TDSO: a new hybrid method for temporal detection of spatial objects from large-scale satellite images

Accurate identification of buildings and roads in large-scale satellite imagery is critical for disaster management, urban transformation, and city planning. This study proposes a TDSO (Temporal Detection of Spatial Objects) approach that considers the temporal changes of spatial objects using different-temporal satellite images from 2020 and 2024. Satellite images were obtained from the server infrastructure of Kayseri Metropolitan Municipality; to make the high-volume data processable, they were published as WMS services via the GeoServer platform and decomposed into sub-images with dimensions of 256 × 256 pixels, preserving geographic reference information, using the TileCache system. Mask R-CNN and YOLO architectures were used to detect building and road objects on the resulting tile images, and object detection performance was evaluated with object-based metrics under different image quality conditions (PNG (0.99), JPEG (0.99), and JPEG (0.75)). Experimental findings show that Mask R-CNN exhibits more balanced and stable performance than YOLO across all scenarios, notably achieving the highest performance under JPEG (0.99) conditions, with F1-Scores of 83.63% and 88.51% for building and road detection, respectively. Furthermore, it has been demonstrated that newly constructed and demolished structures can be reliably detected by integrating asynchronous images; the proposed approach provides a robust decision-support infrastructure for post-disaster damage assessment, urban transformation, and sustainable city planning applications.