Forward-Looking Sonar Image Segmentation Using an Improved U-Net Model

Forward-Looking Sonar (FLS) image segmentation holds significant importance in marine engineering. However, existing image segmentation algorithms struggle to extract features from FLS images due to weak semantic information, complex backgrounds, and environmental noise. This paper proposes an improved model based on U-Net, incorporating transformer modules to effectively segment edge details in FLS images. Experimental results demonstrate that the improved U-Net algorithm achieves superior performance in FLS image segmentation tasks.