Efficient Cross-Modality Insulator Augmentation for Multi-Domain Insulator Defect Detection in UAV Images.

Area of Science:

• Electrical Engineering
• Computer Vision
• Artificial Intelligence

Background:

• Regular inspection of insulators is crucial for power system stability.
• Unmanned aerial vehicle (UAV) inspection is replacing manual methods for transmission lines.
• Deep learning methods for insulator defect detection show promise but face inference complexity challenges.

Purpose of the Study:

• To propose an efficient cross-modality insulator augmentation algorithm for multi-domain defect detection.
• To address the overfitting problem and inference complexity in existing methods.
• To enhance the accuracy of insulator defect detection in real-world scenarios.

Main Methods:

• Developed a High-Resolution Insulator Cross-Modality Translation (HICT) module to generate multi-modality images, reducing modality discrepancy.
• Introduced a Multi-Domain Insulator Multi-Scale Spatial Augmentation (MMA) module to augment images across different spatial scales.
• Integrated fused images and location information to improve defect localization for various scales.

Main Results:

• The proposed algorithm achieved superior performance on public UPID and SFID insulator defect datasets.
• Demonstrated effective alleviation of the overfitting problem without additional inference resources.
• Validated the algorithm's ability to mimic complex scenarios and improve detection accuracy.

Conclusions:

• The novel cross-modality augmentation algorithm significantly enhances insulator defect detection precision.
• The method offers a valuable approach to improve transmission line inspection efficiency and safety.
• This work provides a new perspective for advancing defect detection without increasing computational demands.