表面缺陷检测数据集综述及GitHub开源项目推荐

2&＃xff09;太阳能板&＃xff1a;elpv-dataset

地址&＃xff1a;https://github.com/zae-bayern/elpv-dataset

3&＃xff09;金属表面&＃xff1a;KolektorSDD

The dataset is constructed from images of defected electrical commutators that were provided and annotated by Kolektor Group d.o.o… Specifically, microscopic fractions or cracks were observed on the surface of the plastic embedding in electrical commutators. The surface area of each commutator was captured in eight non-overlapping images. The images were captured in a controlled environment.

offical link: https://www.vicos.si/Downloads/KolektorSDD

百度网盘链接&＃xff1a;https://pan.baidu.com/share/init?surl&＃61;HSzHC1ltHvt1hSJh_IY4Jg

提取码&＃xff1a;1zlb

The dataset consists of:

• 50 physical items (defected electrical commutators)
• 8 surfaces per item
• Altogether 399 images:
  – 52 images of visible defect
  – 347 images without any defect
• Original images of sizes:
  – width: 500 px
  – height: from 1240 to 1270 px
• For training and evaluation images should be resized to 512 x 1408 px

For each item the defect is only visible in at least one image, while two items have defects on two images, which means there were 52 images where the defects are visible. The remaining 347 images serve as negative examples with non-defective surfaces.

4&＃xff09;PCB板检测&＃xff1a;DeepPCB

地址&＃xff1a;https://github.com/Charmve/Surface-Defect-Detection/tree/master/DeepPCB

5&＃xff09;AITEX数据集&＃xff08;面料缺陷&＃xff09;

数据集下载链接&＃xff1a;

https://pan.baidu.com/s/1cfC4Ll5QlnwN5RTuSZ6b7w

提取码&＃xff1a;b9uy

该数据库由七个不同织物结构的245张4096 x 256像素图像组成。数据库中有140个无缺陷图像&＃xff0c;每种类型的织物20个&＃xff0c;除此之外&＃xff0c;有105幅纺织行业中常见的不同类型的织物缺陷&＃xff08;12种缺陷&＃xff09;图像。图像的大尺寸允许用户使用不同的窗口尺寸&＃xff0c;从而增加了样本数量。Internet上的数据库还包含所有具有缺陷的图像的分割mask&＃xff0c;使得白色像素表示缺陷区域&＃xff0c;其余像素为黑色。

6&＃xff09;天池布匹缺陷数据&＃xff08;竞赛&＃xff09;

数据下载链接&＃xff1a;

https://pan.baidu.com/s/1LMbujxvr5iB3SwjFGYHspA

提取码&＃xff1a;gat2

在布匹的实际生产过程中&＃xff0c;由于各方面因素的影响&＃xff0c;会产生污渍、破洞、毛粒等瑕疵&＃xff0c;为保证产品质量&＃xff0c;需要对布匹进行瑕疵检测。布匹疵点检验是纺织行业生产和质量管理的重要环节&＃xff0c;目前人工检测易受主观因素影响&＃xff0c;缺乏一致性&＃xff1b;并且检测人员在强光下长时间工作对视力影响极大。由于布匹疵点种类繁多、形态变化多样、观察识别难道大&＃xff0c;导致布匹疵点智能检测是困扰行业多年的技术瓶颈。本数据涵盖了纺织业中布匹的各类重要瑕疵&＃xff0c;每张图片含一个或多种瑕疵。数据包括包括素色布和花色布两类&＃xff0c;其中&＃xff0c;素色布数据约8000张&＃xff0c;用于初赛&＃xff1b;花色布数据约12000张&＃xff0c;用于复赛。

7&＃xff09;天池铝型材表面瑕疵数据集&＃xff08;竞赛&＃xff09;

数据集下载链接&＃xff1a;

https://tianchi.aliyun.com/competition/entrance/231682/information

数据介绍&＃xff1a;在铝型材的实际生产过程中&＃xff0c;由于各方面因素的影响&＃xff0c;铝型材表面会产生裂纹、起皮、划伤等瑕疵&＃xff0c;这些瑕疵会严重影响铝型材的质量。为保证产品质量&＃xff0c;需要人工进行肉眼目测。然而&＃xff0c;铝型材的表面自身会含有纹路&＃xff0c;与瑕疵的区分度不高。传统人工肉眼检查十分费力&＃xff0c;不能及时准确的判断出表面瑕疵&＃xff0c;质检的效率难以把控。近年来&＃xff0c;深度学习在图像识别等领域取得了突飞猛进的成果。铝型材制造商迫切希望采用最新的AI技术来革新现有质检流程&＃xff0c;自动完成质检任务&＃xff0c;减少漏检发生率&＃xff0c;提高产品的质量&＃xff0c;使铝型材产品的生产管理者彻底摆脱了无法全面掌握产品表面质量的状态。大赛数据集里有1万份来自实际生产中有瑕疵的铝型材监测影像数据&＃xff0c;每个影像包含一个或多种瑕疵。供机器学习的样图会明确标识影像中所包含的瑕疵类型。

