| Abstract |
Deep learning plays a growing and crucial role on the Internet of Things (IoT), especially in intelligent data analysis, decision support, and automation control. YOLOv5, as an efficient model for target detection in deep learning, inherits the high-efficiency features of the YOLO series, and significantly improves the performance of real-time detection applications by enhancing the structure of the network and the optimization of the loss function. The article introduces the network structure of YOLOv5 and emphasizes an analysis of its advantages in target detection tasks. Then, the article details the optimization of the network structure based on YOLOv5, including the attention mechanism(CA) and the bidirectional feature pyramid network (BiFPN) structure. In addition, extensions of the evaluation metrics are explored, especially the improved IoU loss functions (e.g., CIoU, EIoU, and SIoU), which address the limitations of the standard IoU loss functions in dealing with targets of different shapes and sizes by introducing new penalty terms. YOLOv5 has been widely used in the IoT domain, and has demonstrated usefulness in the areas of smart city, agricultural monitoring, industrial automation, and environmental monitoring, etc., showing utility and effectiveness, while it has a promising application in the IoT field. By further optimizing the network structure of YOLOv5 and combining it with other AI techniques, it can drive IoT systems towards greater intelligence and automation. © 2024 IEEE. |
