In photovoltaic power plants, high voltage areas are common, and it is a safety requirement for workers to put on safety helmets when accessing these areas. However, manually detecting compliance with this safety measure is labor-intensive and costly. This paper proposes an improved YOLOv5 safety helmet monitoring system that employs real-time detection. The proposed system leverages the Focal_EIoU_Loss loss, which improves the base YOLOv5 by enabling it to learn high-quality samples and mitigating false detection probability. Additionally, the C2F module is introduced to enhance the network's feature extraction and reduce the likelihood of missing detections. Experimental results show that the improved YOLOv5 model outperforms the base model by .4% in the mAP@.5 index and 1.1% in mAP@.5:.95:.05 index.