Abstract
Change detection is essential for applications such as urban planning, environmental monitoring, and disaster response. Despite advancements in high-resolution satellite imagery, accurate change detection remains challenging due to increased landscape heterogeneity and variable atmospheric conditions. The Mamba model, an efficient state-space model-based architecture, has shown promise in capturing spatiotemporal relationships in high-resolution datasets, addressing the limitations of traditional methods that struggle with the diverse appearances of urban structures. This research investigates applying Mamba to multitemporal Korea Multi-Purpose Satellite (KOMPSAT) imagery, using both real and synthetic data from SyntheWorld, a dataset developed to simulate various change scenarios. This study introduces a synthetic data-augmented mamba-based change detection algorithm (SAMBA), designed to detect structural changes in urban environments using KOMPSAT-3A satellite imagery. The main objectives are to evaluate the Mamba binary change detection (MambaBCD) model’s ability to detect building changes in KOMPSAT-3A images and assess the impact of synthetic data augmentation on performance. Experimental results with MambaBCD-Small and MambaBCD-Tiny models indicate that synthetic data incorporation improves generalization in complex settings, achieving high performance across multiple data and model configurations. Notably, the MambaBCD-Tiny model, with or without synthetic augmentation, outperformed the larger-parameter MambaBCD-Small model, demonstrating enhanced sensitivity in detecting satellite image changes. Performance evaluation metrics yielded an overall accuracy of 99.73%, precision of 98.34%, recall of 96.54%, F1-score of 97.43%, intersection over union of 95.00%, and Kappa coefficient of 97.29%. These metrics were similarly used to test the SAMBA algorithm’s generalization on benchmark change detection datasets, showcasing its potential as a robust tool for highresolution image change detection.