Automatic Rooftop Detection Using a Two-stage Classification

  • Bikash Joshi

Student thesis: Master's Thesis

Abstract

Automatic rooftop detection from satellite images is a popular research area with a variety of applications. One important application of this study is to estimate the available rooftop space for solar Photovoltaics(PV) system installation. This study presents a novel application of machine learning techniques for the automatic detection of building rooftops in satellite images. The images are first segmented into homogeneous regions using k-means algorithm. These segments are then treated as candidate rooftop regions which are presented to a novel two-stage classification process; features are extracted from each segment and submitted to an ANN which serves as the first stage of the classification procedure. New features are then extracted from the outputs of the ANN and these are submitted to an SVM which then performs the second classification pass. In this way, the first classification stage acts as a preprocessing step which, when processed by the SVM significantly reduces the number of false-positives. Additionally, we present a technical and economic analysis of the solar PV system installation in the available area detected by our rooftop detection algorithm.
Date of Award2014
Original languageAmerican English
SupervisorWei Lee Woon (Supervisor)

Keywords

  • Photovoltaic Power Systems; Solar Photovoltaics(PV); Image Segmentation; Artificial Neural Network.

Cite this

'