8&＃xff09;弱监督学习下的工业光学检测&＃xff08;DAGM 2007&＃xff09; Weakly Supervised Learning for Industrial Optical Inspection

数据下载链接&＃xff1a;

https://hci.iwr.uni-heidelberg.de/node/3616

数据集介绍&＃xff1a;

• 主要针对纹理背景上的杂项缺陷。

• 较弱监督的训练数据。

• 包含是个数据集&＃xff0c;前六个为训练数据集&＃xff0c;后四个为测试数据集。

• 每个数据集均包含以灰度8位PNG格式保存的1000个“无缺陷”图像和150个“有缺陷”图像。每个数据集由不同的纹理模型和缺陷模型生成。

• “无缺陷”图像显示的背景纹理没有缺陷&＃xff0c;“无缺陷”图像的背景纹理上恰好有一个标记的缺陷。

• 所有数据集已随机分为大小相等的训练和测试子数据集。

• 弱标签以椭圆形表示&＃xff0c;大致表示缺陷区域。

9&＃xff09;基建表面裂纹检测数据 Cracks on the surface of the construction.

数据集是github上的一个项目&＃xff0c;主要是一些基建&＃xff08;水泥表面裂纹&＃xff09;&＃xff0c;可以访问项目连接&＃xff1a;https://github.com/cuilimeng/CrackForest-dataset 或者百度云链接&＃xff1a;https://pan.baidu.com/s/1108j5QbDr7T3XQvDxAzVpg

提取码&＃xff1a;jajn

Cracks on the bridge(left) and crack on the road surface.

• Bridge cracks. There are 2688 images of bridge crack without pixel-level ground truth. From the authors “Liangfu Li Weifei Ma Li Li Xiaoxiao Gao”. Files can be reached by visiting https://github.com/Charmve/Surface-Defect-Detection/tree/master/Bridge_Crack_Image.

• Crack on road surface. From Shi, Yong, and Cui, Limeng and Qi, Zhiquan and Meng, Fan and Chen, Zhensong. Original dataset can be reached at https://github.com/Charmve/Surface-Defect-Detection/tree/master/CrackForest. We extract the image files of the pixel level ground truth.

10&＃xff09;磁瓦缺陷数据集 Magnetic tile dataset

Magnetic tile dataset by ourselves. Can be downloaded from https://github.com/Charmve/Surface-Defect-Detection/tree/master/Magnetic-Tile-Defect, which was used in our paper “Surface defect saliency of magnetic tile”, the paper can be reach by here or here

Figure 1. An overview of our dataset.

This is the datasets of the paper “Saliency of magnetic tile surface defects”
The images of 6 common magnetic tile defects were collected, and their pixel level ground-truth were labeled.

11&＃xff09;铁轨表面缺陷数据集 RSDDs dataset

RSDDs数据集包含两种类型的数据集&＃xff1a;第一种是从快车道捕获的I型RSDDs数据集&＃xff0c;其中包含67个具有挑战性的图像。第二个是从普通/重型运输轨道捕获的II型RSDDs数据集&＃xff0c;其中包含128个具有挑战性的图像。

两个数据集的每幅图像至少包含一个缺陷&＃xff0c;并且背景复杂且噪声很大。

RSDDs数据集中的这些缺陷已由一些专业的人类观察员在轨道表面检查领域进行了标记。

官方链接&＃xff1a;http://icn.bjtu.edu.cn/Visint/resources/RSDDs.aspx

链接&＃xff1a;https://pan.baidu.com/share/init?surl&＃61;svsnqL0r1kasVDNjppkEwg

提取码&＃xff1a;nanr

12&＃xff09;Kylberg Texture Dataset v. 1.0

Figure 2. Example patches from each one of the 28 texture classes.

Short description

• 28 texture classes, see Figure 2.
• 160 unique texture patches per class. (Alternative dataset with 12 rotations per each original patch, 160*12&＃61;1920 texture patches per class).
• Texture patch size: 576x576 pixels.
• File format: Lossless compressed 8 bit PNG.
• All patches are normalized with a mean value of 127 and a standard deviation of 40.
• One directory per texture class.
• Files are named as follows: blanket1-d-p011-r180.png, where blanket1 is the class name, d original image sample number (possible values are a, b, c, or d), p011 is patch number 11, r180 patch rotated 180 degrees.

offical link:http://www.cb.uu.se/~gustaf/texture/

https://github.com/abin24/Surface-Inspection-defect-detection-dataset

三、Surface Defect Detection Papers

收集一些表面缺陷检测的文章&＃xff0c;主要检测对象是&＃xff1a;金属表面、LCD屏、建筑、输电线等缺陷或异常检物。方法以分类方法、检测方法、重构方法、生成方法为主。电子版论文放在了paper文件的对应日期文件下。

详见本仓库 Papers