关键词: 迁移学习, 卷积神经网络, 电缆隧道, 锈蚀识别, Tensorflow框架, 故障诊断与定位

Abstract: With the continuous development of the power industry, the laying of high-voltage cables and the construction and maintenance of underground cable tunnels have gradually become one of the hot issues in this field. This paper proposes a cable tunnel rust recognition method based on the integration of transfer learning and the classical convolutional neural network (LeNet5), which is able to realize the specific rust recognition of some electrical devices, such as the internal power box, fan and other equipment. The whole recognition process is based on the Tensorflow framework, and is able to solve such problems as insufficient training samples, long training time, and low recognition accuracy. Furthermore, by comparing with four classical target recognition algorithms, the proposed scheme is proved to be superior in terms of training time and recognition accuracy. The entire scheme provides a solid theoretical basis and experimental support for the realization of subsequent cable tunnel inspection robot system.

Key words: transfer learning, convolutional neural network, cable tunnel, rust recognition, Tensorflow framework, fault diagnosis and positioning