复杂背景下的路面裂缝检测的关键技术  

Key technology of pavement crack detection under complex background

作　　者：杨泽 孙静宇[1] YANG Ze;SUN Jing-yu(College of Software,Taiyuan University of Technolgy,Jinzhong 030600,China)

机构地区：[1]太原理工大学软件学院,山西晋中030600

出　　处：《计算机工程与设计》2023年第5期1519-1527,共9页Computer Engineering and Design

基　　金：山西省“1331工程”基金项目(SC9100026)。

摘　　要：针对目前路面裂缝检测方法在复杂环境下识别率较低、鲁棒性较差的问题,提出一种改进网络CBAM-Res-GhostNet对路面裂缝实现有效分类。在卷积神经网络中引入Ghost模块和改进残差模块,加入卷积注意力,避免梯度消失和过拟合现象,实现对路面裂缝的准确判断;在此基础上,提出一种改进网络Self-Attention-UNet对路面裂缝区域进行高精度分割,引入自注意力机制增强模型裂缝特征提取能力,提高分割精度。在EdmCrack600数据集上,所提分类算法准确度达到99.13%,分割算法的精准率和F1值分别为86.85%和86.6%,相较原始方法具有更好的分类分割效果。Aiming at the problems of low recognition rate and poor robustness of current pavement crack detection methods in complex environments,an improved network CBAM-Res-GhostNet was proposed to effectively classify pavement cracks.Ghost module and improved residual module were introduced into the convolutional neural network,and convolution attention was added to avoid gradient disappearance and over-fitting phenomenon,so as to achieve accurate judgment of pavement cracks.On this basis,an improved network Self-Attention-UNet was proposed to segment the pavement crack area with high precision,and the self-attention mechanism was introduced to enhance the crack feature extraction ability of the model and improve the segmentation accuracy.On the EdmCrack600 data set,the accuracy of the proposed classification algorithm reaches 99.13%,and the accuracy and F1 value of the segmentation algorithm are 86.85%and 86.6%,respectively.Compared with the original method,it has better classification and segmentation effects.

关 键 词：计算机视觉 裂缝检测 深度学习 图像处理 U型卷积神经网络 注意力 残差模块 

分 类 号：TP399[自动化与计算机技术—计算机应用技术]