Abstract

Rambutan fruit (Nephelium Lappaceum), a rare tropical fruit cherished for its unique appearance and flavor, has drawn attention for its limited cultivation. The variable sweetness levels of rambutan fruits, stemming from their limited cultivation, often lead to customer dissatisfaction and economic losses for growers, distributors, and stakeholders in the industry. The need for an automated and dependable system for classification and sweetness prediction is evident. To cope with this problem, the project’s goal is to create a consistent deep learning system that uses a variety of features such as color, and size to classify the rambutan fruit into 4 different varieties namely Binjai, E35, N18, and Pulasan and predict the sweetness levels of rambutan fruits into three categories namely sweet, mild sweet and sour accurately. The principal achievement of this study involves the development of a proficient deep-learning model based on the VGG16 architecture. The VGG16 architecture, characterized by its 16-layered Convolutional Neural Network design, demonstrates precision in categorizing rambutan fruits into predetermined varieties and sweetness levels. This model aims to bring about consistency in sweetness levels, ultimately resulting in increased consumer satisfaction and economic benefits for growers and distributors. The implementation included dataset creation of a total of 912 rambutan fruit images, preprocessing, VGG16 model building, training the dataset, and testing them. Hence the results obtained had an accuracy of 88% for classification based on variety and 85% for sweetness classification with 50 epochs of training. This can further be increased with a greater image sample in the dataset and the implementation of diverse preprocessing techniques. In the future, comparative analysis can be done with other deep learning algorithms.