Spaceborne SAR Data for Regional Urban Mapping Using a Robust Building Extractor

Abstract

With the rapid development of urbanization, timely and accurate information on the spatial distribution of urban areas is essential for urban planning, environmental protection and sustainable urban development. To date, the main problem of urban mapping using synthetic aperture radar (SAR) data are that nonbuilding objects with high backscattering cause high false alarms, while small-scale buildings with low backscattering result in omission errors. In this paper, a robust building-area extraction extractor is proposed to solve the above problems. The specific work includes (1) building a multiscale and multicategory building area dataset to learn enough building features in various areas; (2) designing a multiscale extraction network based on the residual convolutional block (ResNet50) and a pyramid-based pooling module to extract more discriminative features of building areas and introducing the focal loss item as the object function of the network to further extract the small-scale building areas and (3) eliminating the false alarms using the Normalized Difference Vegetation Index (NDVI) and Modified Normalized Difference Water Index (MNDWI) index. GF-3 SAR data with a 10-m resolution of four regions in China are used to validate our method, and the regional building-area mapping results with overall accuracy above 85% and kappa coefficient not less than 0.73 are obtained. Compared with the current popular networks and the global human settlement layer (GHSL) product, our method shows better extraction results and higher accuracy in multiscale building areas. The experiments using Sentinel-1 and ALOS-2/PALSAR-2 data show that the proposed method has good robustness with different SAR data sources.

Satellite-Based Human Settlement Datasets Inadequately Detect Refugee Settlements: A Critical Assessment at Thirty Refugee Settlements in Uganda

“…Much human settlement mapping research to date has focused on improving measures of urbanization [7,8,22,23], but there has been very little formal examination of the inclusion of small-scale settlements in broad-scale human settlement products (with exceptions [24,25]). As a result, it remains unclear whether rural, small-scale settlements that are home to approximately 3.4 billion people, including over half of Africa's population [26], are included in satellite-based human settlement datasets.…”

Built-Up Area Mapping for the Greater Bay Area in China from Spaceborne SAR Data Based on the PSDNet and Spatial Statistical Features

“…We can find that our method can achieve similar OA, PA, UA, and F scores in comparison with the WSF product, indicating that the proposed method can obtain satisfactory Bas extraction results based on only one single date Sentinel-1 SAR image. In order to illustrate the superiority of our proposed method over other state-of-theart deep learning methods for SAR image BA mapping, we select six regions from the whole Sentinel-1 SAR image and depict the results of our method, the BA-Unet [31] and PSPNet [30], respectively, in Figure 12. Note that all the methods use the same training samples described in the Section 2.…”

“…Considering that this work focuses on the BAs extraction from SAR data, we select the newly published WSF product for comparison. In terms of the deep learning models, we compare our proposed results with two deep learning models for SAR image BA mapping, which are BA-UNet [31] and the pyramid scene parsing network (PSPNet) model [30].…”

“…In 2017, Gao et al [29] proposed to extract BAs from SAR images on the basis of fully convolutional networks. In 2020, Li et al [30] presented a method for BAs extraction from SAR imagery by using a deep feature convolutional network, i.e., ResNet50 together with multiscale processing, which can extract multi-scale buildings. In 2021, Wu et al [31] proposed a BA mapping method for Gaofen-3 (GF-3) SAR data based on the deep learning method.…”

Satellite-Based Human Settlement Datasets Inadequately Detect Refugee Settlements: A Critical Assessment of Area, Accuracy, and Agreement at Thirty Refugee Settlements in Uganda

“…These products have broadened our foundational awareness of where humans live and work [7][8][9] and made important contributions to population modeling [10][11][12][13][14], development monitoring [8,[15][16][17], and climate hazard mitigation [18][19][20][21]. Much human settlement mapping research to date has focused on improving measures of urbanization [7,8,22,23] but there has been very little formal examination of the inclusion of small-scale settlements in broad-scale human settlement products (with exceptions [24,25]), despite approximately 3.4 billion people living in rural, small-scale settlements, including over half of Africa's population [26].…”