Illegal Buildings Detection from Satellite Images using GoogLeNet and Cadastral Map

Abstract

Automatic illegal building detection from satellite imagery is a specific and important problem for both research community and government agencies, which has not been sufficiently investigated since it combines the challenge of automatic remote sensing data interpretation and verification with a cadastral map. Recovery of building footprints from satellite images is a very complicated process because building areas and their surroundings are represented with various color intensities and complex features. This paper proposes a methodology that incorporates image processing techniques with deep learning tools for the automatization of State Inspection in the field of illegal buildings detection. Our approach integrates four various computer vision-based building detection techniques for generating region proposals with the following classification by retrained GoogLeNet convolutional neural network (CNN) and the final validation whether detected building areas are legal according to a state cadastral map. Our tests with satellite imagery dataset show that the proposed approach achieves acceptable results in both building detection and legality assessment.