市政技术2025，Vol.43Issue(9)：81-91,11.DOI:10.19922/j.1009-7767.2025.09.081
基于轻量化深度学习的沥青路面裂缝智能识别
Lightweight Learning-driven Crack Identification in Asphalt Pavements
摘要
Abstract
In order to deal with a more efficient and rapid road surface crack detection environment,a lightweight model for identifying road surface cracks was proposed.Firstly,after processing the images of the original asphalt pavement defect dataset,traditional image processing technology and auxiliary classifier generative adversarial net-work(ACGAN)were used to effectively expand the asphalt pavement defect original image dataset,which includes 300 transverse cracks,300 longitudinal cracks,72 oblique cracks,36 mesh cracks and 300 broken road markings.The training set images of each type of pavement defect were expanded to 1 000 pieces.Then,Based on Tensor-flow2.0 deep learning framework,the effectiveness of lightweight intelligent methods for automatic pavement crack identification of three types of models of ShuffleNetV2,MnasNet and GhostNet were established and compared to find the optimal classification model.The results show that after 100 iterative training and ACGAN expansion,the classification accuracy of the lightweight deep learning model was significantly improved than the original dataset;After comparison,ShuffleNetV2 lightweight model achieved the optimal recognition accuracy(ACC=92.8％)on the expansion of defect image dataset.Furthermore,the identification accuracy of transverse crack,longitudinal crack,oblique crack,mesh crack and broken road markings was 89％,90％,87％,98％and 100％,respectively by the model.The research results can provide theoretical and technical support to carry out transportation infrastructure inspection and maintenance.关键词
道路工程/路面裂缝检测/智能识别模型/轻量化卷积Key words
road engineering/pavement crack detection/intelligent recognition model/lightweight convolution分类
交通工程引用本文复制引用
史宏宇,杨晨,李亚非,罗代松,李霖,张永升..基于轻量化深度学习的沥青路面裂缝智能识别[J].市政技术,2025,43(9):81-91,11.基金项目
甘肃省科技重大专项(22ZD6GA010) （22ZD6GA010）
国家自然科学基金项目(52308455) （52308455）
国家重点研发计划项目(2022YFE0137300) （2022YFE0137300）
中央级公益性科研院所基本科研业务费项目(20237504) （20237504）
