Wind Turbine Blade Surfaces Dataset

The proposed dataset aims to provide a number of wind turbine blade images for testing and training purposes. The featured datasets can be used for testing and evaluation of Structure from Motion algorithms for 3D reconstruction, as data for machine learning algorithms for detecting damaged areas on blades or for quantifying blade surface roughness. Images are taken using Canon 5Ds DSLR camera with resolution of 8688 x 5792. The images come with EXIF data, containing additional information about the used capturing settings. The dataset is separated into two parts: - A dataset containing 5 wind turbine blade patches of areas of different surface structure. Two of the patches are of damaged areas, two of the patches are of rough areas, without pronounced surface deformations and one of the patches is a reference area, that does not contain any roughness or damages. The final patch contains the whole of the blade's edge area and is comprised of a mix of severely damaged areas, areas of small roughness and clear areas. The dataset also contains ground truth microscopy data for two of the damaged patches. The ground truth is scaled to absolute scale. - A dataset containing images of a small blade segment. The blade has been sand blasted, to imitate prolonged real world use. The images are taken both outdoor and indoor. The indoor images contain patches specifying areas of interest - one containing a rough patch and one containing damages. The outdoor images are focused on the whole blade and do not contain patches. The dataset also contains ground truth microscopy data for the two patches, which is scaled to absolute scale.

本研究所提出的数据集旨在提供多组风力涡轮机叶片图像，以供测试与训练使用。该特色数据集可用于测试评估用于三维重建的运动恢复结构（Structure from Motion）算法，亦可作为机器学习算法的训练数据，用于检测叶片受损区域或量化叶片表面粗糙度。图像采用佳能5Ds（Canon 5Ds）数码单反相机拍摄，分辨率为8688×5792。所有图像均附带EXIF数据，包含拍摄所用的参数设置等附加信息。
本数据集分为两个子数据集：
- 第一子数据集包含5个覆盖不同表面结构区域的风力涡轮机叶片图像块：其中2个为受损区域图像块，2个为无明显表面变形的粗糙区域图像块，剩余1个为无任何粗糙度或损伤的参考区域图像块。最后一个图像块覆盖叶片完整的边缘区域，由严重受损区域、轻微粗糙区域以及洁净区域混合组成。该子数据集还包含2个受损图像块的显微真值标注（ground truth）数据，且该标注数据已校准至绝对尺度。
- 第二子数据集包含小型叶片段的图像。该叶片经过喷砂处理，以模拟长期实际使用后的状态。图像分别在室内与室外环境拍摄：室内图像包含标注了感兴趣区域的图像块——1个为粗糙区域图像块，1个为受损区域图像块；室外图像则聚焦于完整叶片，未包含预标注的图像块。该子数据集同样包含上述两个图像块的显微真值标注（ground truth）数据，且该标注数据已校准至绝对尺度。

搜集汇总

背景与挑战

背景概述

该数据集是一个风力涡轮机叶片表面图像集合，主要用于测试和训练目的，特别是支持Structure from Motion算法的3D重建以及机器学习算法在损伤检测和表面粗糙度量化中的应用。数据集包含高分辨率图像和EXIF数据，分为两部分：一部分涵盖不同表面结构的叶片补丁（包括损伤、粗糙和参考区域），并提供部分区域的绝对尺度地面真实显微镜数据；另一部分包含模拟长期使用的小叶片段图像，室内外拍摄并带有兴趣区域补丁。数据集由研究机构贡献，适用于风力涡轮机维护和表面分析领域。