计算机与数字工程2025，Vol.53Issue(4)：1201-1207,7.DOI:10.3969/j.issn.1672-9722.2025.04.047
基于卷积神经网络的多任务轮毂检测方法
Multi-task Hub Detection Based on Convolutional Neural Network
夏辉 1潘丰 1董进华 2张茂彬2
作者信息
- 1. 江南大学轻工过程先进控制教育部重点实验室 无锡 214122
- 2. 无锡信捷电气股份有限公司 无锡 214072
- 折叠
摘要
Abstract
In order to solve the problems of slow detection speed and low accuracy caused by the variability of hub shape,size and position in wheel hub type classification and valve hole location detection,this paper proposes a single-stage method for auto-matically classifying wheel hubs and positioning valve holes,using a lightweight neural network model to extract features,and adopting a single-stage method in parallel with the classification and segmentation branches.The stage strategy is to directly obtain the category and position information of the target through a single detection,and perform wheel image acquisition,preprocessing,classification,and segmentation to locate the valve hole position.The method in this paper has made the following improvements:1)adopting MobileNetV3 network and further lightweighting improvement;2)jointing training of classification and segmentation,and add SVM as supervision;3)introducing encoder/decoder structure,fusing multi-scale features,learning to differentiate catego-ries efficiently.Good results have been obtained on the test set.In the classification of wheel hub models,combining the advantages of CNN and SVM can achieve 100%top-1 accuracy.In segmentation,the top-1 error is 2.63%.关键词
卷积神经网络/MobileNetV3/轮毂分类/气阀孔定位/语义分割Key words
convolutional neural network/MobileNetV3/wheel classification/valve hole positioning/semantic segmenta-tion分类
交通工程引用本文复制引用
夏辉,潘丰,董进华,张茂彬..基于卷积神经网络的多任务轮毂检测方法[J].计算机与数字工程,2025,53(4):1201-1207,7.
