Master thesis defense by Malte von Mehren
Accelerating Break Detection Methods for Remote Sensing Using GPUs
The field of geography is nowadays faced with a lot data.
While this offers various research opportunities, it presents some challenges with computation time and space requirements. This basically renders existing approaches too slow, which in turn makes it hard to make sense of all the data. This work aims at accelerating a subset of the BFAST method, which depicts a prominent tool for analysing and detecting changes in time series and in particular satellite data. The overall goal of the thesis at hand is to make this approach faster via
(1) better hardware-close implementations and (2), more importantly, via parallel programming using graphics processing units. In the experimental evaluation of our work, we show that our implementation, while getting same results up to machine precision, is up to four orders of magnitude faster than the existing R implementation. This renders the analysis of significantly larger time series data sets possible in seconds or minutes instead of hours or even days.
Supervisor: Fabian Gieseke
Censor: Ira Assent