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

Abstract

Extracting buildings accurately from very highresolution (VHR) remote sensing imagery is challenging due to diverse building appearances, spectral variability, and complex background in VHR remote sensing images. Recent studies mainly adopt a variant of the Fully Convolutional Network (FCN) with an encoder-decoder architecture to extract buildings, which has shown promising improvement over conventional methods. However, FCN-based encoder-decoder models still fail to fully utilize the implicit characteristics of building shapes. This adversely affects the accurate localization of building boundaries, which is particularly relevant in building mapping. A contour-guided and local structure-aware encoder-decoder network (CGSANet) is proposed to extract buildings with more accurate boundaries. CGSANet is a multi-task network composed of a contour-guided (CG) and a multi-region guided (MRG) module. The CG module is supervised by a building contour that effectively learns building contour-related spatial features to retain the shape pattern of buildings. The MRG module is deeply supervised by four building regions that further capture multiscale and contextual features of buildings. In addition, a hybrid loss function was designed to improve the structure learning ability of CGSANet. These three improvements benefit each other synergistically to produce high-quality building extraction results. Experiment results on the WHU and NZ32km2 building datasets demonstrate that compared with the tested algorithms, CGSANet can produce more accurate building extraction results and achieve the best intersection over union (IoU) value 91.55% and 90.02%, respectively. Experiments on the INRIA building dataset further demonstrate the ability for generalization of the proposed framework, indicating great practical potential.

Multi-Layer Fault-Tolerant Protection Strategies for Hybrid Distribution Transformers Integrated Photovoltaic Systems

“…Post-processing based building mapping refines or vectorizes binary building masks using post-processing procedures to refine boundary or vectorize building masks. Since vectorization of building binary masks tends to bring blurred and irregular boundaries, many boundary refinement methods [14]- [17], [34] have been studied to regularize building boundaries by multi-scale feature fusion, building shape information embedding, or other post-processing procedures. BCTNet [35] proposes a bi-branch cross-fusion transformer network by using CNN and transformer to enhance multi-scale features from local and global aspects.…”

“…All these pixel-wise segmentation-based methods fail to obtain accurate building boundaries due to dense buildings and similar backgrounds in remote sensing images. To refine blurred boundaries, some studies [14]- [17] introduce boundary-preserved modules to regularize building boundaries. Although recent pixel-wise segmentation methods produce accurate buildings with precise boundaries, they usually output raster building segmentation masks, requiring a delicate post-vectorization pipeline to meet real-world geographic applications.…”

MDCGA-Net: Multiscale Direction Context-Aware Network With Global Attention for Building Extraction From Remote Sensing Images

“…To address the challenge of the absence of detailed information across multiple scales at boundaries. Some researchers introduce auxiliary modules to refine the boundary information [22]. Alternatively, some other studies introduce multiscale encoder architecture [23] and the atrous spatial pyramid pooling (ASPP) [24] to obtain the multi-scale contextual information.…”

Unsupervised Building Extraction from Multimodal Aerial Data Based on Accurate Vegetation Removal and Image Feature Consistency Constraint

“…Supervised deep-learning-based methods are a possible solution to realise automatic and accurate building extraction from remotely sensed data. The rapid development of deep learning, especially convolutional neural networks [13][14][15][16][17][18][19][20][21] and transformers [22][23][24], has made deep-learning-based methods the mainstream for building extraction, and many impressive results have been achieved. However, deep-learning-based methods still rely on a large number of labelled samples to obtain satisfactory results, and these samples are often manually labelled, which is time and labour consuming.…”

“…If we can extract sufficient information about the buildings from the used data in an unsupervised manner, it is possible to design an unsupervised method to extract buildings automatically and accurately, avoiding manual data labelling and manual parameter(s) tuning. According to the data types used, existing building extraction methods can be divided into three categories: (1) methods based on remote sensing images [20,[31][32][33][34], (2) methods based on three-dimensional data (often LiDAR point clouds) [10,[35][36][37][38][39], and (3) methods that combine remote sensing images and three-dimensional data [6,16,30,[40][41][42][43].…”