An Assembled Detector Based on Geometrical Constraint for Power Component Recognition.

Area of Science:

• Electrical Engineering
• Robotics
• Computer Vision

Background:

• Unmanned Aerial Vehicles (UAVs) have significantly improved the safety, efficiency, and quality of inspecting power lines and hard-to-reach infrastructure.
• Traditional inspection methods are often time-consuming, costly, and pose safety risks to personnel.

Purpose of the Study:

• To analyze power equipment scene characteristics and UAV operational modes for intelligent inspection.
• To develop and evaluate a novel detection method for power components using a combination of virtual and real data.
• To enhance the accuracy and efficiency of power line component detection for improved infrastructure maintenance.

Main Methods:

• Creation of a low-cost virtual scene to generate rapid training samples for power-line components.
• Proposal of an assembled detector based on geometrical constraint (ADGC) focusing on vibration-dampers.
• Integration of Faster R-CNN, Deformable Part Model (DPM), and Haar cascade classifiers, leveraging geometric positional relationships as constraints to utilize contour, shape, and texture features.

Main Results:

• The ADGC effectively detected power components by combining virtual and real data from UAV imagery.
• The proposed detector demonstrated good performance, contributing to the expansion of the training dataset.
• The method showed potential for creating improved power component detection models.

Conclusions:

• The ADGC method, utilizing virtual data generation and multi-classifier integration, offers a robust approach for power component detection.
• This technique enhances the training dataset and leads to better detection models for power infrastructure.
• The methodology is transferable to various power line facilities and inspection scenarios.