Joint Learning of Contour and Structure for Boundary-Preserved Building Extraction

Abstract

Most of the existing approaches to the extraction of buildings from high-resolution orthoimages consider the problem as semantic segmentation, which extracts a pixel-wise mask for buildings and trains end-to-end with manually labeled building maps. However, as buildings are highly structured, such a strategy suffers several problems, such as blurred boundaries and the adhesion to close objects. To alleviate the above problems, we proposed a new strategy that also considers the contours of the buildings. Both the contours and structures of the buildings are jointly learned in the same network. The contours are learnable because the boundary of the mask labels of buildings implicitly represents the contours of buildings. We utilized the building contour information embedded in the labels to optimize the representation of building boundaries, then combined the contour information with multi-scale semantic features to enhance the robustness to image spatial resolution. The experimental results showed that the proposed method achieved 91.64%, 81.34%, and 74.51% intersection over union (IoU) on the WHU, Aerial, and Massachusetts building datasets, and outperformed the state-of-the-art (SOTA) methods. It significantly improved the accuracy of building boundaries, especially for the edges of adjacent buildings. The code is made publicly available.

CGSANet: A Contour-Guided and Local Structure-Aware Encoder–Decoder Network for Accurate Building Extraction From Very High-Resolution Remote Sensing Imagery

“…Building contours implicitly represent building shape and structure features. The contours are learnable since the building region label can easily generate building contour labels [40]. Researchers adopted a multi-task learning framework and designed a hybrid loss function to learn building structure features [15], [24], [41], [42].…”

“…These methods have greatly improved the quality of extracted building boundaries; but they only supervise the building edges in the last layer of the decoder, which is still limited by the loss of detailed spatial information in the encoder layers. Liao et al [40] proposed a boundary-preserved network for building extraction by simultaneously learning building structure and contour. A structural prior constraint module combined with the dice loss function was designed to learn building contour information from the gradient image.…”

A Deep Learning-Based Framework for Automated Extraction of Building Footprint Polygons from Very High-Resolution Aerial Imagery

“…To resolve the problem of blurry object boundaries, Marmanis et al [33] proposed a DCNN models for semantic segmentation of high-resolution aerial images, which explicitly accounts for the boundaries of classes in the segmentation process. Taking the contours into consideration, Liao et al [34] proposed a boundary-preserved building extraction approach. By embedding the contour information in the labels, the proposed approach can enhance the representation of building boundaries and improve the performance on boundaries of adjacent buildings.…”

Attention Enhanced U-Net for Building Extraction from Farmland Based on Google and WorldView-2 Remote Sensing Images

“…With the continuous development of remote sensing technology, the obtained remote sensing images with higher spatial resolution contain abundant building information. Extracting buildings from high-resolution remote sensing images has become a research hotspot [9,[13][14][15][16]. Traditional building extraction methods, based on optical remote sensing images, mainly consider low-level semantic features such as color, texture, and shape to extract buildings.…”

“…Even if the high-resolution remote sensing images are rich in details, the complex types of features, pixel mixing, shadows and other problems within the farmland are serious, making the phenomenon of "same subject with different spectra" or "different subject with same spectra" more common. These methods are limited when solving the problem of building extraction under specific data conditions [14,22].…”