Evaluating Electro-Optical and Infrared Thermal Modalities for Counter-UAV Detection Under Daytime and Nighttime Conditions
The use of unmanned aerial vehicles (UAVs) in recent conflicts and growing threats to critical infrastructure has made reliable automatic drone detection an urgent security priority. Ensuring detection performance across diverse lighting conditions, especially nighttime scenarios where electro-optical (EO) visible-spectrum sensors are limited, remains a key challenge. This paper presents a controlled comparative study of EO and infrared thermal (IR) imaging modalities for UAV detection using the YOLO11n architecture, conducted within the NATO Science for Peace and Security Ai4CUAV project. A synchronized EO-IR dataset was collected under real-world daytime and nighttime conditions. The experiment uses paired daytime EO-IR data for direct EO-IR modality comparison, together with separate IR night-only and IR day-night datasets. Four YOLO11n models were trained under the same configuration. IR achieved higher performance than EO on paired daytime data (mAP@0.50: 0.946 vs. 0.905, F1: 0.950 vs. 0.912), supported reliable nighttime detection (mAP@0.50: 0.874, F1: 0.881), and achieved mAP@0.50 of 0.912 (F1: 0.915) when evaluated across combined day-night conditions. For the scenes and conditions represented in the evaluated dataset, both modalities perform strongly during daytime, while IR remained necessary for nighttime operation.