Sonar Image Simulator Device and Underwater Object Detection Device

Sonar Image Training Data Generation Technology Using Ray Tracing Simulation and Measured Noise Synthesis

This technology is a sonar image simulator that generates virtual sonar images using a 3D model-based ray tracing method, combines them with background noise images captured in actual underwater environments to create training images, and trains models to detect underwater objects.

Previously, it was difficult to secure large-scale sonar image datasets for training neural networks for underwater object recognition. Furthermore, collecting real-world data was time-consuming and costly, and discrepancies between simulated images and real-world environmental noise often led to degraded recognition performance.

This technology proposes a method to generate precise training data by simulating sonar images using 3D object models and sample rays, then synthesizing them with measured backgrounds to which Gaussian blur and noise level parameters have been applied. Applicable to underwater exploration, marine disaster prevention, and port surveillance, it significantly reduces data collection costs while enhancing detection performance.

‍

‍

Key Features:

• A simulator that generates sonar images of virtual objects based on ray tracing
• An imaging unit that captures actual underwater backgrounds to generate background noise images for training
• A synthesis unit that combines simulated sonar images with noise images to create training data
• A training unit that builds models for detecting underwater objects by training on the generated images

‍

This invention was developed with support from the Smart Underwater Tunnel System Research Center of the Ministry of Science and ICT.