Non-Intrusive Liquid Counterfeit Identification With Low-Cost Acoustic Devices

The growing demand for high-quality liquid products, such as liquor and perfume, has intensified liquid counterfeiting, causing economic losses and potential health risks. Existing liquid authentication methods often rely on intrusive sensors, radio-frequency signals, or cameras, making them difficult to deploy conveniently and non-invasively in everyday settings. This paper presents ASLiquid+, a liquid counterfeit identification system that uses low-cost commercial off-the-shelf (COTS) acoustic devices, such as standard earphones. By placing a speaker and a microphone on opposite sides of a container, ASLiquid+ excites the liquid-container system and captures acoustic spectra that encode liquid density and solute composition through resonance and interface absorption effects. To make such sensing practical,ASLiquid^+$addresses three key challenges: hardware heterogeneity among acoustic devices, variability in user operations, and data complexity caused by changing device placements and liquid volumes. Specifically, it introduces a hardware-independent normalization method to suppress circuit noise and compensate for frequency-response differences, a user operation adaptation algorithm to extract robust frames from manual measurements, and a conditional variational autoencoder (CVAE)-based framework for data augmentation and anomaly detection under diverse physical states. We implementASLiquid^+$as an interactive smartphone application and evaluate it with six COTS earphones and a low-cost DIY acoustic device. Experiments show F1 scores of 96.0%–99.1% across five common counterfeit detection tasks and 93.3% under additional liquid-volume variations.ASLiquid^+$can also distinguish liquids with a 1% mass-fraction difference and identify solutions with identical densities but different chemical compositions.