Change detection of buildings using high spatial resolution images and deep learning

Abstract

The main aim of this study is to detect changes in buildings. The data required to achieve this
goal are high spatial resolution images and the method used to achieve the goal is the use of a
deep neural network and a new relevant dataset. In this research, the training and testing of a
deep neural network is investigated to detect changes in buildings such that the accuracy is
improved. The deep learning networks have not been previously used in industrial areas that
have different characteristics such as building structures, building construction methods,
materials and color variety of the roofs of the buildings. In this research, it is assumed that the
development of networks alone does not improve the accuracy of the results considerably, but
the training of previously developed networks with the relevant training dataset greatly
increases the accuracy. Accordingly, a training dataset to detect the changes in the buildings
of industrial areas have been produced to train an already developed deep learning network
with the relevant data to detect the change in buildings. As a result, the trained network with
relevant data was tested, and reasonable average accuracy and recall achieved.