HRCDS: A Benchmark Dataset for High-Resolution Concrete Damage Segmentation

Concrete structures deteriorate due to environmental stressors, aging, and mechanical loads, resulting in cracks, spalling, and corrosion. Early damage detection is essential for ensuring structural integrity and safety. While deep learning has improved automated detection, its effectiveness is constrained by the lack of high-quality datasets with diverse damage types and precise annotations. Existing datasets often suffer from low resolution, limited variability, and inadequate labeling, hindering model generalization. To overcome these challenges, a high-resolution concrete damage segmentation dataset (HRCDS) has been introduced for deep learning applications in structural health monitoring. HRCDS offers pixel-wise annotations for various damage types, including cracks, exposed rebar, corrosion strain, and surface spalling, captured under different lighting conditions and textures. The public release of HRCDS aims to drive advancements in AI-powered structural assessment, fostering innovation in civil engineering, deep learning, and digital twin technologies.

混凝土结构会因环境应力、老化及机械荷载作用发生劣化，进而产生裂缝、表层剥落与钢筋锈蚀等病害。早期病害检测对于保障结构完整性与安全性至关重要。尽管深度学习推动了自动化检测技术的发展，但其实际效能受限于高质量数据集的匮乏——此类数据集需涵盖多样病害类型并具备精准标注。现有数据集普遍存在分辨率偏低、样本变异性不足以及标注不完善等问题，制约了模型的泛化能力。为应对上述挑战，本研究提出了一款面向结构健康监测深度学习应用的高分辨率混凝土损伤分割数据集（HRCDS）。该数据集为裂缝、钢筋外露、锈蚀应变以及表层剥落等多种病害类型提供像素级标注，采集场景覆盖不同光照条件与纹理特征。此次HRCDS的公开发布旨在推动人工智能驱动的结构评估技术进步，助力土木工程、深度学习以及数字孪生等领域的技术创新。

搜集汇总

背景与挑战

背景概述

MDMCS是一个用于混凝土结构多损伤监测的基准数据集，包含1,200张图像，涵盖四种损伤类型（裂缝、剥落、腐蚀和裸露钢筋），并具有多样化的光照条件和材料纹理。该数据集旨在促进基于人工智能的结构监测和机器人辅助自动检测技术的发展。