FG-LiquID: A Contact-less Fine-grained Liquid Identifier by Pushing the Limits of ...
summary
A compact prototype uses a 60 GHz FMCW radar and FFT-based peak-zone extraction to classify liquids by permittivity, training a custom neural net called RC-Net that calibrates multi-antenna (Rx) signals to remove location interference (pairwise Pearson >0.85). With 3,000 samples across five indoor scenarios the system distinguishes 30 liquids—including 1% alcohol differences like 52% vs 53%—and reports about 97.5% overall accuracy and >92.5% on fine-grained liquor pairs, generalizing to an unseen environment.
AI-written summary · about