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

Area of Science:

• * Spectroscopy and Chemometrics
• * Machine Learning and Deep Learning
• * Mycology and Fungal Biotechnology

Background:

• * Accurate identification of bolete species is essential for understanding their medicinal and nutritional properties.
• * Traditional spectroscopic methods like Fourier Transform Mid-Infrared Spectroscopy (FT-MIR) often require complex preprocessing and feature extraction.
• * Existing methods for analyzing spectroscopic data, such as Multilayer Perceptron (MLP) and CatBoost, have limitations in modeling complex spectral information.

Purpose of the Study:

• * To develop a novel, simplified approach for bolete species identification using advanced spectroscopic imaging.
• * To evaluate the effectiveness of deep learning models applied to three-dimensional correlation spectroscopy (3DCOS) projection images.
• * To overcome the limitations of previous two-dimensional correlation spectroscopy (2DCOS) methods in spectral data analysis.

Main Methods:

• * Generation of 9 datasets of synchronous, asynchronous, and integrative 3DCOS projection images using computer methods.
• * Establishment of 18 deep learning models tailored for different sizes of the generated image datasets.
• * Application of deep learning algorithms to analyze 3DCOS projection images for bolete species identification.

Main Results:

• * Synchronous spectral models achieved 100% accuracy in bolete species identification.
• * Asynchronous and integrative spectral models using 3DCOS projection images demonstrated high accuracy (96.97% and 97.98%) on large datasets.
• * The 3DCOS projection image method significantly improved upon the modeling performance of previous 2DCOS studies.

Conclusions:

• * Three-dimensional correlation spectroscopy (3DCOS) projection images offer a powerful and accurate method for bolete species identification.
• * Deep learning models applied to 3DCOS images provide a robust alternative to traditional spectroscopic preprocessing techniques.
• * The developed methodology shows potential for application in other identification fields beyond boletes.