Implementation of Deep Learning for Orange Variety Classification Using the Convolutional Neural Network (CNN) Method Based on Fruit Images

Abstract

This research aims to implement the Deep Learning method in classifying visually similar types of oranges using the Convolutional Neural Network (CNN) algorithm, as well as to determine the model’s accuracy in recognizing orange fruit images. The study classifies four types of oranges: Keprok Brastepu, Keprok Maga, Murcott, and Siam Madu. The dataset used consists of 1.739 initial dataset, which increased to 2.304 images after augmentation and oversampling, with a data split ratio of 80% for training and 20% for testing. The pre-processing stages include data labeling, data organization, remove background, and image resizing to match the input requirements of the Convolutional Neural Network model. The model employed is the InceptionV3 architecture with the Adam optimizer, categorical crossentropy as the loss function, 50 epochs, and early stopping applied at the 20th epoch when accuracy reached 98%. The experimental results show that the model with a learning rate of 0,0001 achieved the best performance, with training accuracy of 0.98 and validation accuracy of 0.98, along with precision of 0.98, recall of 0.97, and an f1-score of 0.97. Meanwhile, a higher learning rate (0,01) resulted in signs of overfitting. Based on these findings, it can be concluded that the Convolutional Neural Network method with the InceptionV3 architecture is capable of classifying visually identical orange varieties with excellent accuracy, making it effective for implementation in image-based fruit identification systems.