Klasifikasi Tingkat Kematangan Buah Kelapa Sawit Menggunakan Metode Convolutional Neural Network

Classifying the ripeness level of oil palm fruits represents a critical aspect of the oil palm industry that dominates Indonesia's economy. This research aims to develop and evaluate a Convolutional Neural Network (CNN) model to automatically, objectively, and accurately classify the ripeness level of oil palm fruits based on digital image analysis. The underlying problem of this research is the manual harvesting practice that relies on subjective assessment by harvesters, resulting in inconsistency and substantial economic losses. The research approach employs a quantitative experimental methodology with a dataset of 1,840 digital images of oil palm fruits balanced across four ripeness categories (unripe, semi-ripe, ripe, overripe). Image preprocessing was performed to standardize input with a data split of 80% training and 20% testing. The implemented CNN model achieved an average accuracy of 76.52% with optimal accuracy of 82.61%, precision of 0.77, recall of 0.77, and F1-score of 0.76 from five independent test runs. RGB profile analysis revealed a significant correlation between color pigment values and ripeness level, with extreme categories (unripe and ripe) achieving accuracy >95%, while transitional categories (semi-ripe) demonstrated higher challenges. Per-category results showed excellent F1-scores (0.946–0.983) for all classes, indicating that the model learned meaningful ripeness indicators based on biological pigment physiology. System implementation was complemented with a user-friendly Graphical User Interface (GUI) based on MATLAB, enabling non-technical operators to use the model directly. Functional black-box testing demonstrated a 100% pass rate, validating the system's readiness for operational deployment. In conclusion, CNN can be implemented as a practical solution for harvest automation that enhances objectivity, consistency, and efficiency in harvest timing determination, with positive implications for product quality, industry sustainability, and economic value added for Indonesian oil palm farmers.   Keywords: Ripeness classification, oil palm fruit, Convolutional Neural Network, digital image processing, harvest automation, machine learning, RGB analysis