A novel approach to automated cloud removal and seamless multisensor satellite image mosaicking

Persistent cloud cover, radiometric inconsistency, and computational scalability remain major challenges in generating seamless mosaics from high-resolution optical satellite imagery. Although numerous studies have addressed cloud removal, radiometric normalization, and seamline optimization individually, relatively few have integrated these components into a unified and automated workflow suitable for large-scale operational applications. This study presents a computationally efficient framework for automated cloud-free mosaicking of multispectral and panchromatic satellite imagery. The workflow combines Dynamic Radiometric Normalization (DRN), a stability-mask-based approach for robust inter-scene radiometric normalization; Sparse-Graph Seamline Optimization, which reduces the computational burden of dense shortest-path seamline searches through graph simplification; and an SSIM-guided cloud reconstruction strategy coupled with Poisson blending to improve structural and radiometric continuity. Rather than introducing entirely new low-level algorithms, the framework integrates and adapts established techniques into a scalable processing pipeline that minimizes manual intervention while maintaining radiometric and geometric consistency. The framework was evaluated using SPOT 6/7, Formosat-2/5, and Pleiades imagery over Taiwan, a region characterized by frequent cloud contamination and complex terrain. Results show cloud removal rates of 97.8% for SPOT and 89.9% for Formosat-2 imagery while maintaining strong radiometric consistency, with inter-tile RMSE values within 11.7% of the 12-bit dynamic range and global radiometric RMSE values of 6.0–9.3% across spectral bands. Seamline quality metrics, including gradient magnitude and Laplacian variance, further demonstrate smooth transitions aligned with natural landscape features. These results indicate that the proposed framework provides an effective and scalable solution for generating cloud-free, analysis-ready mosaics from large multisensor image archives, supporting regional mapping, geohazard monitoring, and time-series remote sensing applications.