A research group headed by Prof. Hu Bingliang and Prof. Yu Tao from the Xi'an Institute of Optics and Precision Mechanics (XIOPM), part of the Chinese Academy of Sciences, has created the physicochemical-informed spectral Transformer (PIST) model alongside ultraviolet-visible-shortwave near-infrared (UV–vis-SWNIR) spectroscopy for monitoring water quality. This work represents their latest study. published in Environmental Science & Technology and was chosen as a featured article.
By incorporating essential physicochemical attributes into the Transformer model, our investigation aims to advance AI technology," stated Professor Liu Jiacheng from XIOPM, who is also the lead author of this work. "These enhanced spectroscopic techniques have significant potential in environmental surveillance, serving as robust instruments.
Spectroscopy, which identifies the absorption of light at particular wavelengths by electrons within pollutant molecules, is extensively utilized in assessing water quality. The primary difficulty involves developing precise and broadly applicable models to establish a quantitative connection between spectral readings and indicators of water quality.
Conventional methods like partial least squares (PLS) and convolutional neural networks (CNNs), despite their widespread use, have constraints when it comes to extracting spectral features and performing well under different conditions.
This research presents the PIST architecture—a pioneering Transformer-based model designed specifically for analyzing water quality through spectroscopy. The design includes a physicochemical-informed segment aimed at facilitating domain adjustment alongside a feature embedding section meant to boost comprehensive feature retrieval, which notably enhances overall model effectiveness.
The experimental verification validated the remarkable precision and versatility of the PIST model in assessing chemical oxygen demand (COD) levels. When integrated with UV–vis-SWNIR spectroscopy, this model offers an efficient, dependable, and cost-effective approach for evaluating water quality. Additionally, once paired with field-based spectrometry equipment, it can facilitate extended ecological surveillance as well as swift reactions to aquatic contamination incidents.
This research represents progress in using AI-powered techniques for environmental conservation, providing fresh possibilities for smart surveillance of water quality.
More information: Jiacheng Liu and colleagues presented a method for rapid COD sensing in complex surface water using a physicochemical-informed spectral transformer combined with UV-Visible-shortwave near infrared spectroscopy. Environmental Science & Technology (2025). DOI: 10.1021/acs.est.4c14209
Provided by the China Academy of Science
This tale was initially released on Massima . Subscribe to our newsletter For the most recent science and technology news updates.