基于改进MobileNetV2的钢筋混凝土桥梁病害分类识别

打开文本图片集

WU Lipeng1，²，YAN Xinyu1.2， JIN Zhecheng1，2

1. School of Civil Engineering， Shijiazhuang Tiedao University， Shijiazhuang O50043，China; 2. Key Laboratory of Roads and Railway Engineering Safety Control （Shijiazhuang Tiedao University），Ministry of Education， Shijiazhuang O5O043， China）

Abstract：Inordertotimelyandefectivelydetectbridgediseases，overcometheinherentdefectsofthecurrntmanual visual inspectionmethod indisease investigation，whichis time-consuming，laborious，easytocause potential safety hazards，and easilyaffectedbysubjectivefactors，soastoprovidepreconditionsfortargetedbridgemanagement，anintellgentmethodwhich issuitableformobilediseaseimagerecognitionisproposed.Whilekeepingthewidthof theoriginalMobileNetV2network model unchanged，byreducing thenumberofbotleneck modulesandembeddingthe SqueezeandExcitationchannelatention mechanism inthebotleneck module，thecorrelationbetweenchannels isestablished，sothatthenetworkcanleamandutlize therelationshipbetweenfeaturechannelsmoreefectivelyTheresultsshowthattheaccuracyof imagerecognitionisimproved， aftertwo improvements inthenetwork modelofreducing botteneck moduleandembedding channelatentionmechanism，the recognition accuracy of 98.78% is achieved，which is 2. 85% higher than before the improvement and the computational cost is also reduced by 23.02M ；The improved network model is introduced into the GUI interface of Matlab，and the code is

visualizedasabridgediseaseidentificationsystemthatcanbeusedindependently，whichprovidesconveniencefortheefficient identification of four bridge diseases： efflorescence， steel corrosion， cracking and holes.

Key words：bridge；disease identification；MobileNetV2；attention mechanism；GUI

0 引言

桥梁是重要的交通基础设施，与国家建设、社会进步和人民福祉紧密相关。（剩余8821字）