Sistem Klasifikasi Kematangan Apel Fuji berdasarkan Warna menggunakan KNN untuk Sortasi Otomatis

Abstract

Manual fruit sorting typically relies on workers' visual observation to assess ripeness. This assessment is heavily influenced by individual experience and lighting conditions, often leading to inaccuracies. Furthermore, manual methods are time-consuming, increase the risk of misclassification, and reduce operational efficiency. Our research aims to develop a color-based Fuji apple ripeness classification application using the K-Nearest Neighbor algorithm that combines RGB and HSV features. Our research is developmental research using the Waterfall model, consisting of requirements analysis, design, implementation, testing, and maintenance. We used 240 fuji apple images sourced from images taken in the Kudus area. Our findings are an automatic classification application capable of classifying apple images into three ripeness levels: unripe, semi-ripe, and ripe. The evaluation results showed an accuracy of 93.75% with balanced precision, recall, and f1-score across all classes, confirming the system's stable performance without any indication of bias. Testing results using the black-box method in three scenarios opening the application, uploading an image, and reclassifying proved that all features performed as expected. The implication is that this application is ready for use in camera-based sorting in horticultural production lines and can be developed for other fruit classifications, supporting widespread post-harvest digitalization.