Building Instance Change Detection from High Spatial Resolution Remote Sensing Images with Improved Instance Segmentation Architecture

The detection of ground surface changes can provide essential and valuable information for experts in the fields of geomatics, emergency management, and urban management. Many deep learning models have been proposed for detecting the change from remote sensing images. However, the majority of studies have not been able to simultaneously accomplish both building change detection and instance segmentation of changed buildings from high spatial resolution images. The information of the building instance change in high spatial resolution remote sensing images is crucial for disaster assessment and urban building management. To address these issues, we proposed a novel building instance change detection architecture based on the Cascade Mask R-CNN framework and designed a swin transformer siamese (STS) backbone to input the bi-temporal images and to improve the detection accuracy, called STS-ConcCMR. Due to the lack of remote sensing mask datasets, two building change detection datasets are transferred into the building instance change detection task. Our method achieves the best performance in F1 of90.7 and 94.5 points on the LEVIER-CD and WHU-CD datasets, respectively. And the performance of the proposed architecture can be effectively improved by substituting the siamese backbone for the non-siamese backbone. The AP and F1 are improved by 0.5 and 0.4 points on the LEVIR-CD test set, respectively, and by 2.6 points and 4.0 points on the WHU-CD test set. Experimental results demonstrated the superiority of the proposed architecture.

I. Introduction

The ground surface is constantly changing due to sudden phenomena or gradual ones. The detection of ground surface changes provides valuable information to geoscience and geomatics experts, called change detection (CD). The technology has important applications in ecological protection [1], environmental monitoring [2], and urban expansion [3]. Buildings are essential places for human activities, building change information is crucial when we detect various categories of change. A fine detection of building change is very helpful in land management, geographic information updating, and disaster assessment. Today, remote sensing images have a spatial resolution of sub-meters, which enables us to clearly distinguish each building in the images. However, the majority of studies have not been able to simultaneously accomplish both building change detection and instance segmentation of changed buildings from high spatial resolution images. And the information of the building instance change in high spatial resolution remote sensing images has not been detected so far. The information is valuable for disaster assessment and urban building management [4]. China suffers from floods almost every summer, and an unprecedented rainstorm hit Henan in 2021. In addition, China is one of the countries with frequent earthquakes. The Wenchuan earthquake caused the collapse of about 5 million houses. The high spatial resolution remote sensing images help implement damage assessment and disaster situation surveys, which can save the workforce and reduce risk. With the continuing development of cities, illegal buildings have become a more prominent problem in urban building management [5]. To address these issues, it is an excellent way to simultaneously accomplish both building change detection between bi-temporal images and instance segmentation of changed buildings for distinguishing each instance, which is called building instance change detection (BICD). This is an essential step in assessing building damage, surveying disaster situations, and identifying illicit buildings.