基于CCG-YOLOv8的施工场景下安全帽佩戴检测

Safety Helmet Wearing Detection Based on CCG-YOLOv8 for Construction Scenarios

现有安全帽佩戴检测模型对小目标和部分遮挡目标存在误检和漏检,对此提出一种基于YOLOv8的安全帽佩戴检测模型CCG-YOLOv8。首先,在YOLOv8颈部添加CA注意力机制,增强算法对特征的提取能力。其次,使用轻量级上采样CARAFE替换最近邻插值上采样,提高细节信息利用率,同时保持轻量化。最后,使用GIOU优化边界框回归损失函数,准确衡量预测框和真实框之间的重叠情况并加快模型收敛速度。结果表明,CCG-YOLOv8模型的mAP达到了85.1%,较YOLOv8提高了4.5%,有效减少了安全帽佩戴检测中的漏检与误检率,满足在环境复杂的施工场景下对小目标和部分遮挡目标的检测要求。

Current helmet-wearing detection models often misdetect or omit small and partially occluded targets.To ad-dress this issue,a helmet-wearing detection model,CCG-YOLOv8,based on YOLOv8 was presented.This model incorpo-rated a coordinate attention(CA)mechanism to the Neck of the YOLOv8 for enhancing feature extraction.It also incor-porated lightweight upsampling,i.e.CARAFE,which improved detail utilization while maintaining model efficiency.Fur-thermore,the bounding box regression loss function using GIOU was optimized to accurately measure overlap between predicted and labeled boxes,speeding up model convergence.The results demonstrate that the CCG-YOLOv8 model a-chieves an mAP of 85.1%,outperforming YOLOv8 by 4.5%.This model effectively reduces the rate of helmet-wearing misdetections and omissions,particularly for small and partially occluded targets in complex construction scenarios.