海底底质分类支持下的WorldView-3多光谱影像浅海海域水深反演

Shallow sea water depth inversion from WorldView-3 multispectral images based on seabed sediment classification

近几十年来,基于遥感影像进行水深反演一直是国内外学者研究的热点。本文使用WorldView-3高分辨率卫星影像,结合卫星测高数据,以中国海南岛附近的蜈支洲岛及其附近海域为主要研究区域,在进行数据预处理、底质分类之后,分别通过多元线性回归模型、Stumpf对数比值模型和BP神经网络集中对岛屿周围0~20 m水域的水深进行反演和结果分析。结果证明,对这3种模型而言,在进行底质分类之后精度都会明显提升。其中,BP神经网络反演水深精度最高(均方根误差范围为0.2~0.7 m),多元线性回归模型次之(均方根误差范围为0.3~0.8 m),对数比值模型精度最低(均方根误差范围为0.6~1.1 m)。

In recent decades,sea water bathymetry inversion method based on remote sensing image has been a hot research topic.This paper uses WorldView-3 high-resolution satellite imagery,combined with satellite altimetry data,to focus on Wuzhizhou island which is near Hainan Island,China,and its adjacent waters as the main study area.After data preprocessing and substrate classification,multiple linear regression model,Stumpf logarithmic ratio model and BP neural network model are used to invert and analyze the water depth around the island.Results show that:for the three model,after the bottom sediment classification accuracy will be improved significantly.Among them,BP neural network model has the highest accuracy(root mean square error range of 0.2~0.7 m),followed by multiple linear regression model(root mean square error range of 0.3~0.8 m),and log ratio model has the lowest accuracy(root mean square error range of 0.6~1.1 m).