Statistically-Guided Dual-Domain Meta-Learning with Adaptive Multi-Prototype Aggregation for Distributed Fiber Optic Sensing

2025-11-21

AI Summary

Distributed fiber-optic sensing (DFOS) faces three key challenges in cross-deployment vibration recognition: significant domain shift induced by heterogeneous deployment environments, extreme scarcity of labeled data in novel scenarios, and difficulty in modeling intra-class diversity within source domains. To address these, we propose Dual-domain Multi-Prototype Meta-learning (DMP-Meta), a novel framework that jointly exploits time-frequency dual-domain features and incorporates statistical guidance for prototype sensitivity modeling. We further design a query-aware adaptive multi-prototype aggregation mechanism to enable few-shot cross-domain knowledge transfer and robust generalization—without requiring any labels from the target domain. DMP-Meta effectively mitigates distributional shift while enhancing prototype discriminability. Evaluated on a cross-deployment DFOS benchmark, it achieves substantial improvements over state-of-the-art baselines using only 1–5 labeled samples per class, attaining +8.2% higher F1 score and superior generalization capability.

Technology Category

• Machine Learning: Transfer, Domain Adaptation, Multi-Task Learning
• Search and Optimization: Distributed Search
• Computer Vision: Multi-modal Vision

Application Category

• Systems and Infrastructure for Web, Mobile and WoT: Experiences and lessons learnt from Web-based algorithms and system deployments
• Search and Retrieval-Augmented AI: Vertical and domain-specific search
• Graph Algorithms and Modeling for the Web: Representation, reconstruction, and subgraph or motif discovery in Web-related graphs

Abstract

Distributed Fiber Optic Sensing (DFOS) has shown strong potential in perimeter security due to its capability of monitoring vibration events across long distances with fine spatial resolution. However, practical DFOS systems face three critical challenges: (1) signal patterns of the same activity vary drastically under different fiber deployment types (e.g., underground, wall-mounted), causing domain shift; (2) labeled data in new deployment scenarios is often scarce or entirely unavailable, limiting model adaptability; and (3) even within source domains, data scarcity makes it difficult to capture intra-class diversity for robust learning. To address these challenges, we propose a novel meta-learning framework, DUPLE, for cross-deployment DFOS activity identification. First, a dual-domain multi-prototype learner fuses temporal and frequency domain features, enhancing the model's generalization ability under signal distribution shifts. Second, a Statistical Guided Network (SGN) infers domain importance and prototype sensitivity from raw statistical features, providing data-driven prior information for learning in unlabeled or unseen domains. Third, a query-aware prototype aggregation module adaptively selects and combines relevant prototypes, thereby improving classification performance even with limited data. Extensive experiments on cross-deployment DFOS datasets demonstrate that our method significantly outperforms baseline approaches in domain generalization settings, enabling robust event recognition across diverse fiber configurations with minimal labeled data.

Problem

• Addresses domain shift in fiber optic sensing under varied deployments
• Solves limited labeled data availability in new deployment scenarios
• Handles intra-class diversity challenges with scarce source domain data

Innovation

• Dual-domain multi-prototype learner fuses temporal and frequency features
• Statistical Guided Network infers domain importance from raw statistics
• Query-aware prototype aggregation adaptively selects relevant prototypes