AI-enhanced multimodal spectroscopy for monitoring volatile organic compounds in pharmaceutical wastewater

Pharmaceutical wastewater commonly contains volatile organic compounds (VOCs) such as methanol, isopropanol, and acetone, which pose serious threats to wastewater treatment efficiency, ecological systems, and human health. Therefore, the rapid and robust quantitative monitoring of VOCs in wastewater has become an urgent necessity to ensure treatment effectiveness and environmental safety. However, traditional detection methods suffer from issues such as complex operations and delayed responses, failing to meet the monitoring requirements of industrial sites. To address these challenges, this study proposes an artificial intelligence-enhanced multimodal spectral fusion framework for efficient and rapid VOC analysis. Specifically, the adaptive weighted feature fusion (AWFF) method integrates near-infrared (NIR) and Raman spectral data to construct a rich and balanced feature representation. Building upon this, a lightweight multi-scale residual network (LMRN) is developed to precisely predict multi-component concentrations. Experimental results demonstrate that the AWFF-LMRN model achieves R² values of 0.9501, 0.9499, and 0.9498 for methanol, isopropanol, and acetone, respectively, with root mean square errors (RMSE) of 375.16, 287.27, and 357.54 mg/L, showing good predictive accuracy and generalization capability. Moreover, the model exhibits strong interpretability by elucidating the intrinsic relationships between spectral features and target concentrations. These findings provide robust technical support for VOC monitoring, with potential to enhance wastewater treatment efficiency and promote environmental sustainability.