Classification Of Palm Oil Maturity Using CNN (Convolution Neural Network) Modelling RestNet 50

Abstract

Accurate classification of palm fruit maturity levels is very important to optimize harvest time and increase production efficiency in the palm oil industry. Traditional methods that rely on visual assessment of factors such as fruit shedding and skin discoloration are prone to human error. To overcome this limitation, this research applies deep learning techniques, specifically using Convolutional Neural Network (CNN) with ResNet-50 architecture, to classify Fresh Fruit Bunches (FFB) into two stages of maturity: unripe and ripe. The model is trained and validated using a combination of data augmentation techniques to improve model performance. Various configurations were tested, including variations in data sharing, optimizer, and learning rate. The optimal configuration—90/10 training and validation data split, Adam optimizer, and learning rate of 0.0001—resulted in excellent model performance. The ResNet-50 model achieved 97% accuracy, with 96% precision, 98% recall, and an F1 score of 97%. This metric reflects the high reliability of the model in classifying palm fruit maturity levels, significantly reducing classification errors compared to traditional methods. This research highlights the transformational potential of deep learning to improve maturity classification in the palm oil industry, by offering a more efficient, accurate and automated approach. Further research should focus on expanding the dataset to increase model robustness as well as exploring real-time implementation to further improve decision making in palm oil production. This approach promises to increase agricultural efficiency by ensuring optimal harvest timing and better resource management.