Vision-based object recognition opens up new possibilities for regulating PPE use by workers exposed to hazardous conditions in industries such as medicine, mining, and construction. With similar color texture and significant size differences between various PPE components for medical staff, existing object detection methods cannot meet the demand. This paper proposes a model named MSZ-YOLO that applies to PPE donning detection for medical staff. This model first addresses the problem of easy confusion between medical staff PPE components due to similar local color textures by adding the BOCSP Trans module to the CSPDarknet backbone network and establishing global information dependencies. The module can combine the local features of the object with global information to assist the detector in identifying differences between components. To address the challenge of large scale disparities among PPE components, a weighted bidirectional mechanism is employed for integrating feature representations across multiple spatial resolutions. In the paper, we also constructed a medical staff dress code dataset for model training and testing. With an experimentally validated mAP of 94.8%, the MSZ-YOLO model offers both high accuracy and real-time performance, making it well-suited for implementation in a wide range of clinical environments.