SimID: Wi-Fi-Based Few-Shot Cross-Domain User Recognition with Identity Similarity Learning.

Area of Science:

• Computer Science
• Signal Processing
• Machine Learning

Background:

• Indoor user identification using Wi-Fi is a growing field for smart homes and IoT.
• Traditional deep learning methods struggle with generalization and require extensive data for new scenarios.

Purpose of the Study:

• To introduce SimID, a few-shot Wi-Fi user recognition framework.
• To overcome the limitations of conventional deep-learning classifiers in Wi-Fi-based identification.

Main Methods:

• SimID employs identity-similarity learning, embedding user-specific signal features into a high-dimensional space.
• It encourages greater pairwise similarity for samples from the same individual.
• New users are recognized by comparing query signals against a small set of stored templates, enabling few-shot identification without retraining.

Main Results:

• SimID achieves high average accuracies on the XRF55 dataset: 97.53% (cross-action), 93.37% (cross-person), 92.38% (cross-action-and-person), and 92.10% (cross-person-and-scene).
• Demonstrates robustness in few-shot scenarios, including unseen users and novel movement patterns.

Conclusions:

• SimID provides a robust and data-efficient solution for indoor identity recognition.
• The framework holds significant promise for applications in smart homes, healthcare, and security.