Akademik Veri Yönetim Sistemi
Tezin Türü: Yüksek Lisans
Tezin Yürütüldüğü Kurum: Hacettepe Üniversitesi, Fen Bilimleri Enstitüsü, Geomatik Mühendisliği A.B.D., Türkiye
Tezin Onay Tarihi: 2015
Tezin Dili: İngilizce
Öğrenci: FATEMEH SAFARLOU
Danışman: Mustafa Türker
Özet:
The automatic and accurate detection of buildings and the changes of buildings in urban environment is important for the efficient updating of geographic databases, decision makers, urban planning and management. In this study, building changes in a selected urban area in the Altındağ district of Ankara, Turkey were detected from a high resolution satellite imagery by using image to map comparison change detection technique. The data used include the WorldView2 satellite image acquired in 2010, the vector map data compiled in 2001, and DSM data generated from stereo aerial photographs taken in 2013.
The method used in the study consists of four main steps. In the first step, a multiresolution image segmentation was carried out using the multispectral image bands and DSM data in eCognition software. In the segmentation process, the parameters Scale, Shape, Compactness and Layer Weight were used. The values for these parameters were defined by trial and error analysis. In the second step, the image was initially classified using the nearest neighbour (NN) method, which was then followed by a rule-based classification. During rule-based classification, the spatial, geometrical, contextual and texure features of the image objects as well as the relations between them were considered. After performing the classification operation, the building class was masked out and a morphological opening operation was applied on building class as a post processing operation. In the third step, the accuracy assessment of the detected building class was performed by means of comparing it with the manually digitized reference data set. With this respect, true positive (TP), true negative (TN), false positive (FP), false negative (FN), building detection percentage (BDP) and quality percentage (QP) measures were computed. In addition, an error matrix was generated based on random samples and an overall accuracy was computed. In the fourth step, the building changes were detected by means of comparing the detected building class with the old vector map data. The change detection process was carried out in ArcGIS software by means of an overlay analysis. The detected changes were categorized into new buildings and demolished buildings. Moreover, the accuracy assessment of the detected changes was also carried out based on the accuracy measures same before using the change reference data set, which was generated by manually on screen digitization of the changes.
Experimental
results demonstrate the effectiveness of both object-based classification and
image to map change detection methods in high resolution satellite imagery. For
the building class extracted through object based classification, the
building detection percentage (BDP) value was computed to be 82.21%. Similarly,
the BDP values for the detected changed and unchanged areas were computed to be
86.06% and 70.64%, respectively. Multiresolution segmentation and the subsequent rule
based classification were found to be quite efficient for successfully detecting
buildings from high resolution satellite imagery. However, the
classification performance was not quite good for small and unclear
buildings as well as those buildings that are surrounded by high vegetation
which strongly effects the results of classification and the subsequent change
detection. The use of true orthoimagery of high resolution satellite data in
classification would increase the accuracy results of both building detection
and the subsequent change detection. The results demonstrate that the method
presented in this study can be efficiently used to detect changes of buildings
and update geodatabases in urban environment.