Rotational Photonic Spin Hall Effect Sensor for Liquid Chemicals Classification via EfficientNet-V2

Abstract

Photonic spin Hall effect (PSHE) has been widely used for sensing tasks; however, its potential appears to be unexplored for the development of a compact yet effective sensor for the classification of liquid chemicals. In this study, a liquid identification scheme is demonstrated based on the recently proposed rotational PSHE, where the weak measurement techniques are no longer required for sensing. A liquid crystal device is fabricated to experimentally validate the rotational PSHE, which provides unique beam patterns for liquid analytes. The collected beam pattern images are used to train an EfficientNet-V2─a fast and efficient deep learning architecture─for classifying the liquid chemicals. Two groups of liquids are identified with accuracy over 99% in the proposed scheme. Moreover, the performances of several deep learning models are compared, demonstrating the fast training speed and high parameter efficiency of the EfficientNet-V2. The proposed approach provides an efficient, accurate, and convenient method for refractive index sensing.