Colorimetric Sensor for Low-Level Ammonia Detection Using Ag-Nanoparticle-Decorated Zinc Oxide Nanopellet/Polydiacetylene Composites with Machine-Learning Assistance

Colorimetric sensors have attracted considerable attention as simple and rapid tools for on-site testing, enabling the detection of specific analytes through visible color changes. In this work, ultrasensitive point-of-care colorimetric sensors integrated with machine learning (ML)-assisted analysis have been developed for the high-accuracy detection of low-level ammonia. The sensor was fabricated by combining the color-responsive properties of polydiacetylene (PDA) with the plasmonic enhancement of silver nanoparticles (Ag NPs) decorated on zinc oxide (ZnO) nanopellets. As a result, the sensor exhibits an ultrasensitive performance, capable of detecting ammonia concentrations as low as 200 ppb. The as-prepared AgZnO/PDA sensors exhibit a distinct blue color that visibly changes to red upon exposure to ammonia. Compared with pristine PDA and ZnO/PDA sensors, they demonstrate enhanced sensitivity and a faster response time. This improvement is attributed to the presence of Ag NPs, which promote stronger interaction with ammonia molecules and facilitate relaxation of the PDA backbone. Quantitative evaluation of the color response can be performed by using optical absorption spectroscopy. In addition, highly accurate qualitative and quantitative identification of ammonia was achieved through ML-assisted analysis of the absorption spectra. For on-site ammonia testing, smartphone-captured color images analyzed using an ML model have been evaluated. The model effectively accounts for variations in camera optics and ambient lighting conditions. Moreover, the practical application of the sensor demonstrated a good predictive performance in detecting ammonia content in aquarium wastewater samples. These results indicate that the AgZnO/PDA composites represent a promising platform for low-level ammonia detection, enabling high-accuracy on-site prediction through ML assistance.