<![CDATA[The level of compliance with the use of Personal Protective Equipment (PPE) in the work environment, especially in the field, remains a serious challenge with a direct impact on worker safety. Manual supervision is considered ineffective because it is subjective and limited in scope, time, and space. This research aims to develop an automatic PPE detection system using deep learning that can recognize five main objects: helmets, vests, gloves, shoes, and people. The system was designed to automatically detect real-time PPE availability and record workers' occupational safety status in the field. This research used the Convolutional Neural Network (CNN) with the YOLOv11 nano variant architecture, due to its advantages in efficiency and inference speed. The dataset comprised 1,471 images collected from PT PLN Aceh field documentation and the Roboflow Universe platform, which were expanded to 7,310 images following data augmentation. The dataset was split into training (96%), validation (2%), and testing (2%) subsets. The model was trained for 150 epochs and deployed on a Raspberry Pi 4 B for real-time inference. Evaluation results show a mean Average Precision at IoU 0.5 (mAP@0.5) of 90%, precision of 91.8%, and recall of 82%. The deployed system operates at 5–8 frames per second (FPS) and automatically logs worker safety status to Excel reports, demonstrating its practicality for real-time occupational safety monitoring.]]>
Copyrights © 2026
