A rapid and effective method for species identification of edible boletes: FT-NIR spectroscopy combined with ResNet

Abstract

The accurate and rapid species identification of edible boletes was associated to prevent food safety incidents caused by eating poisonous boletes and commercial fraud activities. This study aimed to develop an approach to accurately identify boletes species using Fourier transform near-infrared (FT-NIR) spectroscopy. A total of 418 samples were collected for the study, which included five common edible boletes species. For each sample, two portions (cap and stipe) were analyzed by FT-NIR spectroscopy. Firstly, partial least squares-discriminant analysis (PLS-DA) models were developed using the single and fused FT-NIR spectral data of cap and stipe. Then, residual convolutional neural network (ResNet) models were developed using the FT-NIR two-dimensional correlation spectroscopy (2DCOS) images of the individual portion (cap and stipe). The results showed that the ResNet model was more suitable for boletes species identification due to the easy operation and accurate classification. For both ResNet models established by the 2DCOS images of cap and stipe, samples were correctly classified as species with 100 % accuracy in the training set and test set. Furthermore, 42 external validation samples were completely identified as species. To summarize, FT-NIR spectroscopy combined with ResNet could be considered as a rapid and effective approach for identifying edible boletes species, which may a promising analytical method for edible fungi species identification.