Real-time flying object detection with YOLO v8,9,10

Abstract

This paper presents a general model for real-time flying object detection that can be used for transfer learning and further research, as well as a refined model that achieves state-of-the-art results in detecting flying objects. The study trains a dataset containing 40 different types of flying objects with fine-tuned data to address several practical challenges, such as objects appearing at long distances due to camera placement, flying objects frequently observed against sky backgrounds, and visually similar objects (e.g., AG600 and US-2 seaplanes), which make accurate identification difficult. The paper focuses on the application of the YOLO model using its latest versions, including versions 8, 9, and 10. Experimental results demonstrate that the YOLOv8-N model outperforms YOLOv9-T and YOLOv10-N, achieving an mAP50 of 91.1% and an mAP50–95 of 87.3%. In addition, the study shows that the YOLOv10-N model can be effectively customized for object counting tasks under predefined constraints, enabling reliable control of the number of detected objects within a frame. These results indicate the potential applicability of the proposed approach to airport monitoring and aviation security and safety.