Food and Liquid Sensing in Practical Environments Using RFIDs

Explore a groundbreaking conference talk from NSDI '20 that introduces RF-EATS, an innovative system for non-invasive sensing of food and liquids in closed containers using RFID technology. Discover how this MIT-developed solution leverages passive backscatter tags and near-field coupling to identify contents without opening containers or making direct contact. Learn about the advanced learning framework that incorporates variational inference and an RF kernel, enabling robust content identification in practical indoor environments and generalization to unseen scenarios. Delve into the system's ability to adapt to new inference tasks with minimal measurements, and examine its impressive performance in real-world applications such as detecting fake medicine, adulterated baby formula, and counterfeit beauty products. Gain insights into the technical aspects of RF-EATS, including multipath kernel utilization, anomaly detection, and its superior accuracy compared to existing RFID sensing systems.

Intro
Can we sense food and liquids in closed containers?
Approach: Exploit the wireless interaction between an RFID and the content
Experiment in a Different Environment
Decomposing the RFID Channel
Multipath Kernel Allows Generating Environment-only Features
Leverage Multipath Kernel to Detect Anomalies
Implementation
Applications Tested
Training & Testing in Different Environments
Accuracy vs Dielectric
Conclusion