2.2. Acquisition and preprocessing of FT‐MIR spectrum data

The samples were scanned using an FT‐MIR spectrometer (PerkinElmer), and the instrument was equipped with a detector and a diamond single reflection universal attenuated total reflection (ATR) sampling accessory. The used software is SpectrumTM 10 ES. The spectra were acquired in the range 4000–450 cm⁻¹ with a resolution of 4 cm⁻¹. Each sample was scanned three times with each scan being 64 and the average was taken as the final spectrum. The original mean spectra of cultivated and wild G. elata are shown in Figure 2.

Raw mean spectral images of wild and cultivated Gastrodia elata.

Spectroscopic instruments are subject to various interferences in the actual use process. The samples have different particle sizes and uneven distribution, as well as problems with light scattering and baseline instability (Robert & Gosselin, ²⁰²²). The result is that there is some redundant information or invalid information in the obtained spectral data. The absence of spectral data processing increases the time to train the model and reduces the model accuracy. Therefore, it is necessary to preprocess the spectral data before building the model. Yan et al. (²⁰²²) used a variety of preprocessing methods before building the model with the aim of minimizing such physical effects. The use of preprocessing methods can reduce the adverse effects of the physical properties of the sample on the spectroscopic measurement process. MSC and SNV can reduce the scattering effect of the spectra; the use of derivatives can separate overlapping peaks; SG can smooth out spurious peaks (Caused by noise; Kademi et al., ²⁰¹⁹; Robert & Gosselin, ²⁰²²).

The Kennard Stone algorithm was used, and 70% of the total samples (wild: 44, cultivated: 76) were used as the training set and 30% (wild: 19, cultivated: 33) as the test set for subsequent SVM and PLS‐DA model building. Data preprocessing is performed by using 1D, 2D, MSC, SNV, and SG (third‐order derivatives with a window size of 15) and combinations of them.