Original Title: Side-Scan Sonar Image Synthesis Based on CycleGAN with 3D Models and Shadow Integration
Published in Computer Modeling in Engineering&Sciences (CMES)
Publication date: November 2025 (selected as cover paper)
Source institutions: Korea Institute of Marine Science and Technology (KIOST), National University of Busan
Technical interpretation
Side scan sonar is a core equipment for underwater exploration, target search and rescue, and geological survey, but the acquisition of high-quality measured data has long been limited by sea conditions, weather, and high operating costs. A South Korean research team has proposed a side scan sonar image synthesis technology based on generative AI (CycleGAN), providing a new approach to address this pain point.
The core innovation of this technology lies in:
3D modeling and shadow physics modeling: The research team constructs high-precision 3D models of artificial targets such as sunken ships and airplanes, and introduces precise shadow models, taking into account the distance, height, and sound scattering characteristics of sonar and targets, to simulate shadow areas that are close to the level of measured data.
Generative Adversarial Network Transformation: Utilizing CycleGAN to transform rendered images into side scan sonar images with real textures, noise, and reflection patterns, achieving efficient generation of large-scale training data.
Industry significance: This technology significantly reduces reliance on expensive offshore exploration, efficiently simulates diverse underwater environments, and lays the data foundation for AI driven sonar image automatic recognition and interpretation. This achievement marks the evolution of ocean exploration towards deep integration of "AI+physical models".
Original link: https://www.techscience.com/CMES/online/detail/24795
