MicroSlicer: Efficient Inference-Time Slicing for Small-Target Detection in Thermal UAV Search-and-Rescue Imagery

Andrei-Gabriel Manta, Costin-Emanuel Vasile, Călin Bîră

MDPI Journal of Imaging

JurnalQ2IF 3.8În curs de publicare

Abordări noi de programare pentru acceleratoare hardware

Thermal drones help search for missing people in the wilderness, but flying high enough to cover wide areas shrinks each person to just a few pixels, causing standard detectors to miss them. We introduce MicroSlicer, a lightweight method that improves small-target detection by slicing each frame into overlapping tiles at inference time, with no extra training required. Tested on three detector architectures (YOLO26-nano, YOLO26-small, RT-DETR-Large) using a thermal search-and-rescue dataset, MicroSlicer detects more small targets than standard full-frame inference and matches or beats an existing slicing method, SAHI, while running as fast or faster. It is especially strong on the transformer-based detector, where SAHI’s accuracy and speed both collapse at low confidence thresholds while MicroSlicer stays stable. We also show that training directly on cropped patches is less effective than simply combining full-image training with MicroSlicer, by up to 24.8 percentage points in accuracy.