Rapid identification of the storage duration and species of sliced boletes using near‐infrared spectroscopy

• Published in: Journal of food science Vol. 87; no. 7; pp. 2908 - 2919
• Main Authors: Chen, Jian, Li, Jie qing, Li, Tao, Liu, Hong gao, Wang, Yuan zhong
• Format: Journal Article
• Language: English
• Published: Chicago Wiley Subscription Services, Inc 01.07.2022
• Subjects: 2DCOS images; Artificial neural networks; chemometrics; Classification; Discriminant analysis; Economic analysis; Fruit bodies; Identification methods; Image classification; Infrared spectroscopy; Machine learning; Near infrared radiation; Neural networks; NIR spectroscopy; Quality control; rapid identification; ResNet; Shelf life; sliced boletes; Species; Species classification; Spectrum analysis; Storage; Supply chains
• ISSN: 0022-1147; 1750-3841
• DOI: 10.1111/1750-3841.16220

Boletes are recognized as a worldwide delicacy. Adulteration of the expired and low‐value sliced boletes is a pressing problem in the supply chain of commercial sliced boletes. This study aimed at developing a rapid method to identify the storage duration and species of sliced boletes, using near‐infrared (NIR) spectroscopy. In the study, 1376 fruiting bodies of wild‐grown boletes were collected from 2017 to 2020 in Yunnan, containing four common species of edible boletes. A NIR spectroscopy‐based strategy was proposed, that is, identify the storage duration of sliced boletes to ensure that they are within the shelf life firstly; then identify the species of sliced boletes within the shelf life to evaluate their economic value. Three supervised methods, partial least squares discriminant analysis (PLS‐DA), extreme learning machine (ELM), and two‐dimensional correlation spectroscopy (2DCOS) images with residual convolutional neural network (ResNet) model were applied to identify. The results showed that PLS‐DA model cannot accurately identify the storage duration and species of sliced boletes, and the ELM model can identify the storage duration of boletes samples, but cannot accurately discriminate different species of samples. And ResNet model established by 2DCOS images showed superiority in classification performance, 100% accuracy was obtained for both the storage duration and species classification. Moreover, compared to traditional methods, the 2DCOS images with ResNet model was free of complicated data preprocessing. The results obtained in the present study indicated a promising way of combining 2DCOS images with ResNet methods, in tandem with NIR for the rapid identification of the storage duration and species of sliced boletes. Practical Application In the boletes supply chain, the method can be considered as a reliable method for testing the authenticity of boletes slices. The current study can also provide a reference for quality control of other edible mushroom.