期刊名称：American Journal of Geographic Information System
印刷版ISSN：2163-1131
电子版ISSN：2163-114X
出版年度：2019
卷号：8
期号：5
页码：213-220
DOI：10.5923/j.ajgis.20190805.03
语种：English
出版社：Scientific & Academic Publishing Co.
摘要：This research proposed an approach for automatic extraction of buildings from digital aerial imagery and LiDAR data. The building patches are detected from the original image bands, normalized Digital Surface Model (nDSM) and some ancillary data. Support Vector Machines (SVMs) and artificial neural network (ANNs) classifiers have been applied individualey as member classifiers. In order to improve the obtained results, SVMs and ANNs have been combined in serial, parallel and hybrid forms. The results showed that hybrid system has performed the best with an overall accuracy of about 87.211% followed by parallel combination, serial combination, ANNs and SVMs with 84.709, 82.102, 77.605 and 74.288% respectively.
关键词：Building extraction; nDSM; Hybrid system; Image classification; SVMs and ANNs
