Industrial Safety Helmet Detection: Innovative CNN-Based Classification Approach

Abstract

This study presents the development and evaluation of a CNN-based model for detecting safety helmets in industrial settings. Utilizing a dataset from GitHub, which includes images of individuals wearing safety helmets in various industrial environments, the model was trained using the YOLOv8 architecture over 100 epochs. The comprehensive training process involved data augmentation techniques to enhance generalization capabilities. The evaluation results demonstrated high precision (0.92) and recall (0.856) for helmet detection, with an overall mAP50 of 0.766. Visual analysis through precision-confidence curves confirmed the model's high reliability in detecting helmets at higher confidence thresholds. These findings suggest that the implementation of this model in real-time monitoring systems could significantly enhance industrial safety by reducing manual inspection efforts and ensuring compliance with safety regulations