To address the high deployment costs associated with existing safety helmet detection algorithms due to platform resource constraints, an improved YOLOv8-based helmet-wearing detection algorithm, Sub-YOLOv8, has been proposed specifically for substation environments. This approach integrates MobileNetv3 with YOLOv8 while incorporating an efficient enhanced feature pyramid network (EFPN) and a parameter-free attention mechanism, SimAM. These improvements collectively reduce the number of parameters and computational complexity in the backbone network. The enhanced feature pyramid network significantly boosts the model's sensitivity to key features, while SimAM reduces parameter counts and improves detection accuracy for small safety helmets. As a result, the proposed approach achieves low computational requirements and a minimal parameter count, placing minimal demands on execution platform resources. The optimized model size is reduced to 4.39 MB, substantially cutting network deployment costs and broadening its applicability in real-world scenarios. Future research will focus on enhancing detection precision and exploring deployment possibilities on hardware devices, aiming to develop a more convenient and efficient safety helmet detection system for practical applications.