Text Localization and Recognition from Images of Substation Equipment

• Main Authors: Jie-Ling Zheng, 鄭劼靈
• Other Authors: Shih-Hau Fang
• Format: Others
• Language: zh-TW
• Published: 2018
• Online Access: http://ndltd.ncl.edu.tw/handle/4tpwmg

碩士 === 元智大學 === 電機工程學系 === 106 === Substations are an important part of the power system. Because its main equipment is exposed to nature for a long time, it not only bears the effects of normal mechanical loads and electric loads, but also suffers from external forces such as pollution, lightning strikes, strong winds, and bird damage. In order to prevent sudden equipment failures and accidents, it is necessary to conduct regular and irregular regular inspections of important substations and lines. In recent years, with the popularization of unmanned substations, substation inspection technology is gradually replacing manual inspections by robots and drones. The nameplate of the substation equipment is undoubtedly an important part of the power equipment and is an important basis for distinguishing different kinds of power equipment in the substation. The task of this paper is to complete the identification of the electric nameplate suitable for power inspection equipment to locate and identify the substation equipment. This study used deep learning Fater-RCNN algorithm to locate the nameplate of the substation, and used HSV (Hue, Saturation, Value) color space and multi-scale Rtinex method to solve complex light effects. Then use connected area detection for text positioning and character segmentation. Finally, using convolutional neural network for text recognition. The image data used in this study comes from the Xiamen substation and contains 521 substation scene images. The proposed scheme was validated on this data set. The precision of the nameplate positioning was 98.43\%, the text positioning precision was 97.23\%, and the accuracy of text recognition was 96.81\%.