结合光谱信息与空间特征的浅海光学遥感水深反演  

Shallow water bathymetry retrieving of optical remote sensing combined with spectral information and spatial features

作　　者：尹飞 戚甲伟 滕东东 YIN Fei;QI Jiawei;TEN Dongdong(College of Geodesy and Geomatics,Shandong University of Science and Technology, Qingdao Shandong 266590, China)

机构地区：[1]山东科技大学测绘与空间信息学院,山东青岛266590

出　　处：《北京测绘》2022年第5期531-536,共6页Beijing Surveying and Mapping

摘　　要：本文基于四波段的GeoEye-1高分辨率遥感影像,以中国南海甘泉岛附近浅海区域为研究区进行水深反演实验。现阶段的机器学习水深反演方法,大多是把波段反射率信息作为反演因子,忽略了空间特征信息对水深反演的影响。本文结合影像的光谱信息与空间特征,利用极限梯度提升与BP神经网络算法构建水深反演模型,探究空间特征因子对模型性能的影响。研究结果表明:引入空间特征因子的两个模型均方根误差降低了25%~31%,相关系数从0.94提高到0.97。空间特征模型有效地降低了误差聚集性问题,水深反演精度显著提升。Based on the four band GeoEye-1 high-resolution remote sensing image,this paper took the shallow sea near Ganquan island in the South China Sea as the research area to carry out the bathymetry inversion experiment.At present,most of the machine learning bathymetry inversion methods only take the band reflectance information as the inversion factor,ignoring the influence of spatial characteristics information on water depth inversion.In this paper,combined with the spectral information and spatial characteristics of the image,the XGBoost and back propagation(BP)neural network algorithms were applied to construct the bathymetry inversion model,and the influence of spatial feature factors on the model performance was explored.The results showed that the root mean square error(RMSE)of the two models was reduced by 25%~31%,and the correlation coefficient(R2)was increased from 0.94 to 0.97.The spatial feature model reduced the problem of error aggregation effectively,and the accuracy of bathymetry inversion was improved significantly.

关 键 词：水深遥感 极限梯度提升 BP神经网络 空间特征 

分 类 号：P237[天文地球—摄影测量与遥感]