基于改进YOLOv5算法的创伤伤情识别与定位研究

Trauma condition identification and localization based on improved YOLOv5 algorithm

下载PDF

导出

摘要目的:为减少因创伤患者伤情复杂多变而导致的误检和漏检,提出一种基于注意力(attention)机制的YOLOv5算法。方法:以YOLOv5算法为基本框架,分别在特征提取网络和特征融合网络末端嵌入自注意力模块,并在特征融合网络中引入卷积注意力机制模块,构建YOLOv5-attention算法。在Kaggle平台上对YOLOv5-attention算法进行训练验证,并与Fast-RCNN、YOLOv5算法对骨折部位的识别效果进行比较。结果:YOLOv5-attention算法对创伤患者骨折部位识别的平均精度为0.8598,优于Fast-RCNN算法(平均精度为0.6975)和YOLOv5算法(平均精度为0.8471)。结论:YOLOv5-attention算法检测准确率高、鲁棒性好,能够有效识别和准确定位创伤患者伤情。 Objective To propose an attention mechanism-based YOLOv5 algorithm to relieve the wrong or missed diagnosis due to the complexity and variability of trauma conditions.Methods A YOLOv5-attention algorithm was constructed with YOLOv5 algorithm as the basic framework,which introduced the convolutional attention mechanism module into the feature fusion network and embedded the self-attention module at the end of the feature extraction network and the feature fusion network,respectively.The YOLOv5-attention algorithm was trained and validated on the Kaggle platform and compared with Fast-RCNN and YOLOv5 algorithms for determining fracture sites.Results The YOLOv5-attention algorithm achieved an average presicion of 0.8598 for fracture site determination,which behaved better than Fast-RCNN algorithm with an average presicion of 0.6975 and YOLOv5 algorithm with an average presicion of 0.8471.Conclusion The YOLOv5-attention algorithm with high accuracy and robustness can identify and locate trauma conditions effectively and accurately.

作者王钰姝粘永健彭雪谢锦齐君谭瑶 WANG Yu-shu;NIAN Yong-jian;PENG Xue;XIE Jin;QI Jun;TAN Yao(Department of Emergency Medicine,the First Affiliated Hospital of Army Medical University,Chongqing 400038,China;Department of Biomedical Engineering,Army Medical University,Chongqing 400037,China;Department of Otolaryngology,the First Affiliated Hospital of Army Medical University,Chongqing 400038,China;Training Center for Clinical Skills,Army Medical Center,Chongqing 400042,China;Department of Military Preventive Medicine,Army Medical University,Chongqing 400037,China)

机构地区陆军军医大学第一附属医院急诊医学科陆军军医大学生物医学工程系陆军军医大学第一附属医院耳鼻喉科陆军特色医学中心临床技能培训中心陆军军医大学军事预防医学系

出处《医疗卫生装备》 CAS 2024年第9期1-6,共6页 Chinese Medical Equipment Journal

关键词 YOLOv5 注意力机制创伤伤情伤情识别伤情定位 YOLOv5 attention mechanism trauma condition trauma condition identification trauma condition localization

分类号 TP3 [自动化与计算机技术—计算机科学与技术]