Fully Convolutional Network with Intermediate Reservation for Insulator Segmentation
Abstract— Insulator state detection is a challenging problem for facilitating the process of inspecting in
power transmission system. Nowadays, intense interest in applying convolution neural networks in image
analysis is wide spread, its success is impeded by the limitation of the depth of the network and is also
dependent on how to improve the information propagation and how to make full use of all the hierarchical
features. To address these problems, this paper proposed a novel framework, called as the Fully
Convolutional Network with Intermediate Reservation (FIR-Net), for insulator segmentation. In this
framework, Intermediate Reservation has been adopted to solve the problem of gradients disappearance. The
Intermediate Reservation reserves and fuses the intermediate loss of different layers, so as to improve the
propagation of the network. Overall, this framework effectively propagates features both on the shallow
layers and the deep layers, and increase the information diversity for insulator segmentation. By evaluating
the proposed framework, it has achieved the good performance on the dataset provided by STATE GRID
Corporation of China. This work is one of the early attempts of employing the idea of Intermediate
Reservation on insulator segmentation.
Index Terms— Insulator Segmentation, Deep Learning, FCN, Intermediate Reservation.
Zhen Qin, Qingya Chen, Jindou Xu
University of Electronic Science and Technology of China, CHINA
Weifu Peng, Tailong Chen, Mei Ma, Tianlong Yang
Information & Communication Company, State Grid Sichuan Electric Power Corporation, CHINA
Cite: Zhen Qin, Qingya Chen, Jindou Xu, Weifu Peng, Tailong Chen, Mei Ma, Tianlong Yang, "Fully Convolutional Network with Intermediate Reservation for Insulator Segmentation," Proceedings of 2019 the 9th International Workshop on Computer Science and Engineering, pp. 395-400, Hong Kong, 15-17 June, 2019.