[1]卞长庚,郝万君,马文琪.基于 Xception 和 SA 的 YOLOv5 建筑裂缝检测方法[J].计算机技术与发展,2023,33(08):159-164.[doi:10. 3969 / j. issn. 1673-629X. 2023. 08. 023]
　BIAN Chang-geng,HAO Wan-jun,MA Wen-qi.YOLOv5 Building Crack Detection Method Using Xception and SA[J].,2023,33(08):159-164.[doi:10. 3969 / j. issn. 1673-629X. 2023. 08. 023]

点击复制

基于 Xception 和 SA 的 YOLOv5 建筑裂缝检测方法()

分享到：

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

卷:

33

期数:

2023年08期

页码:

159-164

栏目:

人工智能

出版日期:

2023-08-10

文章信息/Info

Title:

YOLOv5 Building Crack Detection Method Using Xception and SA

文章编号:

1673-629X(2023)08-0159-06

作者:

卞长庚; 郝万君; 马文琪

苏州科技大学 电子与信息工程学院,江苏 苏州 215009

Author(s):

BIAN Chang-geng; HAO Wan-jun; MA Wen-qi

School of Electronic and Information Engineering,Suzhou University of Science and Technology,Suzhou 215009,China

关键词:

裂缝检测; Xception; 空洞空间金字塔池化; Shuffle 注意力

Keywords:

crack detection; Xception; atrous spatial pyramid pooling; Shuffle attention

分类号:

TP391

DOI:

10. 3969 / j. issn. 1673-629X. 2023. 08. 023

摘要:

裂缝检测对于建筑的维修和加固、延长其使用寿命具有重要意义。 针对建筑裂缝种类多和尺寸小造成裂缝检测精度低、速度慢的问题,提出了一种改进的 YOLOv5 裂缝检测算法,在提高检测裂缝精度的同时也提升了检测裂缝的速度。首先,引入轻量级网络 Xception 对主干网络轻量化,减少主干网络参数量以

提升检测裂缝的速度;其次,使用空洞空间金字塔池化 ASPP( Atrous Spatial Pyramid Pooling) 模块替换 SPP( Spatial Pyramid Pooling) 模块,扩大感受野范

围,加强主干网络提取裂缝特征的能力,避免因对主干网络轻量化而造成检测裂缝的精度降低;最后,添加 SA( Shuffle Attention) 注意力机制,进一步加强网络

提取裂缝特征的能力,提高裂缝检测的精度。 通过在自制数据集上进行的实验表明,改进的算法 mAP比原算法提高了 1. 6% ,速度为 50. 8 f / s,比原算法提高

了 2. 7 f / s,满足建筑裂缝检测的精度和实时性要求,同时将改进算法与 Faster R-CNN、Mobile-SSD、YOLOv4-tiny 等算法进行对比,证明了该算法的优越性,

更适合部署到硬件平台上。

Abstract:

Crack detection is of great significance for the maintenance and reinforcement of buildings and for extending their service life.Aiming at the problems of?

low accuracy and slow speed of crack detection caused by the variety and small size of building cracks,an improved YOLOv5 crack detection algorithm is proposed,which improves the accuracy of crack detection and the speed of crack detection.Firstly,the lightweight network Xception is introduced to lighten the backbone network and reduce the number of backbone network parameters,thus improving the speed of crack detection. Secondly,the SPP ( Spatial Pyramid Pooling) module is replaced with the ASPP( Atrous Spatial Pyramid Pooling) module to expand the receptive field,strengthen the ability of backbone network to extract fracturefeatures,and avoid reducing the accuracy of crack detection due to the lightweight of backbone network. Finally,SA ( Shuffle Attention)attention mechanism is added to further enhance the ability of network to extract crack features and improve the accuracy of crackdetection. Through experiments on self - made datasets, the mAP of improved algorithm is 1. 6% higher than that of the originalalgorithm,with a speed of 50. 8 f / s,　which is 2. 7 f / s higher than that of the original algorithm,and meets the accuracy and real-time requirements of building crack detection. At the same time, the improved algorithm is compared with Faster R - CNN, Mobile - SSD,YOLOv4-tiny and other algorithms,which proves its superiority and is more suitable for deployment on hardware platforms.

常用功能