YOLORM: An Advanced Key Point Detection Method for Accurate and Efficient Rotameter Reading in Low Flow Environments

Automatic reading of rotameters in low flow and challenging environments poses substantial accuracy and efficiency challenges. To address these issues, this study introduces YOLORM, an advanced key point detection method for rotameters, built upon the YOLOv8n model. We de-veloped a comprehensive YOLO-RM dataset comprising 57,500 images to facilitate model training and evaluation. Several innovations were introduced to the YOLOv8n framework: (1) An ALKBlock-based C2f module incorporating DACov structure to enhance spatial perception; (2) An EffQA-FPN feature pyramid network inspired by QARepVGG to mitigate precision loss during quantization and reparameterization; (3) A DynamicHead attention mechanism with multi dimensional perception capabilities; and (4) EIoU and Adaptive Wing Loss functions to optimize bounding box and key point regression. Experimental results demonstrate that YOLORM achieved exceptional performance on the YOLO-RM dataset, with 99.67% precision, 98.83% recall, 99.48% mAP50, and 98.58% mAP50-95. Compared to the baseline YOLOv8n model, YOLORM achieved a 7.43% increase in detection accuracy and a 15.21% reduction in computational complexity. Moreover, YOLORM exhibited significant reductions in parameter count and computational cost while maintaining or enhancing detection performance relative to state-of-the-art algorithms. Notably, compared to Hourglass, HRNet, and HigherHRNet, YOLORM reduced parameters by 99.3%, 92.9%, and 96.8%, and computational cost by 98.8%, 90.2%, and 95.3%, respectively, while concurrently improving precision and recall. This study presents an efficient and reliable auto-matic reading solution for rotameters in industrial automation, demonstrating robust perfor-mance in complex environments. The proposed YOLORM model has the potential to significantly enhance safety and efficiency in industrial production processes.