Line-YOLO: An Efficient Detection Algorithm for Power Line Angle.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Electrical Engineering

Background:

• Manual power line tilt angle detection is labor-intensive and error-prone.
• Existing automated methods struggle with variable power line shapes and occlusions.

Purpose of the Study:

• To develop an efficient and accurate automated system for power line tilt angle detection.
• To improve upon existing object detection models for power line inspection tasks.

Main Methods:

• An improved algorithm, Line-YOLO, based on YOLOv8s-seg was developed.
• Introduced deformable convolutional DCNv4 to handle variable power line shapes.
• Integrated BiFPN for efficient feature fusion and EMA attention for improved target recognition, especially with overlapping power lines.
• Added a small target detection head to address detection of small or occluded targets.

Main Results:

• Line-YOLO achieved a 6.2% improvement in mAP@0.5 compared to the benchmark model.
• Reduced the number of parameters by 28.2% and enhanced floating-point operations per second by 35.3%.
• Increased the number of detected frames per second by 14 FPS, demonstrating improved efficiency.

Conclusions:

• The enhanced Line-YOLO model significantly improves detection accuracy and efficiency for power line tilt angle tasks.
• The algorithm effectively addresses challenges like variable shapes, target overlap, and small/occluded targets.
• Line-YOLO offers a robust and efficient solution for automated power line inspection.