Acoustic camera-based super-resolution reconstruction approach for underwater perception in low-visibility marine environments

In low-visibility environments, the underwater perception range of optical cameras is severely restricted, and perception operations in ocean engineering often rely on sonar. Acoustic cameras are a type of forward-looking sonar that have attracted considerable attention because of their ability to produce images similar to those of optical cameras. However, owing to the unique imaging mechanism employed by acoustic cameras, the resulting images suffer from insufficient resolution and a loss of feature details. This issue considerably diminishes the precision of downstream visual tasks, limiting the application of acoustic cameras. In this study, we propose a deep-learning-based super-resolution reconstruction approach for acoustic cameras, where the reconstruction process relies only on images, without prior assumptions regarding the detection scenes. We verified the effectiveness of the proposed method for two practical applications: marine debris detection and marine structure inspection. The experimental results show that our proposed method can robustly reconstruct high-resolution sonar images, and the obtained images have superior feature details, which improved the precision of downstream vision tasks. In this study, we aim to provide better solutions for underwater perception in low-visibility marine environments, while exploring the application of acoustic cameras in marine debris detection and structure inspection.