge regression(KRR)and least absolute shrinkage and selection operator(LASSO),to explore the impact of different machine learning models on the accuracy of water depth inversion.For deepth data,when the training sample and the test sample were the same,the grid search method was used to find the optimal parameter combination of the machine learning algorithm,and the water depth inversion experiment was carried out on the shallow waters around Ganquan island.Through the comparison and analysis of the inversion results of four types of machine learning models,the results were as follows:in the shallow water depth area of 20m,XGBoost model had strong learning ability,with correlation coefficient(R^(2))of 0.97,root mean square error(RMSE)of 0.85 m,average absolute error(MAE)of 0.63 m and average relative error(RME)of 19%,which was better than the other three machine learning models,and the overall effect was the best.It can be used to predict the water depth around Ganquan island and provide reference for subsequent water depth inversion research.
作者
饶亚丽
沈蔚
栾奎峰
纪茜
孟然
郝李华
RAO Yali;SHEN Wei;LUAN Kuifeng;JI Qian;MENG Ran;HAO Lihua(School of Marine Science,Shanghai Ocean University,Shanghai 201306,China;Shanghai Estuary Marine Surveying and Mapping Engineering Technology Research Center,Shanghai 201306,China;Nantong Institute of Intelligent Perception,Nantong 226009,China)
出处
《海洋湖沼通报（中英文）》
CSCD
北大核心
2024年第5期56-63,共8页
Transactions of Oceanology and Limnology
基金
国家重点研发计划(2016YFC1400904)
上海市科委重点科研计划(17DZ1204902)
上海市海洋局科研项目(沪海科2019-5,沪海科2020-5)。
作者简介
第一作者:饶亚丽(1996-),女,硕士研究生,主要研究方向为海洋遥感。E-mail:2449899639@qq.com;通信作者:沈蔚(1977-),男,博士,教授,主要研究方向为遥感与测绘。E-mail:wshen@shou.edu.cn。
