A Deep Learning-Based Model for Classifying Sweetness Level of Sky Rocket Melon: A Preliminary Result

Frequently consumed for its tasty, crunchy flesh and distinct aroma, Sky Rocket Melon is a widely grown tropical horticultural crop and is considered a commodity of great economic potential in Indonesia. It is rich with sought-after nutrients like Vitamin C. Typically, the melon’s ripeness is assessed by directly inspecting its visual appearance, its aroma, and the sound produced while tapped. This method is naturally subjective and inconsistent due to disparities between individual assessors. Thus, this work aims to develop a novel, and more objective, method for classifying the sweetness of Sky Rocket melon using deep learning. The classes are divided into sweet and non-sweet melons. 1800 collected datasets are divided into training data, validation data, and testing data with a ratio of 60:20:20, respectively. Combining with different parameters, three architectures of Convolutional Neural Network (CNN): EfficientNet B4, ResNet-50, and ShuffleNet V2, are trained and evaluated. The result shows that the ShuffleNet V2 model outperformed other models, with an accuracy of 88.06%.