Evaluation of the YOLO architecture in object detection from aerial images

Authors:

Abstract
This paper evaluates the effectiveness with which YOLO structures perform object detection in aerial imagery, with an emphasis on applications related to intelligent traffic control. The study compares the performance of Yolov5, YOLOv7, YOLOv8 and YOLOv9 to determine which model is the most accurate. The process involved collecting and annotating a dataset of aerial imagery and then training and evaluating each model. According to the results, the accuracy of Yolov5Aug, which was so named because of its data augmentation and more robust training, was 85% as was that of YOLOv8Aug, while that of YOLOv9Aug and YOLOv5 was 84%. These results demonstrate the effectiveness of YOLO designs in intricate contexts such as aerial imagery. In summary, Yolov5Aug and YOLOv8Aug outperforms Yolo5 and Yolo8 designs in identifying aerial objects, which supports the high accuracy and effectiveness of YOLO architectures. These results hold great promise for future research and real-world applications in traffic control and management systems.