YOLO-METER: An efficient and accurate detection method of pointer meter in airport energy station

Abstract

Accurate assessment of electromechanical system status is essential for the safe and efficient management of airport energy stations, where pointer-meter readings serve as key operational indicators. To address pointer-meter detection under complex lighting and cluttered backgrounds, this study proposes YOLO-METER, a vision-based detection model tailored for airport energy stations. The model integrates a Triple Attention Mechanism (TAM) in the backbone to suppress background interference and employs a weighted bidirectional feature pyramid network (BiFPN) in the neck for efficient multi-scale feature fusion. Furthermore, an Improved Sparrow Search Algorithm (ISSA) is used to optimize 12 hyperparameters, substantially improving convergence and detection performance. An inspection-robot platform was built, and on-site images were collected to construct a dedicated pointer-meter detection dataset. Experimental results show that YOLO-METER achieves mAP@0.5 of 97.6%, Precision of 96.46%, and 224.8 FPS, outperforming multiple YOLO variants. These results indicate that YOLO-METER provides an effective and efficient solution for real-time pointer-meter detection, supporting autonomous inspection in airport energy stations.