Machine learning-assisted three-dimensional fluorescence for heavy metal multi-sensing

Heavy metal pollution is pervasive, complex, and diverse, causing lasting and potentially toxic effects on human health. Therefore, it is necessary to develop high-throughput detection of heavy metals. In this work, we proposed a machine learning-based three-dimensional fluorescent sensor strategy for effective detection of heavy metal pollution. Specifically, a multi-emitting fluorescent "chemical nose/tongue" sensor was first constructed with two different colored CDs and 5,10,15,20-tetrakis(4-trimethylamino) phenyl porphyrin (TAPP) for discriminating nine heavy metals, and then the fluorescence fingerprints were obtained based on the changes in the three-dimensional fluorescence spectra influenced by various heavy metals. The results demonstrated that the sensor easily and rapidly discriminated the categories and concentrations of multiple single heavy metal ions and binary mixed heavy metal ions within the concentration range of 10 μM–0.5 mM. Subsequently, the data input way was optimized, and the results showed that the accuracy of image recognition in predicting categories and concentrations of heavy metal ions was 75.1 % and 64.4 %, and the accuracy of data recognition was 97.8 % and 100 %. After optimizing the learning model, the Random Forest (RF) model can predict the categories and concentrations of a binary heavy metal mixture with accuracies of 100 % and 91.8 %, respectively. In addition, machine learning-assisted fluorescent sensors showed promising applications in real tap water and river samples. In conclusion, compared to previous analysis methods, this approach achieved qualitative and quantitative analysis of heavy metals in complex environments through three-dimensional fluorescence spectra combined with the ML strategy. Notably, the sensor did not involve cumbersome synthesis of multiple sensing channels, making it easy to construct and cost-effective.