ElecsDataset: a 3D semantic segmentation dataset for substation environments

Citation Author(s):
Yu
Xia
Submitted by:
Yu Xia
Last updated:
Fri, 03/28/2025 - 03:26
DOI:
10.21227/18zd-7a87
Data Format:
License:
0
0 ratings - Please login to submit your rating.

Abstract 

ElecsDataset is a specialized 3D semantic segmentation dataset designed for substation environments. It addresses the shortage of domain-specific annotated data in the field of substation 3D semantic segmentation. This dataset offers high-resolution, meticulously annotated point clouds that capture complex equipment structures and real-world occlusions. It consists of data collected from three substations of varying scales. The dataset is systematically partitioned into 6 distinct spatial regions with heterogeneous dimensions. Following standardized evaluation protocols, Area 5 is designated as the testing subset, while the remaining five areas form the training subset. Moreover, the dataset is divided into 10 categories, including non-equipment entities like busbars, electrical wires, the ground, buildings, overhead structures, walls, and miscellaneous objects, as well as substation equipment categorized into small, medium, and large scales. This comprehensive benchmark dataset is optimized for compatibility with deep neural architectures and maintains computational efficiency during model training.

Instructions: 

ElecsDataset is a 3D semantic segmentation dataset designed for substation scenarios. It offers high - resolution, finely - annotated point clouds with complex equipment and real - world occlusions, addressing the data shortage in this field. Collected from three substations, it's systematically partitioned into training and testing subsets, and further into spatially contiguous blocks for better compatibility and computational efficiency.

Dataset Construction:

1. Data Collection: Data is gathered from three substations of different sizes to ensure diversity and representativeness.

2. Regional Division: A systematic structural segmentation method divides the data into six independent spatial regions with different dimensions, adapting to the complex substation layout.

3. Category Division: The dataset has ten categories. Non - equipment entities are split into seven classes: busbars, electrical wires, the ground, buildings, towers, walls, and clutter. Substation equipment is classified into small, medium, and large based on size and point cloud count.

4. Data Annotation: Manual segmentation of substation point cloud data ensures precise labeling and data quality.

Dataset Features:

1. High - Resolution Point Clouds: The dataset contains high - resolution point cloud data that can capture the fine geometric details of substation equipment.

2. Complex Equipment Structures: It covers diverse equipment types and complex layouts in substations, such as densely arranged cables and large transformers.

3. Fine - Grained Annotation: Each point cloud is carefully labeled to ensure the accuracy and reliability of the dataset.

Dataset Applications:

ElecsDataset offers rich data support for 3D semantic segmentation research in substation scenarios and can be used to train and evaluate various algorithms. It propels the development of smart grid technology and lays the groundwork for the intelligent inspection and automation management of industrial infrastructure. By using this dataset, researchers can develop more efficient algorithms to tackle the complex challenges in substation scenarios, improving the accuracy of equipment maintenance, fault diagnosis, and safety management.