ISSD: Improved SSD for Insulator and Spacer Online Detection Based on UAV System

Area of Science:

• Computer Vision
• Artificial Intelligence
• Electrical Engineering

Background:

• Power line inspection relies on identifying insulators and spacers.
• Unmanned aerial vehicle (UAV) inspections are increasingly common but require efficient onboard processing.
• Existing methods struggle with the limited computational resources of UAVs for real-time detection.

Purpose of the Study:

• To develop a lightweight, accurate, and fast detection algorithm for insulators and spacers suitable for UAV-based power inspection.
• To enable online detection of critical power infrastructure components using aerial imagery.

Main Methods:

• An improved Single Shot MultiBox Detector (SSD) algorithm was proposed.
• The MnasNet lightweight network was employed as the feature extraction backbone.
• Two multi-scale feature fusion techniques were integrated to enhance feature representation.
• A custom dataset of aerial images featuring insulators and spacers was created for training and validation.

Main Results:

• The proposed algorithm demonstrated efficient detection of insulators and spacers in real-world aerial images and videos.
• Achieved a high detection accuracy of 93.8%.
• Exhibited a small model size and fast detection speed, with a single image detection time of 154 ms and a capture rate of 8.27 fps on NVIDIA Jetson TX2.
• Outperformed existing algorithms in terms of model size and detection speed.

Conclusions:

• The developed algorithm effectively addresses the need for lightweight, high-performance object detection in UAV-based power inspection.
• The system facilitates real-time online detection of insulators and spacers, enhancing the efficiency and safety of power infrastructure monitoring.
• The combination of MnasNet and multi-scale feature fusion offers a promising approach for resource-constrained embedded vision applications.