A novel fast method for identifying the origin of Maojian using NIR spectroscopy with deep learning algorithms

Area of Science:

• Agricultural Science
• Analytical Chemistry
• Computer Science

Background:

• Maojian tea's origin significantly impacts its market price, leading to fraudulent mixing by merchants.
• Visual identification of Maojian tea origins is subjective and requires expert analysis.
• A rapid and objective method is needed to standardize the Maojian tea market and advance detection technologies.

Purpose of the Study:

• To develop a rapid and accurate method for distinguishing Maojian tea from different origins.
• To apply Near-Infrared (NIR) spectroscopy combined with deep learning algorithms for origin identification.
• To evaluate the performance of different deep learning models in classifying Maojian tea origins.

Main Methods:

• Collected NIR spectral data from Maojian tea samples of various origins.
• Employed deep learning algorithms: Back Propagation Neural Network (BPNN), improved AlexNet, and improved RepSet.
• Classified the spectral data using the trained deep learning models.

Main Results:

• The improved RepSet model achieved the highest classification accuracy of 99.30%.
• Improved RepSet outperformed BPNN by 8.67% and improved AlexNet by 0.70%.
• The study demonstrated the feasibility of NIR spectroscopy and deep learning for accurate Maojian origin discrimination.

Conclusions:

• NIR spectroscopy coupled with deep learning provides an effective and objective method for identifying Maojian tea origins.
• This approach can help promote standardization within the Maojian tea market.
• The developed method serves as a valuable alternative to traditional expert-based identification.