改进YOLOv4的实验室设备检测算法  

Improved algorithm of YOLOv4 for laboratory equipment detection

作　　者：李昊霖 徐凌桦[1] 张航 LI Hao-lin;XU Ling-hua;ZHANG Hang(Electrical Engineering College,Guizhou University,Guiyang 550025,China)

机构地区：[1]贵州大学电气工程学院,贵州贵阳550025

出　　处：《计算机工程与设计》2023年第1期133-140,共8页Computer Engineering and Design

基　　金：贵州省省级教学工程基金项目(2019010);国家自然科学基金项目(61861007);贵州省工业攻关基金项目(黔科合支撑[2019]2152)。

摘　　要：针对实验室设备的检测识别问题,提出一种改进YOLOv4算法。针对K-means聚类算法在尺度分布不均匀场景下的局限性,提出一种将数据集标注框按大小划分区间,分别聚类的IK-means++算法;在主干网络中引入通道注意力模块,提出一种阶梯状特征融合网格加强特征融合能力;以计算机实验室为例构建数据集进行训练。实验结果表明,IK-means++算法聚类效果得到有效提升;改进后的YOLOv4算法检测精度更高,模型复杂度更低,速度更快。Aiming at the problem of laboratory equipment detection and recognition,an improved YOLOv4 algorithm was proposed.In view of the limitations of K-means clustering algorithm in the scene of uneven scale distribution,a method called IK-means++algorithm was proposed to divide the data set labeling box into intervals according to size,which performed clustering respectively.A channel attention module was introduced into the structure of the backbone network,and a ladder-shaped feature fusion grid was proposed to strengthen feature fusion capabilities.A computer laboratory was taken as an example to build a data set for training.Experimental results show that the clustering effect of the IK-means++algorithm is effectively improved.The improved YOLOv4 algorithm has higher detection accuracy,lower model complexity and higher speed.

关 键 词：实验室设备 检测识别 先验框聚类 YOLOv4算法 通道注意力 特征融合 复杂度 

分 类 号：TP242.62[自动化与计算机技术—检测技术与自动化装置]