Abstract

Investigating the composition ratio of ethanol–water molecular clusters in air, which are responsible for anisotropic behavior, is a significant challenge, primarily owing to the difficulty of detecting Rayleigh scattering, an inherently weak signal highly susceptible to external interference. This study overcame these limitations by integrating laser diffraction focusing with artificial neural networks.

We demonstrate a fully non-contact sensing system that circumvents the direct measurement of difficult-to-distinguish Rayleigh scattering signals. Our approach infers ethanol content by utilizing a self-fabricated multi-layered graphene Fresnel lens. An analysis of gaseous ethanol content in the 0.01%–0.1% range revealed that while a wavelength of 405 nm exhibited high sensitivity with up to a 7% intensity change corresponding to ethanol content, a wavelength of 638 nm provided superior stability for deep-learning analysis as its intensity parameter remained fixed. Ultimately, by combining the 638 nm laser with our self-developed self-aware assembly network model, we successfully inferred ethanol content with an R2 of 0.884, even under varied power conditions.

This study creates a new path for recognizing weak Rayleigh scattering signals that were previously difficult to measure and may facilitate the development of rapid and durable sensing systems, such as for the non-invasive diagnosis of gas molecules.