Empirical study based on electro-optical and infrared mixed spectrum imaging: YOLO26 versus YOLO11 for detection of small targets in SAR operation

Detecting sub-20-pixel targets in maritime surveillance presents a critical challenge for Search and Rescue (SAR) operations, where drowning persons and survival equipment appear as microscopic objects in UAV-captured imagery. This paper presents an empirical study comparing YOLO26 and YOLO11 architectures for small-target maritime detection. Using a mixed dataset of 21,288 images spanning electro-optical (EO) and infrared (IR) modalities with five operational classes, we demonstrate that YOLO26m achieves 85.7% mAP50 and 78.2% overall recall, with 94.1% mAP50 for vessel detection in IR — substantial gains over YOLO11s (60.2% mAP50). YOLO26 introduces STAL (Small-Target-Aware Label Assignment) and NMS-free end-to-end inference, directly addressing the recall degradation on sub-pixel scale targets. Experimental results validate that YOLO architecture evolution is crucial for maritime surveillance applications, justifying migration to YOLO26 for the next generation of onboard systems in maritime surveillance applications.