Watermelon Sorting Using Sound Waves: Based on Machine Learning

Abstract

The purpose of this study is to predict the ripeness of watermelon by the hitting sound. It was found that the reverberation time, the thickness of the skin, and the volume of the pulp were significantly correlated with ripeness. The hitting of watermelon can be explained as a damped oscillation, and this was verified through experiments and simulations, and correlation analysis between reverberation time and ripeness. As a result, the reverberation time was selected as an important variable. CNN was used to develop a model that predicts ripeness through the hitting sound of watermelon, which showed 99% accuracy. 4-layer perceptron model using the reverberation time which was selected as an important variable for predicting ripeness was developed. This showed 95.7% accuracy, in conclusion proposing an objective and non-destructive method to determine watermelon ripeness.