基于多维注意力模块的轻量化混凝土裂缝检测方法  

Lightweight Concrete Crack Detection Method Based on Multi-Dimensional Attention Module

作　　者：许华杰[1,2] 郑力文 张品 秦远卓 XU Huajie;ZHENG Liwen;ZHANG Pin;QIN Yuanzhuo(College of Computer and Electronic Information,Guangxi University,Nanning 530004,Guangxi,China;Guangxi Key Laboratory of Multimedia Communications and Network Technology,Nanning 530004,Guangxi,China;Beibu Gulf Port Fangchenggang Terminal Co.,Ltd.,Fangchenggang 538001,Guangxi,China;College of Civil Engineering and Architecture,Guangxi University,Nanning 530004,Guangxi,China)

机构地区：[1]广西大学计算机与电子信息学院,广西南宁530004 [2]广西多媒体通信与网络技术重点实验室,广西南宁530004 [3]北部湾港防城港码头有限公司,广西防城港538001 [4]广西大学土木建筑工程学院,广西南宁530004

出　　处：《计算机工程》2025年第5期351-360,共10页Computer Engineering

基　　金：广西自然科学基金(2024JJA170106);广西重点研发计划项目(桂科AD25069071);国家自然科学基金(52169021)。

摘　　要：为解决当前混凝土裂缝检测模型庞大难以部署到移动端设备且裂缝检测不准及漏检问题,提出一种基于多维注意力模块的轻量化混凝土裂缝检测方法。该方法针对当前主流的裂缝检测模型庞大的问题,采用深度可分离卷积对YOLOv5s中的CBS特征提取模块进行重构,得到轻量化CBS(LCBS)特征提取模块,以减少网络的参数量及计算量;针对裂缝检测不准的问题,提出一种多尺度特征(MSF)提取模块用于替换YOLOv5s第1层的卷积层,以增强网络对不同尺寸裂缝特征的提取能力;针对裂缝漏检问题,提出融合空间及通道信息的多维注意力(MDA)模块,以增强裂缝特征提取能力和减少裂缝漏检。实验结果表明,所提方法比YOLOv5s参数量减少了35.2%,计算量减少了50.9%,模型规模减小了32.8%,且平均精度均值(mAP@0.5)提高了4.2百分点,与目前主流的同类目标检测方法相比,具有较低的参数量和较高的检测精度。Most current concrete crack detection models are too large to be deployed in mobile devices,and the crack detection is inaccurate,with cracks being missed;to solve these problems,a lightweight concrete crack detection method based on a multi-dimensional attention module is proposed.In this method,because most current mainstream crack detection models are large,depth-separable convolution is used to reconstruct the Conv-BN-SiLU(CBS)feature extraction module in YOLOv5s to obtain a Lightweight(LCBS)module.To solve the problem of inaccurate crack detection,a Multi-Scale Feature(MSF)module is proposed to replace the convolution layer of the first layer of YOLOv5s to enhance the ability of the network to extract the features of cracks of different sizes.To address the problem of missed cracks,a Multi-Dimensional Attention(MDA)module,which fuses spatial and channel information,is proposed to enhance the ability of crack feature extraction and reduce the number of missed cracks.Experiments show that,compared with YOLOv5s,the proposed method reduces the number of parameters by 35.2%,computation amount by 50.9%,and model size by 32.8%and increases the average accuracy(mAP@0.5)by 4.2 percentage points.Compared with other mainstream target detection methods of the same type,the proposed method has fewer parameters and higher detection accuracy.

关 键 词：裂缝检测 注意力模块 轻量化 YOLOv5s模型 目标检测 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]