Supervised Contrastive Machine Unlearning of Background Bias in Sonar Image Classification with Fine-Grained Explainable AI

TL;DR

This paper introduces a novel framework for sonar image classification that unlearns background bias and enhances interpretability, leading to better generalization and robustness in object detection tasks.

Contribution

It proposes a Targeted Contrastive Unlearning module and an Unlearn to Explain framework, advancing bias mitigation and explainability in sonar image AI models.

Findings

Significant reduction in background bias effects.

Improved model robustness and generalization.

Enhanced interpretability through localized explanations.

Abstract

Acoustic sonar image analysis plays a critical role in object detection and classification, with applications in both civilian and defense domains. Despite the availability of real and synthetic datasets, existing AI models that achieve high accuracy often over-rely on seafloor features, leading to poor generalization. To mitigate this issue, we propose a novel framework that integrates two key modules: (i) a Targeted Contrastive Unlearning (TCU) module, which extends the traditional triplet loss to reduce seafloor-induced background bias and improve generalization, and (ii) the Unlearn to Explain Sonar Framework (UESF), which provides visual insights into what the model has deliberately forgotten while adapting the LIME explainer to generate more faithful and localized attributions for unlearning evaluation. Extensive experiments across both real and synthetic sonar datasets validate our approach, demonstrating significant improvements in unlearning effectiveness, model robustness, and interpretability.