V3I3P46

NIGHTWING: Real-Time Infrared Detection Framework for Nocturnal Migratory Birds Using Spatiotemporal Anomaly Modeling to Enhance Aviation Safety

Amber Sophia A. Perez¹, Philip Andrew C. Malazarte^1*, Ace Jortherson L. Tejano¹, Aniceto B. Ipanag, Jr. ², Myrill Guia A. Bautista¹, Henry F. Haranay¹

Abstract

Bird strikes constitute a significant aviation safety challenge, particularly during nocturnal operations where conventional detection systems are often ineffective. Current thermal imaging technology, which offers the best modality for night-time sensing, lacks the computational efficiency and speed required for real-time deployment on resource-constrained hardware. To address this critical gap, we developed NIGHTWING, a lightweight, dual-stage framework that utilizes infrared (LWIR) signatures for robust, real-time hazard detection. The framework’s core detector, YOLOv8n, achieved 99.5% mAP@0.5 on the primary validation set; however, performance dropped to 51.05% mAP on an independent challenging subset, indicating a significant generalization gap. The system demonstrated high computational efficiency, achieving 90.9 FPS, which exceeds the 5 FPS operational threshold. The spatiotemporal anomaly system generated critical alerts within 150 milliseconds. While YOLOv8n outperformed alternative architectures in our evaluation, baseline models (EfficientDet-D0 and MobileNetV2-SSD) failed to converge properly, rendering comparative accuracy claims invalid. These results highlight both the promise and limitations of lightweight thermal detection systems for nocturnal wildlife monitoring.

Keywords:

avian detection; thermal imaging; YOLOv8; spatiotemporal anomaly; aviation safety; nocturnal migration; edge AI