Lightweight Transmission Line Fault Detection Method Based on Leaner YOLOv7-Tiny.

Area of Science:

• Electrical Engineering
• Computer Vision
• Artificial Intelligence

Background:

• Existing transmission line fault detection algorithms suffer from high parameter complexity and computational load.
• These limitations hinder the deployment of fault detection systems on resource-constrained platforms like drones.
• Accurate and efficient detection of transmission line faults is crucial for grid reliability.

Purpose of the Study:

• To propose a novel lightweight object detection model, Leaner YOLOv7-Tiny, for swift and accurate transmission line fault detection from aerial imagery.
• To reduce model size and computational requirements for practical drone-based applications.
• To enhance the detection accuracy of small and typical transmission line faults.

Main Methods:

• Developed Leaner YOLOv7-Tiny by modifying the YOLOv7-Tiny architecture.
• Replaced the backbone with depthwise separable convolutions to decrease model parameters.
• Integrated SP attention mechanism for multi-scale feature fusion and improved small target recognition.
• Introduced an improved FCIoU Loss function to balance sample contributions and accelerate convergence.

Main Results:

• Achieved a 20% reduction in model size compared to the original YOLOv7-Tiny.
• Demonstrated superior detection accuracy for small targets compared to existing mainstream lightweight object detection algorithms.
• The model exhibits enhanced detection speed and accuracy for transmission line faults.

Conclusions:

• Leaner YOLOv7-Tiny offers a practical and efficient solution for transmission line fault detection.
• The lightweight design and improved accuracy make it suitable for drone-based inspection systems.
• This approach significantly advances the capabilities for real-time monitoring and maintenance of power transmission infrastructure.