Power Line Segmentation Algorithm Based on Lightweight Network and Residue-like Cross-Layer Feature Fusion.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Electrical Engineering

Background:

• Power line segmentation is crucial for safe UAV inspections.
• Existing deep learning models face challenges with small targets, complex backgrounds, and large parameter counts.

Purpose of the Study:

• To develop a lightweight and accurate deep learning model for power line segmentation.
• To address the limitations of current algorithms in UAV inspection scenarios.

Main Methods:

• Introduced RGS-UNet, a lightweight segmentation model with a modified UNet backbone (ResNet18) and Ghost Module optimization.
• Integrated a SIMAM attention mechanism via residual-like addition for enhanced feature extraction.
• Utilized the Mish activation function to maintain accuracy and prevent overfitting.

Main Results:

• RGS-UNet achieved 2.05% and 2.58% improvements in F1-Score and IoU compared to classical UNet.
• The model's parameter count was reduced to 57.25% of the original UNet.
• Demonstrated superior performance in both accuracy and lightweighting.

Conclusions:

• RGS-UNet offers an effective solution for power line segmentation in UAV inspections.
• The model's lightweight nature and improved accuracy make it suitable for edge-side deployment.
• This research contributes to safer and more efficient transmission line monitoring.