A stability detection technique for subway platform screen doors based on an improved YOLOv8 algorithm

Platform screen doors (PSDs) are critical safety components in modern urban rail transit systems, ensuring passenger protection and operational efficiency. However, traditional methods for assessing the mechanical stability and operational reliability of PSDs rely heavily on manual inspection, which is time-consuming and prone to error. This paper proposes a high-precision, real-time detection framework based on an improved YOLOv8 object detection algorithm tailored to the characteristics of PSD operation. The proposed model incorporates attention-enhanced feature fusion and anchor-free adjustment mechanisms to improve detection of subtle door anomalies, such as misalignment, jitter, and asynchronous closing. A dedicated dataset of PSD operation videos from real metro environments is constructed for training and evaluation. Experimental results show that our method achieves a significant improvement in detection accuracy and speed compared to standard YOLOv8 and other state-of-the-art algorithms. This research provides an effective and scalable solution for intelligent maintenance of subway infrastructure.