3D Classification optimization of Airborne LiDAR data by parameters selection

Abstract

Airborne Laser Scanning (ALS) point cloud segmentation and classification are very important and attractive tasks that still interest many scientists. In this paper, 3D classification optimization method is applied for of airborne LIDAR point cloud, covering an urban area with a low density. Our main contribution is to solve classifier and kernel hyperparameters tuning issue. Typically, choice of parameters is done empirically. In almost all classification studies, no details are given for the classifier parameters choice, which is very important. The parameters of the classifier have a direct impact on the classification results. Finding hyper-parameters leads to optimal classification results. In order to determinate the hyper-parameters, authors propose to apply in the context of LiDAR data, a method proposed in the literature called parameters selection to deduce the best parameters, in order to optimize the classifications results. The results shown that parameters selection does not systematically lead to hyper-parameters, but can be used as first stage, then a fine search can be performed around the obtained interval.