Fotogrametrik nokta bulutu verisinin makine öğrenmesi ile sınıflandırılması

Abstract

Today, thanks to the development of science and technology, the three-dimensional earth reality has been modeled and a new page has been opened in map production. Especially with the development of photogrammetric methods, point clouds have started to be used in many areas. Extraction of information from point clouds has become important with the growth of data sizes and the spread of usage areas. Since point cloud data consists of a large number of points, learning-based approaches have been used to classify these data. When learning-based approaches are used for the classification process, the distinguishability of the objects in the point cloud from each other increases and provides convenience and reliability for studies. The usage of machine learning, which has powerful mathematical algorithms, comes first in order to analyze complex data. Within the scope of this study, point cloud data was produced by using aerial photographs taken from unmanned aerial vehicle of the region determined in Istanbul Technical University Ayazağa Campus and classified according to four classes (building, tree, vehicle and ground level objects). This classification process has been done by using Random Forest (RF) and Multi-Layer Perceptron (MLP) algorithms from machine learning algorithms. The overall accuracy was 78.54% with the RF algorithm, and 89.88% with the MLP algorithm.