Pix2Pix Based RAT Algorithm For Denoising of Forward-Looking Sonar Imagery

Forward-looking sonar is widely used in various underwater tasks, but the complex noise and monotonic color of sonar images make it challenging for operators to identify objects from noise. To address this issue, we propose a ROI Based Adaptive Threshold (RAT) algorithm that can effectively suppress noise in background and highlight regions of interest (ROI). However, its slow inference speed makes practical implementation difficult. Therefore, we utilize the Pix2Pix algorithm to accelerate RAT and simplify the training process. As a result, we propose the Pix2Pix Based RAT (PBR) algorithm, which achieves the best denoising performance in our experiments.