摘要
确保工业生产中的操作人员佩戴安全帽以降低风险至关重要。然而,传统检测方法存在小目标漏检和精度低的问题。故在原YOLOv5s模型上改进,引入多项技术提升性能。首先,融合CBAM注意力和DB_CSP模块,加强模型的关键信息提取能力。DB_CSP增强模型的特征多样性。其次,使用α-CIoU损失替代传统GIoU损失,强化目标定位和边界回归的精度,提高鲁棒性。实验表明,改进模型平均精度达96.3%,较原YOLOv5s提升1.2个百分点。本研究方法满足车间作业需求,提高安全帽检测效果。
It is important to ensure that operators in industrial production wear safety helmets to reduce risks.However,the traditional detection methods have the problems of missing small targets and low precision.Therefore,the original YOLOv5s model is improved and a number of technologies are introduced to improve the performance.Firstly,CBAM attention and DB_CSP modules are integrated to enhance the key information extraction capability of the model.DB_CSP enhances the feature diversity of the model.Secondly,α‑CIoU loss is used to replace the traditional GIoU loss,which strengthens the precision of target positioning and boundary regression,and improves the robustness.Experiments show that the average accuracy of the improved model is 96.3%,which is 1.2 percentage higher than that of the original YOLOv5s.This research method can meet the requirements of workshop operation and improve the effect of safety helmet detection.
作者
王坡
罗红旗
Wang Po;Luo Hongqi(School of Artificial Intelligence,Beijing Technology and Business University,Beijing 100048,China)
出处
《现代计算机》
2023年第24期40-45,共6页
Modern Computer
基金
国家重点研发计划项目(2017YFF0207201)。
