Evaluating YOLO V8 Models for Efficient Drone Detection in Anti-UAV Systems

The rising UAV air presence in areas of restricted airspace now raises concerns in terms of security, safety, and privacy issues. An efficient anti-UAV detection system has gained a high demand. The paper delves into the presentation of a comparative analysis between two variants of YOLOv8, YOLOv8-S and YOLOv8-L, which may be used for the purpose of UAV detection. The performance of both models was checked based on the following parameters: accuracy, precision, F1-score, and mean average precision (mAP). YOLOv8 brings some new innovations compared to previous YOLO variants, such as improved feature extraction, better backbones, improved prediction heads, and so on, resulting in enhanced detection performance. These comparisons indicate that both the variants, YOLOv8-S and YOLOv8-L, are doing outstandingly well at the detection of small, fast-flying drones and this is with huge differences in their precision-speed trade-off. Some degree of distinction between the different variants of YOLOv8 can be observed in terms of size and complexity of the model. Although larger models tend to be more accurate, they are also slower, while the small, less complex models are faster, albeit less accurate. The comparison helps in identifying the trade-offs and guides selection based on optimal models for a range of anti-UAV applications such as airspace security, wildlife protection, border surveillance, and public event monitoring.