吉首大学学报(自然科学版) ›› 2023, Vol. 44 ›› Issue (1): 24-29.DOI: 10.13438/j.cnki.jdzk.2023.01.004

• 计算机与电工技术 • 上一篇    下一篇

改进的YOLOv4肺结节检测算法

林锟煌,李建锋,汪洋,刘志杰,刘哲宇   

  1. (吉首大学通信与电子工程学院,湖南 吉首 416000)
  • 出版日期:2023-01-25 发布日期:2023-04-10
  • 通讯作者: 李建锋(1979—),男,湖南张家界人,吉首大学通信与电子工程学院教授,博士,硕士生导师,主要从事医学图像处理研究.
  • 作者简介:林锟煌(1998—),男,福建泉州人,硕士研究生,主要从事医学图像处理研究
  • 基金资助:
    国家自然科学基金资助项目(61962023);吉首大学研究生科研项目(JDY21078)

Improved YOLOv4 Lung Nodule Detection Algorithm

LIN Kunhuang,LI Jianfeng,WANG Yang,LIU Zhijie,LIU Zheyu   

  1. (College of Information Science and Engineering,Jishou University,Jishou 416000,Hunan China)
  • Online:2023-01-25 Published:2023-04-10

摘要:针对目标检测YOLOv4算法在肺结节检测中存在的小目标漏检和肺结节位置失真等问题,设计了一种改进的YOLOv4肺结节检测算法.在原始YOLOv4网络的基础上,将特征融合网络的上采样过程替换为双线性插值法,并采用张量堆叠的方法使顶层的语义信息与底层的位置信息形成更高通道的特征张量.实验结果表明,与原始的YOLOv4算法相比,改进的YOLOv4算法在公开数据集LUAN16上的平均精确度与预测速度分别提高了4.54%和28.1%,可视化结节位置表达更精准.

关键词: 肺结节, 目标检测算法, YOLOv4, 特征融合

Abstract: An improved YOLOv4 lung nodule detection algorithm is designed to solve the problems of small target missing detection and lung nodule position distortion in the target detection YOLOv4 algorithm.On the basis of the original YOLOv4 network,the up sampling process of the feature fusion network is replaced by the bilinear interpolation method,and the tensor stacking method is used to make the semantic information of the top layer and the location information of the bottom layer to form a higher channel feature tensor.The experimental results show that,compared with the original YOLOv4 algorithm,the average accuracy and prediction speed of the improved YOLOv4 algorithm on the public dataset LUAN16 are improved by 4.54% and 28.1% respectively,and the visualization results have more accurate position expression.

Key words: lung nodules, target detection algorithm, YOLOv4, feature fusion

公众号 电子书橱 超星期刊 手机浏览 在线QQ