LiquidAuth: Reliable and Accurate Liquid Authentication Using GAN-enhanced Acoustic-to-Mass-Spectrum Mapping

Counterfeit and adulterated liquids present significant health risks and economic losses, underscoring the need for effective authentication methods. While acoustic signal-based detection offers a promising non-invasive approach that works without opening containers, its accuracy suffers from variations in container properties and environmental conditions. Furthermore, acoustic features alone lack molecular-level detail needed for definitive identification. We address these challenges with LiquidAuth, a system that maps acoustic signals to mass spectra, providing molecular-level insights for more accurate authentication. LiquidAuth employs a cross-shaped microphone array to mitigate positional variation and introduces an adaptive container compensation algorithm to account for different container characteristics. By integrating Conditional GANs (cGANs), our system effectively maps acoustic signals to mass spectra, enabling reliable molecular-level classification. Experimental evaluations show LiquidAuth achieves an average F1-score of 97.89%, with accuracy between 95.35% and 98.25% across various container materials and storage conditions, demonstrating robust liquid authentication capabilities.