Classification of Airborne Laser Bathymetry Data Using Artificial Neural Networks

Abstract

The current development of technology allows for extensive use of active remote sensing systems in water environment research. The data obtained using airborne bathymetric scanning are mainly used to build digital elevation models of the seabed. As a result, their information potential is largely untapped, because full-waveform data have considerably more information that can be used for classification with different machine learning algorithms. This article presents the process of classification and detection of objects on the seabed using multilayer perceptron neural networks. The features of full waveform and point cloud geometry were considered for network training. The results obtained allow for almost 100% correct classification of water surface and seabed. The seabed object points were also classified with an accuracy of over 80%. The obtained results increase the effectiveness of object detection compared to the other well-known classification algorithms.

Keywords

• Airborne laser bathymetry (ALB)
• classification
• neural networks
• obstacles
• underwater objects detection