【摘要】 针对矿井工作人员安全帽佩戴错检、漏检的问题，提出一种改进安全帽佩戴的检测方法，包括：在YOLOv5s基础上增加小目标检测层，以提高网络对小目标的检测效果；引入一种新的包围框相似度度量，以降低网络对小目标位置变化的敏感程度；重构模型的检测头，以加速网络收敛；重建模型中的特征提取模块，以提升网络对遮挡目标的检测能力。在自建数据集上完成消融实验，实验结果表明：改进后模型的识别精度较原YOLOv5s模型，平均精确率提升了2.1%，平均查全率提升了3.0%，平均查准率提升了1.9%。研究表明，改进后模型具备良好的检测精度，适用于复杂情况下的安全帽佩戴检测任务，对于保证工作人员安全具有积极意义。更多还原

【Abstract】 Aiming at the incorrect and missed detection of safety helmets worn by mining workers, an improved detection method for safety helmets was proposed, including the adding of a small target detection layer on the basis of YOLOv5s to improve the network’s detection performance for small targets, the introduction of a new similarity metric for bounding boxes to reduce the sensitivity of the network to changes in the location of small targets, the reconstruction of the detection head of the model to accelerate network convergence and the rebuilt of the feature extraction module in the model to enhance the network’s ability to detect the occluded targets. The experimental results of this ablation experiment completed on the self-built dataset show that the improved model has a recognition accuracy improvement of 2.1% on average, 3.0% on average recall, and 1.9% on average precision compared to that of the original YOLOv5s model. The research shows that the improved model has good detection accuracy and is suitable for detecting the helmet wearing in complext situation, which will exert a positive significance for ensuring the safety of workers.更多还原