Rapid Identification of Asteraceae Plants with Improved RBF-ANN Classification Models Based on MOS Sensor E-Nose

Area of Science:

• Botany
• Chemosensory Science
• Computational Biology

Background:

• Asteraceae plants are vital in Asian herbal medicine and food, necessitating robust authentication and quality control.
• Electronic nose (E-nose) technology offers a promising alternative for analyzing volatile compounds in plant materials.
• Current classification models require optimization for efficiency and accuracy in distinguishing plant species.

Purpose of the Study:

• To develop an improved radial basis function artificial neural network (RBF-ANN) classification model for Asteraceae plants.
• To enhance the discriminative capability of E-nose data through feature selection techniques.
• To identify key metal-oxide-semiconductor (MOS) sensors crucial for accurate plant classification.

Main Methods:

• Application of feature selection algorithms: Principal Component Analysis (PCA) and BestFirst + CfsSubsetEval (BC).
• Development of improved RBF-ANN models using selected features from E-nose data.
• Evaluation of classification accuracy with reduced data dimensions compared to the original model.

Main Results:

• Improved RBF-ANN models achieved 100% classification accuracy with significantly reduced data dimensions.
• Feature selection identified specific MOS sensors (S1, S3, S4, S6, S7) as highly effective for distinguishing Asteraceae plants.
• The study demonstrated the feasibility of enhancing E-nose models for plant authentication.

Conclusions:

• Feature selection effectively improves RBF-ANN models for E-nose based plant classification.
• Optimized sensor subsets can maintain high classification accuracy while reducing data complexity.
• This approach provides a foundation for advanced E-nose applications in food and herbal medicine quality control.