Accurate and efficient insulator maintenance: A DETR algorithm for drone imagery

Area of Science:

• Electrical Engineering
• Computer Vision
• Artificial Intelligence

Background:

• Power grid stability relies on effective maintenance and inspection of critical components like insulators.
• Drone-based inspections face challenges from adverse weather and imaging conditions, leading to low-quality data.
• Suboptimal image quality hinders accurate detection of insulator defects.

Purpose of the Study:

• To develop an insulator defect detection algorithm for low-quality drone imagery.
• To enhance the accuracy and reliability of power grid inspection systems.
• To improve the efficiency, economy, and security of power grid maintenance.

Main Methods:

• Utilized a patch diffusion model to enhance image quality.
• Introduced an optimized DETR (Detection Transformer) method with a Spatial Information Interaction Module.
• Employed a special convergence network to augment detection capabilities.

Main Results:

• Achieved a 95.8% detection accuracy for insulator defects.
• Significantly outperformed existing defect detection methods in complex environments.
• Demonstrated superior performance in handling low-quality images from drone inspections.

Conclusions:

• The proposed algorithm effectively addresses challenges in drone-based insulator inspection.
• The method offers a more accurate, efficient, and economical solution for power grid maintenance.
• Enhanced computational models lead to improved safety and stability of power grids.