Transmission Line Object Detection Method Based on Contextual Information Enhancement and Joint Heterogeneous...

Area of Science:

• Electrical Engineering
• Computer Vision
• Artificial Intelligence

Background:

• Transmission line inspection is vital for power security.
• Object detection accuracy is hindered by large gaps in transmission line fittings.
• Existing methods struggle with small and obscured objects.

Purpose of the Study:

• To propose an optimized object detection method for transmission lines.
• To address the challenge of detecting small and obscured fittings.
• To improve the accuracy and reliability of transmission line inspection.

Main Methods:

• Utilized contextual information enhancement (CIE) and joint heterogeneous representation (JHR).
• Integrated convolution into the Swin transformer's self-attention for enhanced feature extraction.
• Combined diverse detection methods in the detection head for improved classification and localization.

Main Results:

• Achieved a 5.8% increase in overall mean average precision (mAP).
• Demonstrated an 18.6% increase in the average precision (AP) for normal pin detection.
• Showcased improved detection of small-sized and obscured objects.

Conclusions:

• The optimized method significantly enhances object detection accuracy for transmission lines.
• This improvement is particularly notable for small and obscured components.
• The findings lay the groundwork for real-time transmission line inspection systems.