Performance Comparison of Classification Methods in Lidar Point Clouds
V. International Applied Statistics Congress (UYIK - 2024) , İstanbul, Türkiye, 21 - 23 Mayıs 2024, cilt.1, ss.1114-1123, (Tam Metin Bildiri)
- Yayın Türü: Bildiri / Tam Metin Bildiri
- Cilt numarası: 1
- Basıldığı Şehir: İstanbul
- Basıldığı Ülke: Türkiye
- Sayfa Sayıları: ss.1114-1123
- Dokuz Eylül Üniversitesi Adresli: Evet
Özet
Lidar technology works on the principle of determining the position by calculating the distance between the sensor and the surface point when the rays hitting a surface and return, emitted by the sensor. At the same time, the color values of the lidar points detected with the images obtained with the camera are also collected. Since the beams emitted from the lidar sensor detect the points of the first surface it hits, the point cloud data needs to be classified in order to be used in the desired case. The aim of this study is to classify, model and compare the performance of lidar point cloud data using machine learning algorithms.
Within the scope of the study, data were obtained from aerial lidar point cloud in the Democratic Republic of Congo. In the study, 8 different machine learning algorithms (classification tree, random forest, logistic regression, K-nearest neighbor, Naïve Bayes, support vector machine, artificial neural networks, multilayer perceptron) were used with a training set of 9391 points with 2 geometric variables and 1 variable describing color values. According to the lidar point cloud results, the highest accuracy was 84.0% with the Artificial Neural Networks (ANN) algorithm, while the lowest accuracy was 81.2% with the Naive-Bayes algorithm.