Insulator instance segmentation based on deep learning network Mask RCNN

Abstract

With the continuous improvement of power supply line construction, stable power transmission provides the necessary living security for each family, but large-scale power grid maintenance has become a difficult problem for the relevant enterprise departments and an important direction of technological innovation. Transmission line inspection is an important responsibility of the power grid maintenance department. The traditional way is to manually rely on ground transportation or low-altitude manned flight for real-time monitoring. There are great challenges in labor intensity and operating conditions. Modern power grid inspection appears to be based on unmanned aerial vehicles, robots, but relatively backward, low precision recognition algorithm restricts the development of the field. This paper is based on the high-precision instance segmentation algorithm Mask RCNN to realize the identification and image segmentation of power grid insulators, and provide data sources for further image post-processing. The main idea of this paper is to analyze the working background of power inspection at this stage, analyze the development history of target detection field and explain the innovation module and development process of Mask RCNN, and finally verify the feasibility of insulator detection by building a real scene.