Overview of Image Datasets for Deep Learning Applications in Diagnostics of Power Infrastructure.

Area of Science:

• Engineering
• Computer Science

Background:

• The power sector relies on extensive infrastructure requiring regular maintenance.
• Manual inspection of power infrastructure is labor-intensive and challenging due to vast geographical distribution.
• Advances in deep learning offer potential for automating power line inspection processes.

Purpose of the Study:

• To review existing machine learning datasets relevant to power infrastructure inspection.
• To highlight the necessity of domain-specific datasets for effective deep learning applications.
• To introduce and discuss the potential applications of a novel dataset of power line insulators.

Main Methods:

• Literature review of machine learning and deep learning datasets.
• Comparative analysis of selected image data repositories for power infrastructure.
• Characterization and introduction of an original dataset comprising 2630 high-resolution labeled images of power line insulators.

Main Results:

• Identified and compared various repositories of domain-specific datasets for power infrastructure.
• Emphasized the critical need for curated, expert-labeled datasets in deep learning for specialized fields.
• Presented a new, high-resolution labeled image dataset focused on power line insulators.

Conclusions:

• Specialized, high-quality datasets are essential for advancing deep learning in power sector maintenance.
• The newly introduced insulator image dataset offers significant potential for developing automated inspection tools.
• Further research can leverage this dataset to improve the efficiency and accuracy of power infrastructure monitoring.