一种改进SegNet网络的路面裂缝分割算法
Pavement crack segmentation algorithm based on improved SegNet
廖宁生 1杨雲翔 1朱秘 1彭波1
作者信息
- 1. 重庆理工大学 两江人工智能学院,重庆 401135
- 折叠
摘要
路面裂缝是威胁公路安全运行的常见潜在隐患,经典路面裂缝分割算法存在不同程度的裂缝断裂、薄细裂缝边缘识别不佳等问题.针对上述问题,提出一种Crack SegFormer路面裂缝分割算法,主要由基于裂缝定位注意力的编码器、多层特征金字塔以及基于裂缝锐化注意力的解码器三部分组成.利用 Crack500、Crack200、DeepCrack、CFD 4 个公开数据,对CrackSegFormer模型分割裂缝的有效性进行了验证,结果显示所提出的CrackSegFormer模型能够抑制非裂缝特征、保留细微和末梢裂缝特征.相对于经典SegNet网络,所提出模型的准确度、召回率和 F1-score三类评价指标分别提升了1.14%,3.61%和 4.26%.
Abstract
Pavement cracks are common potential hazards threatening the safety on highways.Classical pavement crack detection algorithms suffer such problems as different degrees of crack breaks and poor recognition of the thin and fine edges of cracks.This paper proposes a Crack SegFormer model based on an improved SegNet network,mainly comprising three parts:an encoder based on crack localization attention,a multilayer feature pyramid,and a decoder based on crack sharpening attention.The effectiveness of the Crack SegFormer model in segmenting cracks is verified based on three publicly available data from Crack500,Crack200,DeepCrack and CFD.Our results show the proposed Crack SegFormer model is able to suppress non-cracking features and retain fine and end-cracking features.Compared with the classical SegNet network,it improves the accuracy by 1.14%,the recall rate by 3.61%and F1-score by 4.26%.
关键词
路面裂缝分割/改进SegNet网络/注意力机制/多层特征金字塔Key words
pavement crack segmentation/improved SegNet network/attention mechanism/multilayer feature pyramid引用本文复制引用
出版年
2024
国家科技期刊平台
NSTL
万方数据