Comparison of different machine learning classifiers for building extraction in LiDAR-derived datasets

Issue Date

2015

Abstract

Building extraction in remotely sensed imagery is an important problem that needs solving. It can be used to aid in urban planning, hazard assessments and disaster risk management among others. Light Detection and Ranging or LiDAR, is one of the most powerful remote sensing technologies nowadays. Many studies have used the fusion of LiDAR data and multispectral images in detecting buildings. This study seeks to maximize the power of LiDAR imagery to be able to classify buildings without the aid of multispectral imagery. This work follows the Object Based Image Analysis (OBIA) approach. Instead of the traditional pixel-based classification methods, pixels are segmented into logical groups called objects. From these objects, features for building extraction are calculated. These features are: the number of returns, difference of returns, and the mean and standard deviation of positive surface openness. These objects are then classified using different machine learning classifiers such as Support Vector Machines, K-Nearest Neighbors, Naïve Bayes Classifier, Decision Trees, and Random Forests. A comparative assessment was done on the performance of these different machine learning classifiers. The classifiers performed similarly with the Random Forest Classifier slightly outperforming the others.

Source or Periodical Title

ACRS 2015 - 36th Asian Conference on Remote Sensing: Fostering Resilient Growth in Asia, Proceedings

Page

1-10

Document Type

Conference Paper

Physical Description

illustrations, graphs

Language

English

Subject

Feature extraction, Object based image analysis

Digital Copy

yes

Link to Full Text

Share

COinS