Detection of defective cocoa beans using machine learning techniques and NIR spectral data fusion

Area of Science:

• Agricultural Science
• Analytical Chemistry
• Food Science and Technology

Background:

• Cocoa bean quality assessment traditionally relies on invasive cut tests, leading to bean destruction.
• There is a need for rapid, non-destructive, and objective methods for cocoa bean grading and defect detection.

Purpose of the Study:

• To develop and optimize a non-destructive method for rapid cocoa bean identification and grading.
• To evaluate the performance of portable NIR spectrometers combined with feature extraction and multivariate algorithms.

Main Methods:

• Utilized two portable NIR spectrometers (740-1070 nm and 900-1700 nm).
• Applied feature extraction techniques: Principal Component Analysis (PCA) and Linear Discriminant Analysis (LDA).
• Employed multivariate algorithms: Support Vector Machine (SVM), Artificial Neural Network (ANN), and Random Forest (RF).
• Analyzed individual and fused spectrometer data for model development.

Main Results:

• The Linear Discriminant Analysis (LDA) combined with Support Vector Machine (SVM) model demonstrated superior performance.
• Achieved over 96.39% accuracy with individual spectrometer data and 100.00% accuracy with fused data.
• LDA-SVM significantly outperformed other tested models (PCA-SVM, PCA-RF, PCA-ANN).

Conclusions:

• Integrating portable NIR spectrometers with LDA and SVM provides a reliable, non-destructive alternative to traditional cut tests.
• This approach enables rapid, onsite cocoa bean grading and defect detection.
• The study highlights the potential for improved quality control in the cocoa industry.