摘要
为了解决安全帽检测算法推理时间长,小目标检测漏检等问题,提出一种改进的YOLOv7实时检测模型。以EIOU(Efficient Intersection Over Union)损失作为损失函数,引入焦点损失,聚焦于高质量的锚箱,达到加快收敛速度、提高回归精度的目的。同时采用全局注意力机制,通过减少信息损失并增强全局交互表示来提高深度神经网络的性能。实验结果表明,改进后的模型m AP@0.5为0.951,比YOLOv5s、YOLOv7和YOLOv7-tiny模型分别高出3.93%、2.92%、5.32%,改进后的YOLOv7模型可以有效地实现复杂环境下安全帽的精确检测,能够准确识别工人安全帽佩戴情况,从而大大降低施工现场的安全风险。
To address the issues of long inference time and missed detection of small objects in safety helmet detection algorithms,an improved real-time detection model of YOLOv7 is proposed.The Efficient Intersection Over Union(EIOU)loss function is introduced with focal loss to focus on high-quality anchor boxes,aiming to accelerate the convergence speed and improve the regression accuracy.Additionally,a global attention mechanism is adopted to reduce information loss and enhance global interaction representation,thereby improving the performance of deep neural networks.Experimental results show that the improved model achieves an mAP@0.5 of 0.951,which outperforms YOLOv5s,YOLOv7,and YOLOv7-tiny models by 3.93%,2.92%,and 5.32%,respectively.Therefore,the improved YOLOv7 model can effectively achieve precise detection of safety helmets in complex environments,accurately recognize whether workers are wearing safety helmets,and thereby greatly reduce safety risks at construction sites.
作者
贺洪江
贺换杨
王双友
HE Hong-jiang;HE Huan-yang;WANG Shuang-you(College of Information and Electrical Engineering,Hebei Engineering University,Handan 056000,China)
出处
《计算机仿真》
2025年第4期309-314,362,共7页
Computer Simulation
基金
国家自然科学基金项目(61802107)。
关键词
目标检测
安全帽
注意力机制
Object detection
Safety helmet
Attention mechanism
