A Light-Weight Network for Small Insulator and Defect Detection Using UAV Imaging Based on Improved YOLOv5.

Area of Science:

• Electrical Engineering
• Computer Vision
• Artificial Intelligence

Background:

• Power transmission line stability relies heavily on effective insulator defect detection.
• Existing object detection networks like YOLOv5 struggle with detecting small insulator defects due to poor detection rates and high computational demands.
• Unmanned aerial vehicles (UAVs) offer a promising platform for power line inspection but require efficient detection models.

Purpose of the Study:

• To develop a lightweight and efficient network for detecting small insulator defects.
• To improve the detection accuracy and reduce the computational load of insulator defect detection systems.
• To enable real-time defect detection using UAVs.

Main Methods:

• Introduced the Ghost module into the YOLOv5 backbone and neck to reduce parameters and model size.
• Incorporated small object detection anchors and layers to enhance small defect detection capabilities.
• Optimized the YOLOv5 backbone using Convolutional Block Attention Modules (CBAM) for improved focus on critical defect information.

Main Results:

• Achieved a mean average precision (mAP) of 99.4% at IoU 0.5 and 91.7% at IoU 0.5-0.95.
• Reduced model parameters to 3,807,372 and model size to 8.79 M.
• Attained a detection speed of 10.9 ms/image, meeting real-time detection requirements.

Conclusions:

• The proposed lightweight network effectively detects small insulator defects with high accuracy and efficiency.
• The model's reduced size and computational load facilitate deployment on embedded devices like UAVs.
• This advancement supports stable power transmission line operation through enhanced, real-time defect monitoring.