Machine learning based on structural and FTIR spectroscopic datasets for seed autoclassification.

Area of Science:

• Pharmacognosy
• Analytical Chemistry
• Biotechnology

Background:

• Complex biological samples often exhibit similar morphology and size, challenging accurate classification.
• Traditional identification methods for medicinal materials may lack efficiency and precision.

Purpose of the Study:

• To develop a fast and accurate protocol for identifying seed medicinal materials.
• To evaluate the effectiveness of combining micro-structural and infrared spectroscopic data for classification.

Main Methods:

• Utilized micro-computed tomography (micro-CT) and Fourier-transform infrared (FTIR) spectroscopy to extract features.
• Applied principal component analysis (PCA) and competitive adaptive reweighted sampling (CARS) for feature selection and optimization.
• Trained a back-propagation neural network (BPNN) using single and mixed feature datasets.

Main Results:

• Micro-CT dataset achieved 89.5% classification accuracy; FTIR dataset achieved 93.3% accuracy.
• The mixed dataset, combining micro-CT and FTIR features, reached a classification accuracy of 99.2%.
• The combined multi-dimensional approach significantly outperformed single-feature datasets.

Conclusions:

• A multi-dimensional characteristic architecture combining micro-CT and FTIR data offers superior performance for classifying Chinese medicinal materials.
• This protocol provides a novel and highly effective method for the identification of medicinal plant resources.
• The developed technique enhances the accuracy and speed of quality control in the pharmaceutical industry.