In heavy rain, the catenary images collected by railway detection devices have a severe noise problem. The rain streaks in the image significantly affect the image quality, decreasing the accuracy and efficiency of the automatic identification of catenary components. Therefore, this paper proposes a deep learning-based deraining method of blurry catenary images under heavy rain to solve the problem. This method adopts a two-stage architecture, consisting of the encoder-decoder structure and single-scale convolution, respectively. And a supervised attention module is added to every stage to improve feature transmission efficiency. The experiment results prove that our method can effectively improve the accuracy of component positioning.

Catenary; Deep learning; Supervised attention module; Image deraining; Component positioning