Computer Vision-Based Non-invasive Sweetness Assessment of Mangifera Indica L. Fruit Using K-means Clustering and CNN

Abstract

Hailing from Guimaras, Philippines, the Carabao mango has recognition as the sweetest mango in the world. The Philippines should naturally be a top global mango exporter, for that matter. The distribution system and workforce of the country, however, are lacking. Marketing and labeling yellow, ripe Carabao mangoes as sweet when some are sour easily mislead the human eye. The automated non-invasive sorting of ripe Carabao mangoes as Super Sweet, Sweet, or Sour relative to their yellow hue, Brix value, and the range the mangoes belong under can create leverage for the Philippine mango distribution. Sixty images garnered from two (2) sides of 30 ripe Carabao mango test samples went first through Convolutional Neural Network (CNN) to segment the mango from other unnecessary fragments. The grouped most dominant colors of K-means clustering then produce RGB values in Carabao mangoes. Those RGB values correspond to Brix values, and the higher the Brix values, the sweeter the mango. Classifications of the computer vision system achieved 83.33% accuracy and 16.67% misclassification.