Investigation of oil palm fruit bunch ripeness classification using machine learning classifiers

Abstract

The palm oil industry, particularly in Southeast Asia, relies heavily on accurate ripeness classification of oil palm fruit bunches to ensure high-quality oil production. Despite advances in palm oil classification, distinguishing between different ripeness levels remains challenging due to subjective human judgment and labor-intensive traditional methods. This study proposes an intelligent classifier using color-based features to classify oil palm fruit bunches into three categories: ripe, half-ripe, and unripe. This framework involved capturing images of oil palm fruit bunches at Felda Chuping 2 using commercial camera, followed by image pre-processing such as resizing and cropping. Color-based features by means HSV-, RGB- and YCbCr-based features were extracted and used as significant features. The mean and standard deviation of colour-based features were then subjected to k-Nearest Neigbour (kNN) and Support Vector Machine (SVM) classifier utilizing two different strategies of hold-out and 10-fold cross-validation. Based on the results obtain, the YCbCr based features using kNN classifier achieved 97.40% (hold-out) and YCbCr based features using SVM classifier gives the highest recognition which is 100% (10-fold). The results shows that the use of colour space features able in distinguishing the ripeness levels of oil palm fruit bunches, thus considered as promising approach to be implemented in real-time application.