Machine Learning Quantification of Liberica Coffee Morphology in Lipa City

Investigating the morphological characteristics of Liberica green coffee beans presents challenges due to their irregular shapes and potential overlaps within images. This study aimed to develop a robust image analysis pipeline for quantifying green bean morphology and exploring potential variations between farms within the Lipa City region. A comprehensive coffee bean collection protocol was established, and a diverse image dataset was created. Image preprocessing, including normalization, augmentation, and enhancement techniques, optimized the data for analysis. A U-Net convolutional neural network was developed for semantic segmentation, achieving remarkable performance with an accuracy of 99.83% and an IoU score of 0.985. This enabled the precise extraction of morphological features from individual bean instances. Principal Component Analysis (PCA) revealed the primary contributors to morphological variance, and Multivariate Analysis of Variance (MANDVA) demonstrated statistically significant differences in the overall combination of morphological features between farms. These findings suggest that farm origin, potentially due to variations in environmental conditions, cultivation practices, or bean varieties, influences the physical characteristics of Liberica green coffee beans. This study provides a powerful framework for future investigations into the relationship between morphology, quality assessment, and optimization of processing techniques within the coffee industry.