Vertex Aided Building Polygonization from Satellite Imagery Applying Deep Learning

Kurzfassung

Building extraction is an important task in many fields. The use of convolutional neural networks has been proven to be of great success in building extraction from satellite images. This paper presents a deep learning based vertex aided building polygonization method, which takes RGB satellite images as input and outputs building polygons. Unlike other methods which rely on vertex extraction followed by polygonization, our method requires neither pre-defined number of vertices nor thresholding to obtain extracted vertices. The proposed method has the advantage of simplicity in sense of model complexity, and achieved good performance with average precision of 48.1% and intersection over union of 84.1%.