Geographical discrimination of Boletus edulis using two dimensional correlation spectral or integrative two dimensional correlation spectral image with ResNet

Boletus edulis (B. edulis) is a well-known edible mushroom species in the world due to its high nutritional values. However, its nutritional value varies greatly depending on geographical origins. This study aimed to discriminate the geographical regions of B. edulis by using a novel digital image method based on two dimensional correlation spectra (2DCOS) or integrative two dimensional correlation spectra (i2DCOS). In our research, 106 fruiting bodies of wild-grown B. edulis mushrooms were collected from 2011 to 2014 in 6 geographical regions. We intercepted 1750-400 cm−1 fingerprint regions from their mid-infrared (MIR) spectra, and converted them into 2DCOS or i2DCOS spectra with matlab2017b. Then, a residual convolutional neural network (ResNet) was established with 95 (90%) spectral images. In our model, the discrimination of geographical regions of the Boletus was using directly synchronous 2DCOS, asynchronous 2DCOS or i2DCOS spectral images instead of data matric from these spectra. In the synchronous 2DCOS spectra model, these 95 samples could be correctly classified as their respective regions with 100% accuracy in the train set and 100% accuracy in the test set, and all 11 (10%) samples of external validation set were discriminated correctly. The results indicated that the synchronous 2DCOS spectra model has good discrimination performance, and the new analytical method in this paper can be used for quality control of food, herb and agricultural products.