绝缘子数据集 - 中国电力线绝缘子数据集 (CPLID)

该数据集提供了无人机捕获的正常绝缘体图像和合成缺陷绝缘体图像。 该数据集分为两部分： 1、Normal_Insulators 包含无人机捕获的正常绝缘体。正常绝缘子图像的数量为 600。 2、Defective_Insulators 包含有缺陷的绝缘体。有缺陷的绝缘子图像的数量是 248。由于我们没有太多的有缺陷的绝缘子，因此应用了数据增强方法。这些图像通过以下过程合成： 2.1 使用TVSeg中的算法从一小部分原始图像中分割出缺陷绝缘子，分割结果为掩模图像； 2.2 使用仿射变换对原始图像及其掩码进行增强，增强结果是大量的原始-掩码图像对； 2.3 使用这些图像对来训练 U-Net；使用经过训练的 U-Net 分割图像的其余部分； 2.4 在不同的背景下连接绝缘体。 这两个目录都包含两个子目录，一个叫做images，包含图像文件，另一个叫做labels，包含VOC2007格式的注释。 1、Normal_Insulators 的标签只包含绝缘体的注释； 2、Defective_Insulators 的标签不仅包含绝缘体的注释，还包含绝缘体上缺陷的注释。 图片由国家电网公司提供，数据集由王子浩制作。如果您对此数据集有任何疑问，请随时联系 [e-mail]。

This dataset provides drone-captured normal insulator images and synthetic defective insulator images. The dataset is divided into two parts:
1. Normal_Insulators: contains drone-captured normal insulators, with a total of 600 normal insulator images.
2. Defective_Insulators: contains defective insulators, with 248 defective insulator images. As only a limited number of defective insulators were available, data augmentation methods were applied. These images are synthesized via the following procedure:
2.1 Use algorithms in TVSeg to segment defective insulators from a small subset of original images, and the segmentation results are mask images;
2.2 Perform affine transformation on the original images and their corresponding masks for augmentation, resulting in a large number of original-mask image pairs;
2.3 Train a U-Net model using these image pairs, then use the trained U-Net to segment the remaining parts of the images;
2.4 Paste the insulators onto different backgrounds to synthesize new images.
Both directories contain two subdirectories: one named "images" that stores image files, and the other named "labels" that contains annotations in VOC2007 format.
1. Annotations under Normal_Insulators only include insulator annotations;
2. Annotations under Defective_Insulators include both insulator annotations and defect annotations on the insulators.
The images were provided by State Grid Corporation of China, and the dataset was compiled by Wang Zihao. If you have any questions regarding this dataset, please feel free to contact [e-mail].

搜集汇总

背景与挑战

背景概述

该数据集包含无人机捕获的正常和合成缺陷绝缘体图像，用于电力线绝缘子缺陷检测。数据集分为正常和有缺陷两部分，分别包含600张和248张图像，并提供了VOC2007格式的注释。