Rapid image recognition for construction worker hard hat compliance – a practical application of convolutional neural networks

Head injuries often result in severe and potentially life-altering injuries, highlighting the importance of head protection. However, limited attention has been given to practical deployment constraints and the need for reliable performance in dynamic construction environments. This study utilizes machine learning (ML) image recognition models to detect construction worker compliance with wearing head protection correctly. This study addresses this gap by developing a practical VGG16-based image classification framework, balancing accuracy, computational efficiency, and deployability under limited data conditions. The VGG16 model effectively analyzes inconsistencies between predicted and actual values, with prediction accuracy, precision, and recall rates respectively of 98.83%, 99.16%, and 98.5%. The results indicate that the proposed custom VGG16 model can quickly and accurately determine worker compliance with wearing hard hats properly. This study makes several contributions: First, it addresses the gap between traditional image recognition approaches and deep learning-based methods. Second, it provides a methodological contribution by exploring deep learning performance. Third, it offers practical insights into the trade-off between accuracy and computational performance, showing that near-real-time monitoring can be achieved without high-end hardware. Finally, the study contributes to practical implementation by proposing a framework for integrating image classification models into construction safety management workflows.