Classification of Local Fruit Types using Convolutional Neural Network Method (Study Case: Lombok Island)

Abstract

Indonesia, with its natural beauty and abundant resources, has significant potential for producing food and horticultural crops, particularly on Lombok Island, West Nusa Tenggara. This region is crucial in supplying tropical fruits such as Mangosteen, Pisang Kepok, and Rambutan Lebak Bulus. However, the agricultural sector in NTB faces challenges in post-harvest handling, especially in classifying fruit ripeness, impacting distribution and supply sustainability. To address this, researchers developed a fruit classification model using digital image processing with the Convolutional Neural Network (CNN) method. This model serves as a preliminary step before creating a fruit maturity classification model. Evaluation results showed that the RGB format model achieved 95% accuracy, while the HSV format reached 97%. Comparing three models in HSV format revealed: the proposed model (0.97), MobileNetV2 (0.96), and ResNet50 (0.97). These results indicate that implementing this model could enhance post-harvest efficiency in NTB, ensuring better fruit supply management.