AN EFFECTIVE METHOD FOR DETECTING POWER LINES, ELECTRIC POLES, AND EQUIPMENT ON THE 110 kV HIGH-VOLTAGE TRANSMISSION GRID USING DEEP LEARNING TECHNIQUES

Abstract

In the past few decades, the research, development, and application of smart grid technologies to modernize and upgrade power grids, as well as new technologies to inspect and evaluate the damage of overhead power transmission lines, have been heavily promoted toward the goal of automating the monitoring process of operation status, ensuring safety and efficiency. In this general context, the use of unmanned aerial vehicles (UAVs) to inspect and monitor high-voltage power grids is currently a matter of concern not only for the electricity industry in Vietnam but also for the electricity industry worldwide. This article proposes an effective approach to solving the problem of automatic detection of power lines and devices on the 110 kV high-voltage power grid based on the Mask R-CNN model. The effectiveness of the method is tested on a large dataset of 52,500 frames (images). The dataset includes 27,500 images collected directly on the grid in completely natural environmental conditions (partly cloudy, sunny, rainy, light wind, moderate wind) at different times of the day (morning, noon, and afternoon) and 25,000 images generated using transformation techniques, synthesis, and GAN network structures. The experimental results show that the proposed method achieves an accuracy exceeding 97%, stability, and is less sensitive to noise.