Multi-scale edge constrained for building semantic segmentation using high resolution remote sensing images

Building segmentation from high spatial resolution remote sensing images is still a great challenge in the field of remote sensing image processing. This paper proposes an end-to-end deep convolutional neural network method for building semantic segmentation using high resolution remote sensing images. In this method, a multi-scale edge enhancement module(MEEM) with the Laplacian operator as the core is embedded to extract the building edge, which is used as an auxiliary method of semantic information to improve the segmentation accuracy. Based on the experimental results of the WHU Aerial imagery dataset, it is shown that the proposed method can not only achieve the same performance as some of the best building segmentation methods but also solve the problem that small buildings are difficult to be correctly detected.