Abstract: In an effort to provide reference for the identification and quality control of edible fungi in Yunnan, Fourier transform mid-infrared spectroscopy combined with support vector machine (SVM) was used to identify different wild bolete species in Yunnan. The influence of different data mining methods on model classification performance was determined. Infrared spectra of 8 species with 827 samples of eight common wild bolete species were acquired and analyzed for spectral characteristics, and a discriminant model was established using SVM. Spectral information mining was performed by preprocessing, feature variable extraction or their combination, and the classification performance of the models developed was compared with each other to find the optimal method for species identification of wild bolete. The results showed that there was a large amount of noise and interference information in the original data, which reduced the classification performance of the model. All tested data mining methods could remove non-effective information but in different extents, improving model classification performance. Preprocessing combined with characteristic variables extraction had the highest ability to mine spectral information, providing the best model classification performance. The SVM model developed had excellent goodness of fit with high classification accuracy and wide applicability, allowing the accurate and quick identification of the eight wild bolete species.

Key words: wild bolete; species identification; data mining; Fourier transform mid-infrared spectroscopy; support vector machine