University of Twente Student Theses


An FCN-based approach to analyse dynamics of urbanizing areas

Yang, Xujiayi (2020) An FCN-based approach to analyse dynamics of urbanizing areas.

[img] PDF
Abstract:Urbanization rates are accelerating considerably in recent years. Presently, research on urbanization dynamics mainly focuses on large cities, while small urbanizations are not well covered. Also, not much data is available on urbanization processes in small urbanizing areas. For example, global datasets on built-up areas often omit small urbanizations. Remote sensing images are increasingly used for the spatial analysis of the dynamics of urban areas and could fill this information gap. Such data combined with machine learning algorithms such as fully convolutional networks (FCN) can classify land cover/use classes from satellite images and further extract contiguous built-up areas. Although there are projects that manually delineate contiguous built-up areas, they are time-consuming and labour-intensive. The objective of this study is to develop an FCN-based approach to semi-automatically delineate and analyse the spatial dynamics of urbanizing areas. It uses the example of the Barharia cluster, a small region located in the Bihar state of India, which consists of six villages. This study regards contiguous built-up areas with more than 10,000 people as urbanizing areas. To delineate urbanizing areas, this study takes advantages of deep learning and applied FCN with dilated kernels to classify built-up areas, roads and non-built-up areas from very high resolution (VHR) images in 2005, 2010 and 2018. The contiguous built-up areas were derived by aggregating classified built-up areas with a gap of less than 200 metres. Meanwhile, the population of each contiguous built-up area was estimated based on the census data of Bihar in 2001 and 2011, and the administrative boundaries of settlements in the study area. Finally, spatial metrics was used to analyse the dynamics of urbanizing areas. The classification accuracy of built-up areas in all three years obtained a F1-score of more than 84%. Based on the used definition, one urbanizing area was identified respectively in 2010 and 2018, and there was no urbanizing area in this study area in 2005. Results of spatial metrics calculation indicated that contiguous built-up areas aggregated over time and the urbanizing areas expanded from 2010 to 2018. Moreover, the population density of urbanizing areas decreased from 2010 to 2018 and the land use efficiency of this study area was also decreased over time. This study concludes that the developed FCN-based approach can delineate urbanizing areas in a semi-automatic manner, and based on the analysis of the classification results, expanding urbanizing areas (from 137.44 ha in 2010 to 314.62 ha in 2018) were found in the study area. However, in the developed approach, the assumption made for population estimation and census data disaggregation is quite simple. Nevertheless, the method used for contiguous built-up area extraction is more time-efficient than manual delineation. This shows a potential to develop approaches for the efficient delineation and spatial dynamic analysis of urbanizing areas in small regions.
Item Type:Essay (Master)
Faculty:ITC: Faculty of Geo-information Science and Earth Observation
Programme:Geoinformation Science and Earth Observation MSc (75014)
Link to this item:
Export this item as:BibTeX
HTML Citation
Reference Manager


Repository Staff Only: item control page