Multispectral methods for remote sensing image have been widely applied to shallow water bathymetry by researchers. In nonideal conditions, even with the same spectral radiance, the points still have a very wide range of water depths. This means that spectral features alone are insufficient for water bathymetry. Hence, we need to extract other valuable features from a remote sensing image. This letter introduces a spatial feature for water bathymetry using remote sensing images. We propose a model that utilizes a multilayer perceptron (MLP) to integrate the spectral and spatial location features. Experimental results demonstrate that the proposed model yields a substantial performance improvement. The mean relative error is only 8.41%, and the root mean square error is reduced by 34%-68% when compared with three other models. Furthermore, the proposed model addresses well the problems caused by heterogeneous bottom types.

Accurate bathymetric data in shallow water is of increasing importance for navigation safety, coastal management, and marine transportation. Satellite-derived bathymetry (SDB) is widely accepted as an effective alternative to conventional acoustic measurements in coastal areas, providing high spatial and temporal resolution combined with extensive repetitive coverage. Many previous empirical SDB approaches are unsuitable for precision bathymetry mapping in various scenarios, due to the assumption of homogeneous bottom over the whole region, as well as the neglect of various interfering factors (e.g., turbidity) causing radiation attenuation. Therefore, this study proposes a bottom-type adaption-based SDB approach (BA-SDB). Under the consideration of multiple factors including suspended particulates and phytoplankton, it uses a particle swarm optimization improved LightGBM algorithm (PSO-LightGBM) to derive depth of each pre-segmented bottom type. Based on multispectral images of high spatial resolution and in situ observations of airborne laser bathymetry and multi-beam echo sounder, the proposed approach is applied in shallow water around Yuanzhi Island, and achieves the highest accuracy with an RMSE value of 0.85 m compared to log-ratio, multi-band, and classical machine learning methods. The results of this study show that the introduction of water-environment parameters improves the performance of the machine learning model for bathymetric mapping.

Enteromorpha prolifera is as a frequent marine ecological environment disaster. How to quickly and accurately monitor Enteromorpha prolifera is of great significance to its management and protection of the marine ecological environment. The detection of Enteromorpha prolifera from high spatial resolution remote sensing images (HSRIs) is an important technical means for monitoring Enteromorpha prolifera disasters. With respect to the difficulty of accurate detection of Enteromorpha prolifera area boundary in HSRIs, this paper proposes an Enteromorpha prolifera detection method for HSRIs based on boundary-assisted dual-path convolutional neural networks (CNN). First, a large-scale HSRIs Enteromorpha prolifera detection dataset, FIO-EP, is created and made publication to facilitate the field of HSRIs Enteromorpha prolifera detection. Then, a boundary-assisted dual-path CNN framework is designed to detect Enteromorpha prolifera in HSRIs based on the shape distribution characteristics of Enteromorpha prolifera. In the CNN framework, accurate detection of Enteromorpha prolifera areas in HSRIs is achieved by fusing initial detection and boundary detection results of Enteromorpha prolifera. The proposed method is compared with some state-of-the-art Enteromorpha prolifera detection algorithms using the FIO-EP dataset. The experimental findings demonstrate that the proposed method can obtain 88.28% F1-score and 79.02% intersection-over-union (IOU), and is superior to other state-of-the-art Enteromorpha prolifera detection algorithms.