FastGAS: a UAV-Enabled framework for fast and robust gravel auto-sieving in coastal and mountainous fluvial environments

The grain size distribution (GSD) of gravels plays a crucial role in understanding fluvial processes in coastal and mountainous areas. Conventional UAV-based sieving algorithms face two limitations: (1) Pixel calibration errors caused by elevation-dependent scaling in areas with significant slope variations; (2) Challenges in the efficiency and accuracy of batch processing of cyclical monitoring images. This study presents a fast gravel automated sieving (FastGAS) method incorporating calibration spheres to establish pixel-size correspondence, simultaneously reducing slope-induced calibration errors and serving as waypoint benchmarks. The proposed framework enables automatic batch processing through sphere detection, combined with optimized seed generation and four neighborhood search algorithms for efficient gravel segmentation and size distribution inversion. Validation conducted with 35 images from coastal and mountainous fluvial environments demonstrated strong agreement with manual measurements (NRMSE = 0.07–0.58), further confirmed by one-month continuous monitoring in Moon Bay, Yantai City, China. Comparative analysis showed FastGAS outperformed PebbleCountsAuto (0.24–0.98), pyDGS (0.57–2.69), and SediNet (0.73–9.13) in accuracy while maintaining competitive processing speed (28 s vs. SediNet’s 10 s). The method’s advantages in precision, stability, and computational efficiency suggest its strong potential for automated long-term monitoring of coastal and mountainous rivers.