Simulated Sonar Image Generation Method Based on Improved CycleGAN

Abstract

The stylization method based on CycleGAN network can easily synthesize simulated sonar images when the dataset is small. However, the CycleGAN network does not retain the structure of the real sonar image, resulting in a large gap between the structure and texture of the generated image and the real image. To solve this problem, this paper proposes an improved CycleGAN algorithm. By fusing Unet and Otsu segmentation algorithms, the segmentation effect of sonar image is improved. And by adding SSIM loss, the structural characteristics of the generated image are further limited. Through the evaluation of multiple indicators, it is proved that the algorithm in this paper has improved the image similarity compared with the original CycleGAN, and better retains the structural characteristics of the original image.