GIPPE-YOLO: a global-aware and IoU-optimized method for personal protective equipment detection

The detection of Personal Protective Equipment (PPE) in complex construction site scenes faces significant challenges. Due to issues such as occlusion and lighting conditions, PPE detection is often affected by complex backgrounds, causing detectors to confuse PPE with background. This leads to misclassification and hinder real-time warning systems. To address this challenge, we propose GIPPE-YOLO, a Global-aware and IoU-optimized PPE detector based on YOLO11. Specifically, we design an enhanced Global Spatial Pyramid Pooling Fast (GSPPF) module. By adding two parallel pooling layers and introducing a Universal Perception Large-Kernel ConvNet Block (UnipLKNetBlock), our model can better capture global image features and expand the receptive field. Additionally, we improve the detection box localization accuracy by combining Inner-IoU with Complete IoU (CIoU), while using auxiliary bounding boxes to accelerate training convergence. Comprehensive experiments on the SHD and PPEs datasets demonstrate that GIPPE-YOLO outperforms state-of-the-art models, achieving an increasement of 1.3% and 1.4% in mAP50 on the SHD and PPEs dataset, respectively. These results confirm that our method effectively enhances PPE detection robustness in real-world complex environments.