Method of detection the quadcopters and octocopters based on YOLOv8 model
M.D. Kamak, Volodymyr Kazymyr, Dmytro Kamak · Mathematical machines and systems · 2024
The relevance of the topic of the article is determined by the growing prevalence of unmanned aerial vehicles (UAVs) not only in the civilian but also in the military field, where they create significant problems for air defense. The paper presents a practical test of the possibility and effectiveness of using the YOLOv8 model for detecting quadcopters and octocopters in video streams, which is designed for real-time object detection and image segmentation. A training method is used that adapts the YOLOv8 model by applying a «transfer learning» approach to adjust pre-trained YOLOv8 weights to drone-specific data. The ability of the model to work ef-fectively in various environmental conditions is investigated, too. The conducted experiments demonstrate the efficiency of the developed method to accurately identify UAVs in various sit-uations, which makes it important for improving aerial surveillance tools and security mecha-nisms. The study also explores the model’s adaptability to changing observation conditions, en-suring its robustness in processing images and video. The results indicate the high potential of the YOLOv8 model in enhancing the capabilities of air defense systems and bolstering security measures against UAV threats. Additionally, the article discusses the computational efficiency of the model, highlighting its ability to process video streams in real time with minimal compu-tational resources. The precision and recall metrics of the model are evaluated, demonstrating its ability to accurately detect UAVs while minimizing false positives. Overall, the findings un-derscore the importance of leveraging advanced machine-learning techniques for effective UAV detection.