Machine learning and deep learning based on the small FT-MIR dataset for fine-grained sampling site recognition of boletus tomentipes.

This study proposed the necessity of identifying the sampling sites for Boletus tomentipes (B.tomentipes) in combination with cadmium content and environmental factors. Based on fourier transform mid-infrared spectroscopy (FT-MIR) preprocessing by 1st, 2nd, MSC, SNV and SG, five machine learning (ML) algorithms (NB, DT, KNN, RF, SVM) and three Gradient Boosting Machine (GBM) algorithms (XGBoost, LightGBM, CatBoost) were built. To avoid complex preprocessing, we construct BoletusResnet model, propose the concepts of 3DCOS, 3DCOS projected images, index images in addition to 2DCOS, and combine them with deep learning (DL) for classification for the first time. It shows that GBM has higher accuracy than ML and DL has better accuracy than GBM. The four DL models presented in this paper achieve fine-grained sampling sites recognition based on small samples with 100 % accuracy, and a computer application system was developed on them. Therefore, spectral image processing combined with DL is a rapid and efficient classification method which can be widely used in food identification.

A new effective method for identifying boletes species based on FT-MIR and three dimensional correlation spectroscopy projected image processing.

This study proposed the necessity of identifying the species for boletes in combination with the medicinal value, nutritional value and the problems existing in the industrial development of boletes. Based on the preprocessing of Fourier transform mid-infrared spectroscopy (FT-MIR) by 1st, 2nd, SNV, 2nd + MSC and 2nd + SG, Multilayer Perceptron (MLP) and CatBoost models were established. To avoid complex preprocessing and feature extraction, we try deep learning modeling methods based on image processing. In this paper, the concept of three-dimensional correlation spectroscopy (3DCOS) projection image was proposed, and 9 datasets of synchronous, asynchronous and integrative images are generated by computer method. In addition, 18 deep learning models were established for 9 image datasets with different sizes. The results showed that the accuracy of the three types of synchronous spectral models reached 100%, while the accuracy of the asynchronous spectral and integrative spectral models of 3DCOS projection images were 96.97% and 97.98% in the case of big datasets, which overcame the defects of poor modeling effect of asynchronous spectral and integrative spectral in previous two-dimensional correlation spectroscopy (2DCOS) studies. In conclusion, the modeling results of 3DCOS projection images are perfect, and we can apply this method to other identification fields in the future.

The Storage Period Discrimination of Bolete Mushrooms Based on Deep Learning Methods Combined With Two-Dimensional Correlation Spectroscopy and Integrative Two-Dimensional Correlation Spectroscopy.

Boletes are favored by consumers because of their delicious taste and high nutritional value. However, as the storage period increases, their fruiting bodies will grow microorganisms and produce substances harmful to the human body. Therefore, we need to identify the storage period of boletes to ensure their quality. In this article, two-dimensional correlation spectroscopy (2DCOS) images are directly used for deep learning modeling, and the complex spectral data analysis process is transformed into a simple digital image processing problem. We collected 2,018 samples of boletes. After laboratory cleaning, drying, grinding, and tablet compression, their Fourier transform mid-infrared (FT-MIR) spectroscopy data were obtained. Then, we acquired 18,162 spectral images belonging to nine datasets which are synchronous 2DCOS, asynchronous 2DCOS, and integrative 2DCOS (i2DCOS) spectra of 1,750-400, 1,450-1,000, and 1,150-1,000 cm-1 bands. For these data sets, we established nine deep residual convolutional neural network (ResNet) models to identify the storage period of boletes. The result shows that the accuracy with the train set, test set, and external validation set of the synchronous 2DCOS model on the 1,750-400-cm-1 band is 100%, and the loss value is close to zero, so this model is the best. The synchronous 2DCOS model on the 1,150-1,000-cm-1 band comes next, and these two models have high accuracy and generalization ability which can be used to identify the storage period of boletes. The results have certain practical application value and provide a scientific basis for the quality control and market management of bolete mushrooms. In conclusion, our method is novel and extends the application of deep learning in the food field. At the same time, it can be applied to other fields such as agriculture and herbal medicine.

Deep learning for species identification of bolete mushrooms with two-dimensional correlation spectral (2DCOS) images.

Bolete is well-known and widely consumed mushroom in the world. However, its medicinal properties and nutritional are completely different from one species to another. Therefore, the consumers need a fast and effective detection method to discriminate their species. A new method using directly digital images of two-dimensional correlation spectroscopy (2DCOS) for the species discrimination with deep learning is proposed in this paper. In our study, a total of 2054 fruiting bodies of 21 wild-grown bolete species were collected in 52 regions from 2011 to 2014. Firstly, we intercepted 1750-400 cm-1 fingerprint regions of each species from their mid-infrared (MIR) spectra, and converted them to 2DCOS spectra with matlab2017b. At the same time, we developed a specific method for the calculation of the 2DCOS spectra. Secondly, we established a deep residual convolutional neural network (Resnet) with 1848 (90%) 2DCOS spectral images. Therein, the discrimination of the bolete species using directly 2DCOS spectral images instead of data matric from the spectra was first to be reported. The results displayed that the respective identification accuracy of these samples was 100% in the training set and 99.76% in the test set. Then, 203 samples were accurately discriminated in 206 (10%) samples of external validation set. Thirdly, we employed t-SNE method to visualize and evaluate the spectral dataset. The result indicated that most samples can be clustered according to different species. Finally, a smartphone applications (APP) was developed based on the established 2DCOS spectral images strategy, which can make the discrimination of bolete mushrooms more easily in practice. In conclusion, deep learning method by using directly 2DCOS spectral image was considered to be an innovative and feasible way for the species discrimination of bolete mushrooms. Moreover, this method may be generalized to other edible mushrooms, food, herb and agricultural products in the further research.