Radiation Safety in Nuclear Power Plants: ResNet-Based Glove Image Classification

Ensuring the safety of personnel working in nuclear power plants remains a paramount concern, with a primary focus on mitigating radiation exposure risks. An essential safety measure involves monitoring and verifying that employees are wearing the appropriate gloves while engaged in their tasks. This study aims to design an advanced algorithm capable of accurately identifying the glove status of personnel upon their entry into the workspace by analyzing photographs taken at the entrance. Our classification system includes six categories: false detection, indeterminate, latex glove, white glove, red glove, and no glove. To enhance data diversity and training efficacy, we employ the mixup technique in data preprocessing. Our model of choice is a 50-layer residual network, and we utilize meticulously designed training methods for optimal results. Extensive data validation demonstrates the effectiveness of our approach, achieving an impressive accuracy rate of 92.84%. This research presents a valuable tool for enhancing radiation safety measures within nuclear power plants, with the potential to significantly reduce accident risks and uphold the well-being of plant personnel.