Powerline Components and Faults Dataset
Overview
The Powerline Components and Faults Dataset is a dataset designed for object detection tasks involving powerline components and associated faults. It provides images of powerline infrastructure along with annotated bounding boxes for various components and faults.
This dataset is augmented with mosaic augmentation useful for training and evaluating models on powerline inspection, maintenance, and safety applications.
Dataset Structure
The dataset is organized into the following directories:
train/: Contains training images and their corresponding annotation files.validation/: Contains validation images and their corresponding annotation files.test/: Contains test images and their corresponding annotation files.
Each image file has a corresponding .txt file in the same directory, which contains the annotations in YOLO format.
Data Format
Images
- Format: JPEG/PNG
- Resolution: Various resolutions
Annotations
Annotations are provided in YOLO format, where each line in a .txt file corresponds to an object in the image. The format is:
class_id x_center y_center width height
class_id: The ID of the object class.x_center,y_center: The center of the bounding box (normalized between 0 and 1).width,height: The dimensions of the bounding box (normalized between 0 and 1).
Usage
You can use this dataset with popular machine learning frameworks and libraries. Below is an example of how to load the dataset using the Hugging Face datasets library:
from datasets import load_dataset
# Load the dataset
dataset = load_dataset("docmhvr/powerline-components-and-faults")
# Access the train, validation, and test splits
train_dataset = dataset['train']
val_dataset = dataset['validation']
test_dataset = dataset['test']
License
This dataset is provided under the MIT License. See the LICENSE file for more details.
Acknowledgements
This dataset was created as part of the research work on powerline inspection and fault detection. Data was collected using DJI Mini drone and manually compiled and annotated using Roboflow.
Research reference
You can find the related Research work published in IEEE, full text avaliable on researchgate here,
Contribution
If you would like to contribute to this dataset, please feel free to open an issue or submit a pull request on the GitHub repository.