A Comprehensive Review of CNN-Based Methods for Earthquake-Induced Building Damage Detection Using Remote Sensing Imagery

Abstract

Earthquakes are highly destructive and sudden natural disasters. The primary cause of casualties and major economic losses is the collapse of buildings due to ground shaking. Quick and accurate assessment and localization of damaged buildings is crucial for the rapid deployment of post-earthquake rescue operations and disaster reconstruction tasks. This paper focuses on building damage detection, discussing the application and current developments of deep learning Convolutional Neural Networks(CNNs) in this field. Based on dual-temporal and single-temporal image sources trained by networks, this study compares the characteristics and advantages of different network architectures. In terms of single-temporal imagery-based methods, this study individually examines the detection performance and application contexts for distinct types of detection tasks: semantic segmentation, object detection, and instance segmentation. Furthermore, this paper also analyzes and compares the main types of remote sensing data used in earthquake building damage detection tasks. Finally, this study summarizes the challenges faced by CNNs in the task of automatic building damage detection and the corresponding strategies, aiming to guide future improvement directions. This review offers researchers in the field of disaster assessment and emergency response post-earthquake a reference for decision-making solutions.