Image-based Super Resolution of Underwater Sonar Images using Generative Adversarial Network

Sonar sensors are widely used for underwater observations because they can be used in a turbid stream and have a long operating range. However, images taken with sonar sensors are difficult to identify because of low resolution. In this paper, we proposed a method based on a generative adversarial network to increase the resolution of the underwater sonar image. We built the network of 16 residual blocks and eight convolutional layers. We then trained it with sonar images cropped in several ways. As a result, we could improve the resolution of sonar images from various scenes and recorded a higher peak signal-to-noise ratio than interpolation. The proposed method could help to identify the underwater object without losing the working range of sonar.