Automatic Defect Detection of Ground Line from Images Acquired by UAV

Hiroshi Ohta, Kenta Takaya, Valeri Kroumov

Inspecting ground lines on transmission towers is necessary to maintain stable power transmission. We have already developed a UAV that automatically acquires images of the ground line for inspections. In this paper, we propose a method for detecting the ground line's defect locations from the UAV's images. In order to identify the defect locations, a ground line region from the image must be extracted in advance. The proposed method finds the ground line from multiple line segments in the image by using the edge intensity of the image. A deep neural network (DNN) identifies defect locations for each divided image of a ground line region. As a result of the experiment, a success rate of 0.978 was obtained for the extraction of a ground line from UAV images. In case of using the GoogLeNet as the DNN, an accuracy rate of 0.975 was obtained for defect detection.

IEEJ Transactions on Electronics, Information and Systems

The Institute of Electrical Engineers of Japan

145

3

408

416

0385-4221

10.1541/ieejeiss.145.408

https://www.jstage.jst.go.jp/browse/ieejeiss/list/-char/ja