ARG-Mask RCNN: An Infrared Insulator Fault-Detection Network Based on Improved Mask RCNN

Area of Science:

• Electrical Engineering
• Computer Vision
• Artificial Intelligence

Background:

• Manual inspection of power equipment defects is inefficient and prone to errors.
• Automated defect detection is crucial for ensuring power system reliability.

Purpose of the Study:

• To develop an automated, accurate, and efficient defect-detection method for power equipment, specifically damaged insulators.
• To improve upon existing Mask RCNN models for enhanced performance in defect identification.

Main Methods:

• An Attention, Rotation, Genetic algorithm Mask RCNN (ARG-Mask RCNN) model was developed using infrared imaging.
• The ResNet101 backbone was enhanced with an attention mechanism for better small target detection.
• A rotation mechanism was integrated into the loss function for precise fault localization.
• Genetic Algorithm Combined with Gradient Descent (GA-GD) optimized model hyperparameters.

Main Results:

• The ARG-Mask RCNN model achieved an average accuracy of 98% for insulator fault detection.
• The system processed at 5.75 frames per second (FPS).
• The method demonstrated superior performance compared to traditional techniques.

Conclusions:

• The proposed ARG-Mask RCNN method offers a highly accurate and efficient solution for power equipment defect detection.
• This approach significantly enhances the safety, stability, and reliability of power systems.