Lidar signal processing for under-water object detection

Abstract

This paper presents Artificial Neural Network (ANN) based architecture for underwater object detection from Light Detection And Ranging (Lidar) data. Lidar gives a sequence of laser backscatter intensity obtained from laser shots at various heights above the earth surface. Lidar backscatter can be broadly classified into three different classes: water-layer, bottom and fish. Multilayered Perceptron (MLP) based ANN architecture is presented, which employ different signal processing techniques at the data preprocessing stage. The Lidar data is pre-filtered to remove noise and a data window of interest is selected to generate a set of coefficient that acts as input to the ANNs. The prediction values obtained from ANNs are fed to a Support Vector Machine (SVM) based Inference Engine (IE) that presents the final decision. © Springer-Verlag Berlin Heidelberg 2005.