Hybrid YOLOv8 and SSD for Real-Time Digitalization of PPE Usage Compliance Detection in Workers (Non-Maximum Suppression Method)

This study aims to develop an automated deep learning-based system to monitor compliance with the use of Personal Protective Equipment (PPE) in the manufacturing industry. Manual monitoring, which has been carried out so far, is considered inefficient and prone to error. This system compares three approaches: the YOLOv8 model, SSD Mobile Net, and a hybrid method that combines the two. The dataset consists of 700 images covering eight classes related to PPE use. The results show that the hybrid method performs best with: 1. Accuracy: 95.1%, 2. Precision: 98.7%, 3. Recall: 97.2%, and F1-Score: 94.5%. Although its detection speed (18 FPS) is slightly lower than SSD (29 FPS), its detection quality is superior. The system has been implemented in a web application that can run in real-time using a webcam, equipped with an alarm and “SAFE” or “NO SAFE” notifications. This system is expected to be an accurate and efficient digital solution to improve work safety.