Master Thesis | gulnazbekkarnayeva

Object Based Change Detection of Informal Settlements in Very High Resolution Data

Supervisor: Dr. Peter Hofmann, ZGIS, PLUS

The aim of the project was to identify changes of informal settlements in Cape Town regions: Nyanga, Philippi, Gugulethu, and Crossroads between 2000 and 2015. Following objectives had to be reached:

1. Detection of informal settlements in a selected area of Cape Town in a time-series data.

2. Adaptation and advancement of the rule set due to multiple image data with analysis of the rule set´s transferability.

3. Analysis and documentation of changes based on temporally linked image objects.

Study area and datasets

Images from 2000 to 2015 years of Ikonos-2, QuickBird, WorldView-2, 3 (provided by European Space Imaging EUSI) for Cape Town (Nyanga, Philippi, Gugulethu, and Crossroads regions).

Methodology

The overall workflow included pre-processing, processing, accuracy assessment, and time-series analysis. Images were classified with an existing rule set for 2000 year's image (provided by Dr. Peter Hofmann), and adapted due to various spatial and spectral resolutions, and acquired times of satellite imagery. Hierarchical classification with the same set of features but varying membership functions was applied for all images. Further, accuracy assessment was performed using TTA masks, which resulted in overall 70%. Finally, time-series analysis was performed with object links. Changes of informal settlements were assessed by transitions to classes, sizes, density of sub-objects, and internal changes.

Results

In overall, informal settlements were demolishing each time period, however between 2000 and 2004 most informal settlements became unclassified. On the contrary, informal settlements were emerging every year, and the majority of them appeared in 2002, 2015, and 2004. Between 2000 and 2002 8% of informal settlements became formal, and twice of this percentage were demolished. Following two years, 15% of informal settlements were taken down. In addition, in the period between 2012 and 2013 most formal settlements became informal.

Conclusion

According to accuracy assessment of image classifications, and modified membership functions, the rule set was proven to be transferable. Within this research textural parameters showed efficient transferable results with little adaptation. In general, one rule set was able to provide comparatively the same results in classification of various images for the same region. Object links showed valuable results in informal settlements´ change detection.