报告详情
SF-CrackNet: A Spatial-Frequency Dual-Branch Network for Pavement Crack Segmentation
编号：13
访问权限：仅限参会人
更新：2026-09-11 21:48:19 浏览：12次
张贴报告
报告开始：暂无开始时间（Asia/Shanghai）
报告时间：暂无持续时间
所在会场：[暂无会议] [暂无会议段]
暂无文件
摘要
Pavement crack segmentation remains challenging because cracks are typically thin, low-contrast, spatially discontinuous, and easily confused with shadows, stains, and complex pavement textures. To address these issues, this paper proposes SF-CrackNet, a spatial-frequency dual-branch network that combines a spatial detail encoder (SDE) with a frequency-aware Transformer (FAT). The SDE employs multidirectional convolutional modeling to preserve local crack edges, directional patterns, and elongated structures, while the FAT captures long-range contextual dependencies and frequency-aware representations. Multiscale features from the two branches are aligned at corresponding stages, fused by parameter-free element-wise addition, and progressively decoded to recover pixel-level crack regions. Experiments are conducted on the Crack500 and Crack300 datasets under a unified training-from-scratch protocol. SF-CrackNet achieves an F1-score of 80.26% and an IoU of 67.03% on Crack500, together with a Recall of 83.91% on Crack300. These results indicate that the proposed framework provides competitive segmentation performance while maintaining effective crack-region recall across different pavement scenes.
关键词
Semantic segmentation; pavement crack segmentation; dual-branch network; Transformer; frequency-aware modeling; additive fusion
报告人
Xinyi Liu
Student Soochow University稿件作者
全部评论
重要日期
-
会议日期
11月06日
2026
至11月08日
2026
-
10月15日 2026
初稿截稿日期
主办单位
IEEE Instrumentation and Measurement Society
承办单位
Sichuan University
联系方式
- Icsmd
- ic******@163.com
历届会议
-
2026年11月13日 中国 Chengdu
2026 International Conference on Sensing, Measurement & Data Analytics in the era of Artificial Intelligence (ICSMD) -
2025年11月21日 中国 Guangzhou
2025 International Conference on Sensing, Measurement & Data Analytics in the era of Artificial Intelligence -
2023年11月02日 中国 China
International Conference on Sensing, Measurement & Data Analytics in the era of Artificial Intelligence -
2023年11月02日 中国 Xi'an
2023 International Conference on Sensing, Measurement & Data Analytics in the era of Artificial Intelligence -
2022年11月30日 中国 Harbin
International Conference on Sensing, Measurement & Data Analytics in the era of Artificial Intelligence -
2021年10月21日 中国 Nanjing
2021 International Conference on Sensing, Measurement and Data Analytics in the era of Artificial Intelligence -
2020年10月15日 中国 Xi'an
International Conference on Sensing, Measurement and Data Analytics in the era of Artificial Intelligence
发表评论