胡鹏¹, 赵露露¹, 高磊¹, 朱金山^1,2,3
1.山东科技大学测绘科学与工程学院, 山东 青岛 266590;2.地理信息工程国家重点实验室, 陕西 西安 710054;3.自然资源部海洋测绘技术重点实验室, 山东 青岛 266590

摘要:
在多光谱遥感浅海水深反演过程中，考虑到水体和底质影响，水深值和海水表面辐射亮度之间的线性关系不成立。本文以甘泉岛南部0~25 m范围的沙质区域为研究区域，利用GeoEye-1多光谱遥感影像和多波束实测水深数据构建XGBoost非线性水深反演模型，研究了XGBoost算法用于水深反演的性能。以决定系数（R²），均方误差（MSE）和平均绝对误差（MAE）作为评价指标，并与3种传统线性回归模型进行了对比分析。结果表明，XGBoost非线性水深反演模型的R²、MSE和MAE分别为0.991、0.33 m和0.44 m，拟合程度最好，精度优于线性回归模型。为进一步探究各模型在不同水深的反演精度，将水深范围分成3段（0~8 m，8~15 m，15~25 m）分别进行精度验证和误差分析。结果表明，XGBoost模型在各分段的精度均优于线性回归模型，MSE依次为0.56 m，0.14 m和0.43 m。可见，在单一底质区域下XGBoost模型的水深反演精度更高，且反演效果更稳定。
关键词:光学浅海水深反演XGBoost算法非线性回归模型底质类型
DOI：10.11759/hykx20191226002
分类号:P237
基金项目:地理信息工程国家重点实验室开放研究基金资助项目（SKLGIE2017-Z-3-3）；国家重点研发计划课题（协作）-极区海域水声环境观测与声场特性研究（2018YFC1405903）；测绘地理信息公益性行业科研专项（201512034）

Application of XGBoost algorithm on multispectral shallow water bathymetry retrieval
HU Peng¹, ZHAO Lu-lu¹, GAO Lei¹, ZHU Jin-shan^1,2,3
1.Geomatics College, Shandong University of Science and Technology, Qingdao 266590, China;2.State Key Laboratory of Geographic Information Engineering, Xi'an 710054, China;3.Key Laboratory of Marine Mapping Technology, Ministry of Natural Resources, Qingdao 266590, China
Abstract:
In optical shallow water bathymetry retrieval, due to the influence of the water and sediment types, the relationship between the water depth and sea surface reflectance is nonlinear. In this study, we built a nonlinear depth inversion model that uses the XGBoost algorithm. The research area was a 0-25 m sandy area around Ganquan Island in the South China Sea. GeoEye-1 multispectral data and in-situ multibeam data were used to investigate the depth inversion performance of the XGBoost algorithm. To evaluate the retrieved bathymetry results, we calculated the correlation coefficient (R²), mean square error (MSE), and mean absolute error (MAE) values. We then compared the XGBoost bathymetry results with those of three linear regression models, and found the XGBoost nonlinear depth inversion model to have the best fitting performance and better precision, with R², MSE and MAE values of 0.991, 0.33, and 0.44 m, respectively. To further explore the performance of each model at different depths, we divided the water depth into three ranges (0-8 m, 8-15 m, 15-25 m). The results show that, in each depth range, the XGBoost model's accuracy was better than those of the linear regression models. The MSE values in each depth range are 0.56, 0.14, and 0.43 m, respectively. Based on these results, we can conclude that, compared to other models, the depth inversion accuracy of the XGBoost model is higher, and its inversion effect is more stable in a single sediment region.
Key words:optical shallow water depth inversionXGBoost algorithmnonlinear regression modelsediment type





PDF全文下载地址:

点我下载PDF