Combination of gray level features with deep transfer learning for copra classification using machine learning and...

Area of Science:

• Agricultural Science
• Computer Science
• Image Processing

Background:

• Copra, dried coconut, is vital for oil and by-product production.
• Traditional sun-drying and industrial sulphur fumigation impact copra quality.
• Accurate classification of copra types is crucial for buyers.

Purpose of the Study:

• To develop and evaluate a robust method for classifying sulphur-fumigated copra versus normally dried copra.
• To enhance transparency and benefit copra farmers and buyers through accurate quality assessment.

Main Methods:

• Collected and segmented images of copra from drying industries.
• Combined Gray-Level Co-Occurrence Matrix (GLCM) features with transfer learning model features.
• Evaluated feature sets using machine learning classifiers and neural networks.

Main Results:

• Neural Network-based Pattern Recognition (NNPR) achieved the highest accuracy (99.6%), sensitivity (99.64%), specificity (99.64%), F1-Score (99.6), and Kappa score (0.99).
• Other classifiers like Random Forest (98.9% accuracy), Logistic Regression (98.3%), and KNN (98.3%) also showed high performance.
• The proposed methodology significantly outperforms existing literature in copra classification.

Conclusions:

• The developed approach provides a highly accurate and reliable method for classifying sulphur-fumigated copra.
• This classification system offers practical utility for stakeholders in the copra industry.
• The study demonstrates the effectiveness of combining GLCM and transfer learning features for image-based copra quality assessment.