Fig. 1.

The dataset composition and the algorithm used for the Raman spectrometer. a. Components of 5 datasets: Dataset 1, full spectrum dataset, 400–3800 cm⁻¹; Dataset 2, fingerprint region dataset, 800–1800 cm⁻¹; Dataset 3, HW region dataset, 2800–3800 cm⁻¹; Dataset 4, background dataset; and Dataset 5, all data dataset, Datasets 1 + 2 + 3 + 4. b. Schematic diagram of the SL-Raman algorithm. The main process is as follows. (1) For each dataset, the model with the highest accuracy is selected as the base model. (2) The based model is trained using each dataset, and predictions are obtained; then, the predictions are combined into a new characteristic dataset through five-fold cross-validation. (3) The new feature dataset is input into each of the five meta-models, and the meta-model with the highest accuracy is selected for use with SL-Raman.