Author: Rodrigo da Silva Ferreira
Original title: InterIMAGE Cloud Platform: The Architecture of a Distributed Platform for Automatic, Object-Based Image Interpretation
The general objective of this thesis was the development of a distributed computational architecture for the automatic, object-based interpretation of large volumes of remote sensing image data, focusing on data and processing distribution in a cloud computing environment. Two specific objectives were pursued: (i) the development of a novel distributed architecture for image analysis that is able to deal with vectors and rasters at the same time; and (ii) the design and implementation of an open-source, distributed platform for the interpretation of very large volumes of remote sensing data. In order to validate the new architecture, experiments were carried out using two classification models – land cover and land use – on a QuickBird image of an area of the São Paulo municipality. The classification models, proposed by Novack (Novack09), were recreated using the knowledge representation structures available in the new platform. In the executed experiments, the platform was able to process the whole land cover classification model on a 32,000×32,000-pixel image (approximately 3.81 GB), with approximately 8 million image objects (approximately 23.2 GB), in just one hour, using 32 machines in a commercial cloud computing service. Equally interesting results were obtained for the land use classification model. Another possibility of parallelism provided by the platform s knowledge representation structures was also evaluated.