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

2025-12-01

AI Summary

Underwater sonar image classification suffers from model bias and poor generalization due to overreliance on seafloor background cues. Method: This paper proposes a novel framework integrating contrastive machine unlearning with explainable AI (XAI). It introduces a Target-Contrastive Unlearning (TCU) module to suppress background bias and establishes the “Unlearn-to-Explain Sonar” framework, adaptively coupling the unlearning process with LIME for fine-grained, trustworthy attribution visualization. Additionally, an enhanced triplet loss and supervised contrastive learning are incorporated to improve unlearning precision. Contribution/Results: Experiments on both real and synthetic sonar datasets demonstrate that the method significantly reduces background bias, enhances model robustness and cross-scene generalization, and provides verifiable, interpretable decision rationales—establishing a new paradigm for safe and trustworthy underwater AI recognition.

Technology Category

• Machine Learning: Calibration & Uncertainty Quantification
• Humans and AI: Explainable AI (XAI) for Human Understanding
• Computer Vision: Remote Sensing / Geospatial AI

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Security and Privacy: Security and privacy of machine learning and AI applications

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.

Problem

• Mitigates seafloor background bias in sonar classification
• Enhances model generalization via targeted contrastive unlearning
• Provides explainable AI for visualizing forgotten features

Innovation

• Targeted Contrastive Unlearning reduces seafloor background bias
• Unlearn to Explain Sonar Framework visualizes forgotten model features
• Adapted LIME explainer provides faithful attributions for unlearning evaluation