Historical Long-Term Ultra-Fine-Scale Analysis of Urban Landscape Change with KH-9 PCS Images

Declassified panchromatic images from the Keyhole (KH)-9 HEXAGON panoramic camera system (PCS; hereafter KH-9 PCS) at ultra-fine spatial resolution (0.6–1.2m) during the 1970s can make a considerable contribution to urban landscape change detection research. Such images require comprehensive data curation and processing, however, due to inherent challenges. This research presents a novel multicomponent analytical framework for revealing up to a half-century of urban landscape changes with ultra-fine-scale details through pairing KH-9 PCS 1974 and ultra-fine spatial resolution (UFSR) panchromatic (PAN) aerial 2023 images. This framework includes four components: georeferencing, denoising, change detection, and quantification. First, the KH-9 PCS image was georeferenced to the UFSR-PAN aerial image using integration of SuperPoint + SuperGlue (SPSG) and spline transformation. Second, a curvelet transform framework was applied to denoise the KH-9 PCS image while maintaining its original image properties. For change detection, new bitemporal multiscale features were synthesized, and random forest and deep learning SegNet were applied to the generated features. Four continuous landscape measures were fitted to quantify any changes in the structures, patterns, and heterogeneities of the urban landscape. An ablation investigation was conducted to assess the performance of the proposed framework. Our results suggest that georeferencing and denoising procedures can increase the utility of KH-9 PCS imagery for subsequent procedures. The pairing of KH-9 PCS 1974 and recent UFSR remotely sensed data (e.g., UFSR-PAN aerial 2023) in tandem with appropriate change detection and quantification approaches thus revealed mega-urbanization over half a century in ultra-fine-scale detail.