ROSAR: An Adversarial Re-Training Framework for Robust Side-Scan Sonar Object Detection

TL;DR

ROSAR is a new adversarial re-training framework that significantly improves the robustness and accuracy of deep learning object detection models for side-scan sonar images, addressing noise challenges in underwater environments.

Contribution

It introduces a novel adversarial re-training framework combining knowledge distillation and adversarial datasets for robust SSS object detection, along with three new publicly available datasets.

Findings

Robustness improved by up to 1.85%

Significant accuracy gains under SSS noise conditions

Effective adversarial attack strategies demonstrated

Abstract

This paper introduces ROSAR, a novel framework enhancing the robustness of deep learning object detection models tailored for side-scan sonar (SSS) images, generated by autonomous underwater vehicles using sonar sensors. By extending our prior work on knowledge distillation (KD), this framework integrates KD with adversarial retraining to address the dual challenges of model efficiency and robustness against SSS noises. We introduce three novel, publicly available SSS datasets, capturing different sonar setups and noise conditions. We propose and formalize two SSS safety properties and utilize them to generate adversarial datasets for retraining. Through a comparative analysis of projected gradient descent (PGD) and patch-based adversarial attacks, ROSAR demonstrates significant improvements in model robustness and detection accuracy under SSS-specific conditions, enhancing the model's robustness by up to 1.85%. ROSAR is available at https://github.com/remaro-network/ROSAR-framework.