基于改进YOLOv4-Tiny算法的机械零件识别  

Recognition of mechanical parts based on improved YOLOv4-Tiny algorithm

作　　者：杨一帆 靳伍银[1] 薛文亮 王浩浩 YANG Yifan;JIN Wuyin;XUE Wenliang;WANG Haohao(School of Mechanical and Electrical Engineering,Lanzhou University of Technology,Lanzhou 730050)

机构地区：[1]兰州理工大学机电工程学院,甘肃兰州730050

出　　处：《机械设计》2024年第7期61-65,共5页Journal of Machine Design

基　　金：甘肃省高等学校产业支撑计划项目(2022CYZC-24);基于癫痫脑电信号的分类与识别研究(21JR11RM050)。

摘　　要：为实现机械零件的精准快速识别,文中提出了一种基于改进的YOLOv4-Tiny算法的机械零件识别方法。该方法融合了注意力机制和K-means++聚类算法,采用CSPDarknet53-Tiny网络作为主干网络,并将卷积注意力机制模块(Convolution Block Attention Module, CBAM;Global Attention Mechanism, GAM)加在YOLOv4-Tiny主干网络与特征金字塔的连接处及其上采样处,在不影响主干网络的条件下,对每个通道的特征信息重新压缩并提取,过滤掉冗余特征信息,保留重要特征信息,并重新分配权重;再用K-means++聚类算法得到一组与机械零件图像数据集相匹配的先验框参数。试验结果表明,与传统的YOLOv4-Tiny算法相比,改进后的YOLOv4-Tiny算法在保证实时性的前提下,平均召回率和平均准确率分别达到99.43%和99.41%,可以准确检测并定位机械零件图像的位置。In this article,in order to ensure accurate and rapid recognition of mechanical parts,a method based on the YOLOv4-Tiny algorithm is proposed.It combines the attention mechanism and the K-means++clustering algorithm,with the CSPDarknet53-Tiny network as the mainstream;Convolution Block Attention Module(CBAM)and Global Attention Mechanism(GAM)are added to the connection between the CSPDarknet53-Tiny main network and the feature pyramid as well as the upper sampling point.When the main network is not affected,the feature information of each channel is compressed and extracted once again.As a result,the redundant feature information is filtered out,the key feature information is retained,and the weight is reallocated.Then,a set of prior frame parameters matching the data set of mechanical parts is obtained with the help of the K-means++clustering algorithm.The experimental results show that the compared with the traditional YOLOv4-Tiny algorithm,the improved YOLOv4-Tiny algorithm has desirable real-time performance,with the average recall rate and the average precision rate of 99.43%and 99.41%respectively.This algorithm is helpful to detect and recognize the mechanical parts in an accurate manner.

关 键 词：YOLOv4-Tiny算法 机械零件识别 CBAM GAM K-means++聚类算法 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程] TP391.41[自动化与计算机技术—控制科学与工程] TS958.03[轻工技术与工程]