The detection of building edges from very high resolution (VHR) remote sensing imagery is essential to various geo-related applications, including surveying and mapping, urban management, etc. Recently, the rapid development of deep convolutional neural networks (DCNNs) has achieved remarkable progress in edge detection; however, there has always been the problem of edge thickness due to the large receptive field of DCNNs. In this paper, we proposed a multi-scale erosion network (ME-Net) for building edge detection to crisp the building edge through two innovative approaches: (1) embedding an erosion module (EM) in the network to crisp the edge and (2) adding the Dice coefficient and local cross entropy of edge neighbors into the loss function to increase its sensitivity to the receptive field. In addition, a new metric, E_ne, to measure the crispness of the predicted building edge was proposed. The experiment results show that ME-Net not only detects the clearest and crispest building edges, but also achieves the best OA of 98.75%, 95.00% and 95.51% on three building edge datasets, and exceeds other edge detection networks 3.17% and 0.44% at least in strict F1-score and E_ne. In a word, the proposed ME-Net is an effective and practical approach for detecting crisp building edges from VHR remote sensing imagery.

The automatic vectorization of building shape from very high resolution (VHR) remote sensing imagery is fundamental in many fields such as urban management and geodatabase updating. Recently, deep convolutional neural networks (DCNNs) have been successfully used for building edge detection but the results are raster images rather than vectorized maps, and do not meet the requirements of many applications. Although there are some algorithms for converting raster images into vector maps, such vector maps often have too many vector points and irregular shapes. This study proposed a building shape vectorization hierarchy, which combined DCNNs-based building edge detection and a corner extraction algorithm based on principal component analysis (PCA) for rapidly extracting building corners from the building edges. Experiments on the Jiangbei New Area Buildings and Massachusetts Buildings datasets showed that compared with the state-of-the-art corner detectors, the building vector corners extracted using our proposed algorithm had fewer breakpoints and isolated points, and our building vector boundaries were more complete and regular. In addition, the building shapes extracted using our hierarchy were 7.94% higher than the non-maximum suppression (NMS) method in terms of relaxed overall accuracy (OA) on the Massachusetts dataset. Overall, our proposed hierarchy is effective for building shape vectorization.