An improved YOLOv11 algorithm for personal protective equipment detection

To address the issues of low efficiency in manual detection of personal protective equipment (PPE), missed detection of small objects, and poor adaptability to complex scenarios, this paper proposes the COU-YOLOv11 detection model based on an improved YOLOv11. The innovations lie in three aspects: First, a dual-branch residual structure is reconstructed from the UIB module of MobileNetV4 to enhance cross-channel feature interaction capabilities; second, a cascaded group attention (CGA) mechanism is introduced, using input segmentation and cross-head cascading strategies to optimize feature diversity, reducing computational redundancy while improving model capacity; finally, the dynamic scaling OREPA module replaces traditional convolutional layers to strengthen multi-scale feature capture capabilities. Experiments show that the model significantly improves detection accuracy on the PPE detection dataset, especially in small object recognition and complex background scenarios, providing an efficient solution for intelligent PPE safety detection.