基于深度学习的水下混凝土结构表观缺陷智能化识别研究

打开文本图片集

中图分类号：TU317 文献标志码：A DOI： 10.15935/j.cnki.jggcs.202504.0004

AbstractApparent defects in underwater concrete structures are significantly challnged by complex environmental factors including water turbidity，variablelighting conditions，and flow velocity. These interferences lead to diffculties in defect localization and low recognition accuracy during underwater inspections.To address these limitations，this study proposes an inteligent recognition framework based on deep learning.The methodology integrates three key components：generation of a multi-scenario defect database replicating complex underwater environments；application of small-sample expansion and image enhancement algorithms for robust preprocessing； implementation of the YOLOv5 target detection algorithm for multi-category defect identificationand localization.Experimental results demonstrate that the proposed approach achieves a mean average precision（mAP） of 83% and a recognition precision exceeding 83% . This framework effectively mitigates accuracy degradation caused by underwater environmental complexities and limited sample sizes，providing areliable technical solution for automated structural health monitoring of submerged infrastructure.

Keywordsunderwater structures，apparent defects， identification，YOLO algorithm，deep learning

0 引言

在水下检测过程中，因水体对光的吸收和散射、水体浑浊度等因素的影响，会出现成像雾化模糊、比度差、清晰度低等问题，使得检测过程中缺陷定位以及识别难度增大。（剩余8361字）