ADSCN: Adaptive dense skip connection network for railway infrastructure displacement monitoring images super-resolution

Railway infrastructure displacement monitoring (RIDM) has a pivotal role in the safety of train operation. However, due to the limitations of monitoring distance and instrument cost, the visual displacement monitoring system tends to obtain low-resolution and low-quality images, especially for key monitoring regions, which can seriously affect the monitoring performance. Improving RIDM image quality and resolution thus becomes a critically important task. In this paper, we present a novel Adaptive Dense Skip Connection Network (ADSCN) for image super-resolution to improve the quality of displacement monitoring image and the precision of displacement measurement. Specifically, by embedding dense skip connection into the generator, the low-level feature information can be fully utilized to generate high-quality super-resolution (SR) image. Furthermore, we introduce the adaptive mechanism into each skip connection to select low-level features for further performance enhancement. Finally, the discriminator is used to discriminate whether the input is a real high-resolution image or a generated SR image, which helps the generator learn to achieve better performance. Experimental results using nature images and different types of RIDM images demonstrate that our ADSCN is superior to interpolation-based and deep learning-based image SR algorithms, both in image quality and interpretation precision.