基于改进布谷鸟搜索优化深度卷积网络的隧道衬砌裂缝检测算法研究
Research on Crack Detection Algorithm of Tunnel Lining Based on Improved Cuckoo Search Optimization Deep Convolutional Network
韩青松 1李生勇 2徐红梅2
作者信息
- 1. 内蒙古交通职业技术学院,内蒙古赤峰 024000
- 2. 河套学院,内蒙古巴彦淖尔 015000
- 折叠
摘要
针对隧道衬砌裂缝检测算法准确度低的问题,提出改进布谷鸟搜索优化深度卷积网络的隧道衬砌裂缝图像检测算法.首先,基于EfficientNet卷积块堆叠网络和使用深度可分离卷积移动倒置残块(Mobile Inverted Residual Block,MBConv),进行多尺度高效提取裂缝图像语义特征;同时引入改进的卷积块注意力模块(Convolutional Block Attention Module,CBAM)增强关键特征的影响;为避免区域边缘细节特征丢失,运用边缘强化模块(Boundary Enhancement Module,BEM)来调整边缘位置特征细节权重;最后,使用轮盘赌改进自适应布谷鸟搜索优化分割阈值θ,进而得到衬砌裂缝图像检测算法.消融实验结果表明,各种优化改进模块可有效提高算法模型效果,在有干扰和无干扰条件下,准确率分别达到95.74％和97.26％;对比其他算法,该算法模型的裂缝检测准确率达94.91％,均优于Mask R-CNN和DeepLabv3等算法.
Abstract
In response to the issue of low accuracy in tunnel lining crack detection algorithms,this paper proposes an improved cuckoo search optimization deep convolutional network algorithm for detecting tunnel lining crack images.First,based on the EfficientNet convolutional block stacking network and using the Mobile Inverted Residual Block(MBConv)with depthwise separable convolution,the algorithm efficiently extracts semantic features of crack images at multiple scales.Additionally,an improved Convolutional Block Attention Module(CBAM)is introduced to enhance the impact of key features.To prevent the loss of edge detail features,a Boundary Enhancement Module(BEM)is utilized to adjust the weight of boundary position feature details.Finally,a roulette wheel improved adaptive Cuckoo search is used to optimize segmentation threshold θ,resulting in a lining crack image detection algorithm.Ablation experiment results indicate that various optimized improvement modules can effectively enhance the performance of the algorithm model,achieving accuracy rates of 95.74％and 97.26％under interference and non-interference conditions,respectively.Compared to other algorithms,the accuracy rate of crack detection for this algorithm model reaches 94.91％,which is superior to algorithms such as Mask R-CNN and DeepLabv3.
关键词
衬砌裂缝/隧道结构/检测算法/图像特征Key words
lining cracks/tunnel structure/detection algorithm/image features引用本文复制引用
基金项目
内蒙古自治区教育厅自然科学项目(NJZY21190)
内蒙古自治区教育厅研究专项项目(STAQZX202320)
乌梁素海流域山水林田湖草生态保护修复试点工程支持计划项目(2019HYYSZX)
出版年
2024
NETL
NSTL
万方数据