Rapid identification of total phenolic content levels in boletes by two-dimensional correlation spectroscopy combined with deep learning

Abstract

Wild boletes, as a nutritious food extensively used in food fields. It is rich in polyphenolic compounds, which is it can promote human health, improve physical performance and reduce the risk of developing diseases. But the existing methods could not evaluate the total polyphenol content (TPC) quickly and accurately. In this study, two-dimensional correlation spectroscopy (2D-COS) images were generated by high performance liquid chromatography (HPLC) and Fourier transform near-infrared (FT-NIR) spectroscopy using a generalized two-dimensional correlation algorithm. In addition, the TPC of all samples was determined. Residual convolutional neural network (ResNet) models were then established to identify different levels of TPC. The results shown that compared with HPLC, FT-NIR combined with 2D-COS algorithm had excellent identification performance and greatly reduces the time of data processing. The accuracy rates of training sets and testing sets were 100%. The established method can be effectively applied to Lanmaoa asiatica, Butyriboletus roseoflavus and Rugiboletus extremiorientalis. This work provides a novel and comprehensive strategy for other nutritious foods.