设为首页 联系我们 加入收藏

当前位置: 网站首页 期刊分类目录 2025第1期 正文

一种轻量级混凝土裂缝识别网络

作者:时间:2025-03-05点击数:



全文下载: 20250102.pdf


文章编号: 1672-6987(2025)01-0144-07 DOI: 10.16351/j.1672-6987.2025.01.019

潘澳1 韩笑蕾2* 赵佰亭1*1.安徽理工大学 电气与信息工程学院,安徽 淮南 232001;2.上海商学院 物联网工程系,上海 201400)

摘要: 旨在为混凝土裂缝的类型识别提供一种有效的解决方案,为破损评估提供有力依据。针对现有经典分类网络模型存在参数量大和分类精度不足等问题,提出了一种面向嵌入式平台的轻量化PED-CNN分类算法。算法首先采用空洞卷积扩大模型的感受野,以捕获上下文信息;其次,通过混合池化模块提取裂缝的线性特征;最后,结合部分卷积、普通卷积、深度可分离卷积和ECA注意力机制,设计了一种轻量化PED模块。实验结果表明,所提出的裂缝识别模型以0.76 M的参数,在分类准确率上达到了96.41%,实现了混凝土裂缝的高精度、高效率、智能化的识别与分类。此外,在Jetson Nano嵌入式平台上,该模型峰值帧率高达19.67 Hz,具有较强的研究价值和广泛的应用前景。


关键词: 裂缝识别; 分类算法; 轻量级; 嵌入式平台


中图分类号: TD 26; TP 183; TP 391.41        文献标志码: A


引用格式: 潘澳, 韩笑蕾, 赵佰亭.一种轻量级混凝土裂缝识别网络[J].青岛科技大学学报(自然科学版), 2025, 46(1): -.


PAN Ao, HAN Xiaolei, ZHAO Baiting.A lightweight concrete crack identification network[J].Journal of Qingdao University of Science and Technology(Natural Science Edition), 2025, 46(1): -.

A Lightweight Concrete Crack Identification Network

PAN Ao1 HAN Xiaolei2 ZHAO Baiting11.School of Electrical and Information Engineering, Anhui University of Science and Technology, Huainan 232001, China;2.Institute of IoT Engineering, Shanghai Business School, Shanghai 201400,China)

Abstract: This study aims to provide an effective solution for the identification of concrete crack types, offering robust support for damage assessment. Addressing the issues of large parameter size and insufficient classification accuracy in existing classic classification network models, a lightweight PED-CNN classification algorithm is proposed, tailored for embedded platforms. The algorithm initially employs dilated convolution to expand the model's receptive field, capturing contextual information; subsequently, a mixed pooling module is utilized to extract linear features of cracks. Finally, a lightweight PED module is designed by integrating partial convolution, regular convolution, depthwise separable convolution, and ECA (efficient channel attention) attention mechanism. Experimental results demonstrate that the proposed crack identification model, with a parameter count of only 0.76 M, achieves a classification accuracy of 96.41%, realizing high-precision, efficient, and intelligent identification and classification of concrete cracks. Moreover, on the Jetson Nano embedded platform, the model achieves a peak frame rate of 19.67 Hz, indicating significant research value and broad application prospects.


Key words: crack identification; classification algorithm; lightweight; embedded platform

收稿日期: 2024-09-07

基项目: 国家自然科学基金项目(52174141);苏州市关键核心技术攻关项目(SGC2021069).

作者简介: 潘澳 1999—), 男, 硕士研究生.     * 通信联系人.


Copyright © 2011-2017 青岛科技大学学报 (自然科学版)