Automated Classification of Military Aircraft Using Deep Neural Networks

Abstract

Military aircraft detection in aerial imagery is critical for defence operations, airspace management, and automated surveillance. This paper presents a dataset of 74 military aircraft types, including high-value models like the F-35, B-52, and Rafale, annotated with precise bounding boxes across diverse conditions. The proposed approach, leveraging YOLOv10, achieves a precision of 82 %, recall of 66.4 %, and an F1 score of 0.687. Model evaluation yields a Mean Average Precision (mAP) of 76.4 % at Intersection over Union (IoU) 0.5 and 68.7 % across IoU 0.5–0.95, demonstrating robust detection performance. Real-time feasibility is ensured with an inference speed of 3.8 ms per image. Confusion matrices, PR curves, and annotated results highlight model strengths and areas for improvement, particularly in distinguishing visually similar aircraft. This study positions YOLOv10 as a strong candidate for real-time military aircraft recognition, contributing to defence surveillance and threat monitoring advancements.