Building change detection via semantic segmentation and difference extraction method

Google Earth with high-resolution imagery basically takes months to process new images before online updates. It is considered as a time consuming and slow process especially for post-disaster application. In this study, we aim to develop a fast and accurate method of updating maps by detecting local differences occurred over different time series; where only region with differences will be updated. In our system, aerial imageries from Massachusetts's building open datasets are used as training datasets; meanwhile Saitama district datasets are used as input images. Semantic segmentation is then applied to input images to get predicted map patches of building. Semantic segmentation is a pixel-wise classification of images by implementing convolutional neural network technique. Convolutional neural network technique is implemented due to being not only efficient in learning highly discriminative image features such as buildings, but also partially robust to incomplete and poorly registered target maps. Next, in order to understand overall changes occurred in an area, both semantic segmented images from the same scene are undergone change detection method. Lastly, difference extraction method is implemented to specify the category of building changes. The results reveal that our proposed method is able to overcome current time-consuming map updating problem. Hence map updating will be cheaper, faster and more effective especially post-disaster application, by leaving unchanged region and only updating changed region.