Mung bean seed classification based on multimodal features and Kepler-optimized stacking ensemble learning model.

Area of Science:

• Agricultural Science
• Data Science
• Spectroscopy

Background:

• Accurate mung bean seed classification is crucial for optimizing crop yields and nutritional value.
• Current classification methods are often slow, inaccurate, and lack diverse feature sets.

Purpose of the Study:

• To develop a rapid, accurate, and non-destructive method for mung bean seed variety classification.
• To create a multimodal dataset integrating Raman spectral and image-based features.
• To optimize a stacking ensemble learning model for improved classification performance.

Main Methods:

• A multimodal dataset was created by early fusion of Raman spectral data (44 features) and image-based data (15 features), totaling 59 features selected via Competitive Adaptive Reweighted Sampling (CARS).
• The Kepler Optimization Algorithm (KOA) was employed to optimize parameters for various machine learning models (DT, SVM, KNN, BPNN, RF, GBDT).
• A stacking ensemble learning model was constructed using the optimized models to enhance classification accuracy.

Main Results:

• The Kepler-optimized stacking ensemble model achieved a classification accuracy of 90.71% for mung bean seeds.
• This represents a significant improvement over existing methods, outperforming KOA-RF by 3.24% and KOA-GBDT by 1.59%.
• The proposed multimodal approach demonstrated superior efficiency compared to baseline models.

Conclusions:

• Combining multimodal features (Raman spectra and images) with a KOA-optimized stacking ensemble learning model offers a powerful solution for precise mung bean seed classification.
• This study highlights the potential of artificial intelligence in revolutionizing agricultural practices.
• The developed method provides valuable technical support for the agricultural industry, enabling better seed selection and management.