To address the issues of low detection accuracy, false positives, and missed detections found in traditional reflective vest detection algorithms used in construction, military, and defense technology, we propose a PDEW-YOLOv8n algorithm suitable for reflective vest detection. This algorithm adds a small object detection layer and detection head to improve accuracy. Additionally, ordinary convolutions in the backbone network are replaced with deformable convolutions to accommodate the irregular deformations of reflective vests. An Efficient Multi-Scale Attention Module (EMA) is introduced in the neck network, enhancing the model's focus on reflective vest information and improving the network's feature representation capability. Furthermore, the loss function CIOU is replaced with WIOU to accelerate network convergence. Results indicate that compared to YOLOv8n, the PDEW-YOLOv8n model maintains similar complexity while achieving increases of 0.6%, 5.7%, 4.2%, and 5.2% in accuracy, recall, mAP@0.5, and mAP@0.5:0.95, respectively. Thus, it effectively meets the requirements for low-complexity and high-accuracy real-time reflective vest detection.