[1]罗 杨,万黎明,李 理,等.基于改进 U-Net 网络的半监督裂缝分割方法[J].计算机技术与发展,2022,32(12):179-184.[doi:10. 3969 / j. issn. 1673-629X. 2022. 12. 027]
　LUO Yang,WAN Li-ming,LI Li,et al.A Semi-supervised Crack Segmentation Method Based on Improved U-Net Network[J].,2022,32(12):179-184.[doi:10. 3969 / j. issn. 1673-629X. 2022. 12. 027]
点击复制

基于改进 U-Net 网络的半监督裂缝分割方法()
分享到：

《计算机技术与发展》[ISSN:1006-6977/CN:61-1281/TN]

卷:
32
期数:
2022年12期
页码:
179-184
栏目:
人工智能
出版日期:
2022-12-10

文章信息/Info

Title:
A Semi-supervised Crack Segmentation Method Based on Improved U-Net Network
文章编号:
1673-629X(2022)12-0179-06
作者:
罗 杨1 ; 万黎明1 ; 李 理1 ; 刘知贵1; 2
1. 西南科技大学 计算机科学与技术学院,四川 绵阳 621000;
2. 西南科技大学 信息工程学院,四川 绵阳 621000
Author(s):
LUO Yang1 ; WAN Li-ming1 ; LI Li1 ; LIU Zhi-gui1; 2
1. School of Computer Science and Technology,Southwest University of Science and Technology,Mianyang 621000,China;
2. School of Information Engineering,Southwest University of Science and Technology,Mianyang 621000,China
关键词:
裂缝分割; 半监督; 注意力机制; 深度学习; U-Net
Keywords:
crack segmentation; semi-supervised; attention mechanism; deep learning; U-Net
分类号:
TP391
DOI:
10. 3969 / j. issn. 1673-629X. 2022. 12. 027
摘要:
裂缝反映了结构的受力状态,是结构健康检测的重要关注对象之一。 基于数字图像利用深度学习方法进行结构表面裂缝自动识别具有速度快、精度高等优势,不过深度学习方法严重依赖像素级标注信息,为此,提出一种基于半监督学习的改进 U-Net 方法。 使用特征提取能力更佳的残差网络作为主干特征提取网络代替 U-Net 中由卷积层和池化层进行简单堆叠而成的下采样部分;在主干网络中插入池化窗口长且窄的条带池化注意力辅助下采样进行特征的细化,增强特征提取能力;针对裂缝图像中裂缝区域的亮度普遍暗于背景区域的情况,网络中的池化操作均采用平均池化使网络能更好地处理裂缝图像;利用半监督学习,在训练时同时训练两个网络并利用其分割结果相互监督从而使深度学习分割方法降低对标签数据的依赖度。 改进的 U-Net 分割方法在自建裂缝数据集上进行了对比实验,结果表明,相较于原始 U-Net网络,改进方法具有更高的分割精度,训练时可使用更少的标签数据。
Abstract:
Cracks reflect the stress state of a structure and are one of the important objects of concern for structural health inspection. Theautomatic recognition of cracks based on digital images using deep learning methods has the advantages of high speed and accuracy,butthe deep learning methods rely heavily on pixel-level annotation,so an improved U-Net method based on semi-supervised learning isproposed. A residual network with better feature extraction capability is used as the backbone feature extraction network instead of thedown-sampling part of U-Net,which consists of simple stacking of convolutional and pooling layers. A strip pooling attention with longand narrow pooling window is inserted into the backbone to assist down-sampling for feature refinement and enhance feature extractioncapability. For the situation that the brightness of crack region in a crack image is generally darker than that of the background region,allof the pooling operation in the network is average pooling so that the network can better handle the crack images. Using semi-supervisedlearning,two networks are trained at the same time and their segmentation results are used to supervise each other so that the deep learningsegmentation method can reduce the dependence of labeled data. The improved 　U -Net segmentation method is tested on a self - builtcrack dataset,the experimental results show that the proposed method has higher accuracy than the original 　U - Net network and fewerlabel data needed for training.

相似文献/References:

[1]刘炯 徐同阁.基于NetFlow的应用协议半监督识别算法[J].计算机技术与发展,2010,(07):9.
　LIU Jiong,XU Tong-ge.A Semi-Supervised Clustering Algorithm for Application Protocol Recognition Based on NetFlow[J].,2010,(12):9.
[2]李亚娥,汪西莉.一种自适应的半监督图像分类算法[J].计算机技术与发展,2013,(02):112.
　LI Ya-e,WANG Xi-li.An Adaptive Semi-supervised Image Classification Algorithm[J].,2013,(12):112.
[3]宛艳萍,闫思聪,于海阔,等.结合 SS-GAN 和 BERT 的文本分类模型[J].计算机技术与发展,2023,33(02):187.[doi:10. 3969 / j. issn. 1673-629X. 2023. 02. 028]
　WAN Yan-ping,YAN Si-cong,YU Hai-kuo,et al.A Text Classification Model Based on Semi-supervised Generative Adversarial Networks and BERT[J].,2023,33(12):187.[doi:10. 3969 / j. issn. 1673-629X. 2023. 02. 028]

更新日期/Last Update: 2022-12-10