Density saliency for clustered building detection and population capacity estimation

摘要

Building detection is a critically important task in the field of remote sensing and it is conducive to urban construction planning, disaster survey, shantytown modification, and emergency landing, it etc. However, few studies have focused on the task of the clustered building detection which is inescapable and challenging for some relatively low space resolution images. The appearance structures of those buildings are not clear enough for the single-building detection. Whereas, it has been found that the distributions of clustered buildings are mostly dense and cellular, while the backgrounds are not. This clue will be beneficial to the clustered building detection. Motivated by the fact above and other similar density estimation applications, this work mainly focuses on the information mining mechanism of dense and cellular structure. Firstly, we propose a concept of Clustered Building Detection (CBD), which contributes to develop clustered building detection techniques of remote sensing images. Secondly, a saliency estimation algorithm is proposed to mine the prior information for the clustered buildings. Thirdly and most notably, combining with the CBD and the density saliency map, a Population Capacity Estimation (PCE) method is presented. The PCE can be easily used to estimate the population carrying capacity of certain areas and future applied for national land resource management. Moreover, a Clustered Building Detection Dataset (CBDD) from Gaofen-2 satellite is annotated and contributed for the public research. The experimental results by the representative detection algorithms manifest the effectiveness for the clustered building detection. (c) 2021 Elsevier B.V. All rights reserved.