Novel Approach for Rooftop Detection using Machine Learning Techniques

Abstract

In view of the increasing attention on the need for renewable energy sources, the UAE has actively positioned itself to become one of the major players in this field. By the year 2020 the UAE expects to generate 7% of its power from renewable energy sources. In an important step towards this goal a large number of solar photovoltaic (PV) stations have been installed on residential or commercial buildings to convert sunlight into electricity, with further rapid growth expected in the near future. A crucial step when planning for rooftop PV deployments is the determination of available rooftop area. In this research a new method is presented to achieve accurate automatic rooftop detection in satellite images. The proposed method is based on the use of machine learning techniques used for clustering and classification. Firstly, the images are segmented into a set of homogeneous regions, which are potentially associated with rooftop areas. Next, to achieve improved accuracy, a novel two-pass classification process is used to determine which amongst the candidates correspond to true rooftops. In the first pass a trained classification algorithm (i.e., Support Vector Machine (SVM), Artificial Neural Network (ANN) and Random Forest (RF)) is used in the normal way to distinguish between rooftop and non-rooftop regions. However, this can be a challenging task, resulting in a relatively high rate of misclassification. Hence, the second pass, which is called the histogram method was devised with the aim of detecting rooftops which were missed in the first pass. In addition, an economic evaluation of the PV installations for two test regions is performed using the Net Present Value (NPV) of comparable installations using three different types of cells. While the results are still preliminary, they point to the clear economic potential of these regions for solar energy generation.

Date of Award	2014
Original language	American English
Supervisor	Wei Lee Woon (Supervisor)