Implementation of Color Feature Extraction and KNN Algorithm for Oil Palm Fruit Ripeness Classification

Abstract

Determining the ripeness level of oil palm fresh fruit bunches (FFB) is crucial for Crude Palm Oil (CPO) quality, but current manual inspection in the field remains highly subjective and inconsistent. This study aims to develop and evaluate an automatic classification system for oil palm fruit ripeness using global HSV color feature extraction and the K-Nearest Neighbor (K-NN) machine learning algorithm. The research utilized a comprehensive public dataset comprising 4,728 images categorized into five ordinal ripeness classes (unripe, underripe, ripe, overripe, and decayed). The dataset was split into 80% training and 20% testing data. Images were resized, and global average values of the Hue, Saturation, and Value channels were extracted without prior background segmentation. The classification was executed using the K-NN algorithm optimized with the Euclidean Distance metric. Experimental results demonstrated that the optimal model configuration was achieved at a neighbor value of k=7, yielding an overall accuracy of 54.86%. The system showed the highest reliability in identifying the ripe class with an F1-score of 64.66% due to its prominent red-orange characteristics. However, utilizing global color averages without image segmentation led to misclassifications, particularly in transition stages and due to background noise. Therefore, it is highly recommended for future implementations to apply automatic image segmentation techniques to eliminate background interference and to combine HSV features with texture descriptors like GLCM to significantly enhance classification accuracy.