PPE Recognition System Based on Improved YOLOv11-Safety Algorithm and Frontend-Backend Architecture Integration

With the expansion of global urbanization, the construction industry, a pillar of the national economy, remains a high-risk sector. Although the number of work safety accidents and fatalities in China decreased year-on-year in 2024, housing construction projects still have high accident and fatality rates. Personal Protective Equipment (PPE) is the last effective defense for workers, but PPE compliance management faces challenges such as limited manual supervision and complex working environments. Leveraging advancements in AI and computer vision, this research designs an AI-driven deep learning system centered on the YOLOv11-Safety algorithm to realize real-time and accurate identification of 9 categories of PPE-related targets on construction sites. The system integrates Spring Boot (backend), Vue.js (frontend), and Flask (AI middleware) to build a practical intelligent safety supervision solution. Through improvements in feature extraction, feature fusion, loss function, and inference optimization, the YOLOv11-Safety algorithm achieves higher detection accuracy, faster inference speed on edge devices, and stronger robustness. The system promotes the digital transformation of construction site safety management from passive response to active prevention, providing an effective intelligent solution for workplace safety supervision.