To address the difficulties in detecting reflective vest wearing in apron monitoring scenarios caused by tiny target dimensions, strong background noise interference, and varied personnel postures, a reflective vest detection method for apron workers based on an improved YOLOv11n is proposed. First, the detection head scale was reconstructed by adding a high-resolution P2 detection layer and pruning the redundant P5 layer, so that the texture and contour features of distant tiny targets were preserved. Second, an asymmetric padding convolution module was introduced to construct the backbone network, and asymmetric receptive fields were utilized to enhance the geometric feature extraction capability for the varied postures of non-rigid human targets. Third, a dilated residual structure was integrated with a context aggregation mechanism. While the efficiency of multi-scale feature aggregation was improved, feature weights were dynamically calibrated via global context information to suppress environmental background noise. Finally, an adaptive threshold focal loss function was adopted to resolve the issue of severe imbalance between positive and negative samples under large backgrounds. Experimental results based on a self-built dataset show that the precision, recall, and mAP@0.5 of the improved algorithm reach 87.4%, 75.8%, and 84.3% respectively. Compared with the baseline YOLOv11n model, these represent improvements of 2.5, 27.2, and 13.3 percentage points, respectively, with a 33.8% reduction in parameter count. A theoretical basis and technical support for the refined safety supervision of smart airports are provided by this method.