摘要
为了提高线缆表面缺陷检测正确率,本文提出一种改进Deeplabv3+网络的图像分割方法并将其应用于线缆表面缺陷检测。该方法基于Deeplabv3+网络骨架不变,将空间金字塔结构由4个空洞卷积改为8个空洞卷积并在其后增加1×1的卷积环节;同时在解码融合后用一个并联结构来减少整个网络传输过程的信息丢失。利用改进的算法对线缆表面缺陷图片数据集训练和测试,结果表明改进算法在准确度和平均交并比(MIOU)较原始的Deeplabv3+分析效果更好;相较于边缘分割和阈值分割等算法,改进算法提高了线缆表面缺陷检测的准确率。
In order to improve the accuracy of cable surface defect detection,an improved image segmentation method of Deeplabv3+network is proposed and applied to cable surface defect detection.Based on Deeplabv3+network skeleton unchanged,the spatial pyramid structure is changed from 4 dilated convolutions to 8 dilated convolutions and then 1×1 convolution is added.At the same time,a parallel structure is used to reduce the information loss during the whole network transmission process after decoding and fusion.The improved algorithm is used to train and test the cable surface defect image data set,and the results show that it is better than the original Deeplabv3+analysis in accuracy and mean intersection over union(MIOU).Compared with edge segmentation,threshold segmentation and other algorithms,the proposed algorithm improves the detection accuracy of cable surface defects.
作者
陈亮
杨贤昭
刘惠康
Chen Liang;Yang Xianzhao;Liu Huikang(Engineering Research Center for Metallurgical Automation and Measurement Technology of Ministry of Education,Wuhan University of Science and Technology,Wuhan 430081)
出处
《高技术通讯》
CAS
2021年第9期986-992,共7页
Chinese High Technology Letters
基金
国家重点研发计划(2017YFC0805100)
国家自然科学基金(61703314)资助项目。
