[1]庄永强.融合注意力机制与轻量化网络的桥梁裂缝分类[J].福建工程学院学报,2023,21(04):327-331.[doi:10.3969/j.issn.1672-4348.2023.04.004]
 ZHUANG Yongqiang.Bridge crack classification based onfusion of attention mechanism and lightweight network[J].Journal of FuJian University of Technology,2023,21(04):327-331.[doi:10.3969/j.issn.1672-4348.2023.04.004]
点击复制

融合注意力机制与轻量化网络的桥梁裂缝分类()
分享到:

《福建工程学院学报》[ISSN:2097-3853/CN:35-1351/Z]

卷:
第21卷
期数:
2023年04期
页码:
327-331
栏目:
出版日期:
2023-08-25

文章信息/Info

Title:
Bridge crack classification based onfusion of attention mechanism and lightweight network
作者:
庄永强
福州市交通建设集团有限公司
Author(s):
ZHUANG Yongqiang
Fuzhou Communications Construction Group Co., Ltd.,
关键词:
桥梁裂缝检测轻量化卷积神经网络注意力机制CBAM
Keywords:
bridge crack detection lightweight convolution neural network attention mechanism CBAM
分类号:
TU997
DOI:
10.3969/j.issn.1672-4348.2023.04.004
文献标志码:
A
摘要:
提出一种基于注意力机制融合轻量化网络的桥梁裂缝图像分类方法。以轻量化卷积神经网络为理论基础分类识别桥梁裂缝图像,并在轻量化网络中加入注意力机制以解决网络无法自主关注所感兴趣区域的问题。根据桥梁图像中裂缝所占比例较小且边缘突出的特点,选用适合于识别桥梁裂缝的注意力机制——CBAM(convolutional block attention module),并将其嵌入轻量化卷积神经网络Ef?ficientNetv2中,建立CBAM?EfficientNetv2模型。实验结果表明:CBAM?EfficientNetv2模型与VGG16、ResNet34等常用深度学习模型对比,可获得最优的桥梁裂缝图像分类效果,其分类识别准确率达到95.64%。
Abstract:
A bridge crack image classification method based on attention mechanism and lightweight network was proposed. This algorithm classifies and recognizes the images of bridge cracks based on the lightweight convolutional neural network, and adds an attention mechanism into the lightweight network to solve the problem that the network cannot focus on the area of interest independently. According to the characteristics of small proportion of cracks and prominent edges in bridge images, the attention mechanism CBAM (convolutional block attention module) suitable for identifying bridge cracks was selected and embedded into the lightweight convolution neural network EfficientNetv2 to establish the CBAMEfficientNetv2 model. Experimental results show that compared with the common deep learning models such as VGG16 and ResNet34, the proposed model can obtain the optimal classification effect of bridge crack images, with the classification recognition accuracy being 95.64%.
更新日期/Last Update: 2023-08-25