基于水深分段选择因子的多光谱影像反演水深

岛礁周边海底地形精密测量是海洋测绘的难点问题之一，利用多光谱卫星能精确测定岛礁周边水深地形，是传统岛礁水深测量有效的补充方式之一。本文以蜈支洲岛周围水域为研究区域，提出了基于BP神经网络水深分段选取反演因子的方法，利用WorldView-2数据和多波束实测数据建立多模型，整合预测的各段水深获得海底地形。实验表明，本文方法能充分利用水体的光谱特性，与单一BP神经网络模型结果相比，MRE降低了6.4%，RMSE降低了1.2 m，是一种可行的多光谱水深反演的方法

Multi-spectral image inversion of bathymetry based on water depth segmentation selection factors

The precise measurement of seafloor topography around islands and reefs is one of the difficult problems in marine mapping, and the use of multispectral satellites can accurately determine the bathymetric topography around islands and reefs, which is one of the effective complementary ways to traditional island bathymetry. In this paper, taking the waters around Wuzhizhou Island as the study area, we propose a method based on BP-net as well as water depth segmentation to select the inversion factors, and establish a multi-model using WorldView-2 data and actual measurement data to integrate the predicted water depths of each segment to obtain the seafloor topography. The experiment shows that the method in this paper can make full use of the spectral properties of the water body, and the MRE is reduced by 6.4% and the RMSE is reduced by 1.2 m compared with the results of a single BP net model, which is a feasible method for multispectral water depth inversion.