Enhanced Grading of 'Carabao' Mango for Exportation Using Three-Way Rule-Based Ensemble Method

Lead Researcher(s): Ladlennon C. Banuag, Ritchie B. Malinao, Jojimar S. Rosales, Jonnifer R. Sinogaya
Status: Published, Chito L. Patiño, and Drandreb Earl Juanico

Abstract/summary: The ‘Carabao’ mango, a significant agricultural export from the Philippines, requires accurate grading to ensure consistent quality and maximize market value. Traditional methods for mango grading, reliant on manual inspection, are prone to human error and inconsistency. While widely used for image classification, conventional Convolutional Neural Network (CNN) are limited to analyzing images of a single side of the mango, which restricts their ability to fully grade the fruit by not providing a comprehensive visual assessment of the entire mango. The research highlights a novel three-way rule-based ensemble approach of grading ‘Carabao’ mangoes using multiarchitecture CNN. The proposed method leverages multipleview analysis to improve the robustness of grading mango. By integrating ensemble method, the developed approach combines the outputs from various CNNs and utilizes different perspectives of the same mango to create a more reliable grading system and a more comprehensive analysis. The test accuracies of the three individual CNN models are 95.33%, 89.33%, and 92.67% for side, top and bottom model respectively. The three-way rule-based ensemble method achieved an accuracy of 94.33%. Despite not having a higher accuracy, it excels with a higher true negative rate and lower average false positive rate, enhancing the robustness and consistency of the mango grading process.

Keywords:

• rule-based ensemble method
• CNN
• multi-architecture
• ‘Carabao’ Mango
• multi-view analysis