Underwater Object Identification Method and Apparatus Using Sonar Images

Automatic Underwater Object Identification Based on Sonar Image Registration and Difference Image Segmentation

This technology is a method and apparatus for identifying underwater objects by removing noise from sonar images, performing registration based on reference image feature points, extracting candidate object regions from background-subtracted difference images, and identifying target objects through feature analysis and eigenvector recognition.

Sonar images are often noisy and feature complex seafloor backgrounds, making manual interpretation by operators time-consuming and prone to error. This has historically limited the ability to quickly locate hazardous objects like mines to secure safe navigation routes.

This technology reduces noise using mean, Gaussian, and median filters, registers images using features such as SIFT, SURF, and HOG, and classifies objects through linear combinations of eigenvectors based on covariance matrices. It can be applied to naval mine countermeasures, marine exploration, and underwater structure inspection, helping to rapidly secure optimal safe routes (Q-routes) for vessels.

‍

‍

Key Features:

• A step of removing noise from an image acquired using sonar by using at least one of a mean, Gaussian, or median filter
• A step of registering two images by extracting and combining features that maximize similarity between a reference image and the noise-reduced image
• A step of extracting candidate object regions from a difference image from which a background image consisting of one or more of brightness, color, texture, depth, or motion properties has been removed
• A step of detecting an object based on the features of the candidate region, decomposing it into eigenvectors, and determining whether it is the target object through linear combination recognition

‍