Conventional approaches for the identification of liquid chemicals are bulky and harmful to the environment, detect a limited number of chemical species, produce high false alarm rates, or rely on complex/expensive spectrometers. In this study, a spectrometer-free, accurate metasurface-mediated liquid identification scheme was demonstrated based on optofluidic refractive index (RI) sensing in conjunction with vision intelligence algorithms. A metasurface device integrated into an optofluidic channel provides a polarization-independent focused vortex beam at a single wavelength of 1550 nm, which is highly sensitive to liquid chemicals. The beam patterns respond to the RI and transmission of chemicals, and thus effectively serve as their unique optical “fingerprints”. To realize vision intelligence, two deep-learning architectures─a convolutional neural network and a vision transformer─were adopted and trained to classify the beam patterns. A variety of liquid chemicals were successfully identified in situ with over 99% accuracy, requiring no spectrometers. The proposed approach is expected to corroborate the feasibility of artificial intelligence-powered detection schemes that can classify at single wavelengths, unlike conventional instrument-intensive techniques that are attentive to entire spectral responses.

中文翻译：

---

结合超表面的光流控折射率传感，用于通过视觉智能识别液体化学品

用于识别液体化学品的传统方法体积庞大且对环境有害，检测的化学物质数量有限，误报率高，或者依赖于复杂/昂贵的光谱仪。在这项研究中，基于光流控折射率 (RI) 传感与视觉智能算法相结合，展示了一种无光谱仪、准确的超表面介导的液体识别方案。集成到光流控通道中的超表面设备可提供 1550 nm 单一波长的偏振无关聚焦涡旋光束，该光束对液体化学品高度敏感。光束模式响应化学物质的 RI 和传输，因此有效地充当其独特的光学“指纹”。实现视觉智能，采用并训练了两种深度学习架构——卷积神经网络和视觉转换器——对波束模式进行分类。无需光谱仪，就地成功鉴定了多种液体化学品，准确率超过 99%。所提出的方法有望证实人工智能驱动的检测方案的可行性，该方案可以在单个波长下进行分类，这与关注整个光谱响应的传统仪器密集型技术不同。