YOLO–DE: An enhanced YOLOv11 model for the study of pipeline flow monitoring technology

Abstract The traditional flow monitoring methods rely on a single dimension of a sole physical quantity for flow measurement, resulting in insufficient measurement accuracy and high susceptibility to changes in flow conditions. To solve these problems, an enhanced target detection model, YOLO-DE, for pipeline flow monitoring is proposed, based on the YOLOv11 classify task model (YOLOv11-cla) with dual optimization. First, the C3k2 is reconstructed module by introducing Deformable Convolution into the backbone network, enhancing feature extraction capability for complex pipeline structures through dynamic receptive field adjustment. Secondly, the spatial-channel synergetic attention module (PSASCSA) is introduced to effectively suppress background noise interference through cross-dimensional feature recalibration. The dataset obtained from the pipeline flow monitoring experiments demonstrate that YOLO-DE outperforms the original YOLOv11-cla model. Specifically, the classification precision improves from 0.884 to 0.917, and the F1-score rises from 0.892 to 0.917. These results validate the engineering applicability of the proposed optimization strategy in industrial inspection scenarios. This work provides an intelligent technical solution for the efficient and secure operation of heating systems.