.png){kind=icon}
返回
Weakly supervised pavement crack semantic segmentation based on multi-scale object localization and incremental annotation refinement
.png){kind=icon}
PRE
AI
DOI：10.1007/s10489-022-04212-w.png)
.png){kind=icon}
原文求助.png){kind=icon}
分享.png){kind=icon}
收藏
摘要
En 中文
Automatic and accurate pavement crack detection is essential for cost-effective road maintenance. Deep convolutional neural networks (DCNNs) are widely used in recent methods for pavement crack segmentation. Although DCNNs can segment pavement cracks with great accuracy, the requirement for huge pixel-level labels is demanding. In this article, we propose a novel weakly supervised framework for pavement crack segmentation based on multi-scale object localization and incremental annotation refinement. A trained pavement crack classification network is used to produce initial annotations using multi-scale class activation mapping strategy. Then, a new segmentation network (U-2-Net) with triplet attention (TA) module and multiple loss functions is trained using initial annotations. The TA module is developed to emphasize important features and ignore unimportant features, whereas multiple loss functions are employed to assist crack segmentation for a clean and full mask. Moreover, incremental annotation refinement (IAR) is proposed for iteratively optimizing the segmentation network and refining segmentation masks. Comparative experiments on DeepCrack and Crack500 datasets demonstrate that the proposed framework bridges the performance gap between weakly and fully supervised pavement crack segmentation methods, outperforms existing weakly supervised pavement crack segmentation methods, and achieves state-of-the-art performance while reducing human labeling efforts.
Keyword:
Pavement crack segmentation
Class activation map (CAM)
Weakly-supervised learning
Deep learning
期刊
IF:
3.5
论文数:
7.6K
被引数:
1.7W
机构
引用论文
Optimized segmentation with image inpainting for semantic mapping in dynamic scenes.png)
APPLIED INTELLIGENCE
APPLIED INTELLIGENCEIF3.5
PREAI
分享收藏Distant regulatory elements in a Sox10‐βGEO BAC transgene are required for expression of Sox10 in the enteric nervous system and other neural crest‐derived tissuesSox10-βgeo BAC转基因中的远距离调控元件是肠神经系统和其他神经源性组织中 Sox10 表达所必需的
OAAI
OAAI分享收藏DeepCrack: A deep hierarchical feature learning architecture for crack segmentationNEUROCOMPUTING
NEUROCOMPUTINGIF6.5
PREAI
分享收藏U2-Net: Going deeper with nested U-structure for salient object detectionU2-Net: 基于嵌套U结构的显著性目标检测PATTERN RECOGNITIONOAAI
PATTERN RECOGNITIONIF7.6
OAAI分享收藏
查看更多内容