Rapid Prediction of Melon Sweetness Using Image Processing Techniques and Algorithmic Models

Abstract

Rapid prediction of melon sweetness using image processing techniques and algorithmic models is studied. The sweetness of melon is usually greater than 12.5 Brix, which can be measured with a digital sweet meter where most consumers do not know such a problem. This work proposed an image processing technology combined with an algorithmic model to determine the sweetness of the melon. The melon sweetness measurement method collects 100 samples of full-leaf melon photographs from 4 sides, top, bottom, left and right, consisting of weight, circumference, stalk size, and small circle axis size. The size of the core to the edge of the large circle net is height, light intensity, and relative humidity, color temperature and intensity using image processing techniques. All of this data is learned using Random Forest (RF) and eXtreme Gradient Boosting (XGB) techniques, one of the machine learning algorithms used to predict melon sweetness. The test results showed that the R-squared value of model learning in the XGB algorithm was 99%, while that of the RF algorithm was 95%. It could be concluded that the XGB algorithm was effective in model learning better than the RF algorithm, with the weight of the melon being the most prominent. The melon color value was the least significant in interpreting the melon sweetness value. For the actual test, a model of the XGB algorithm was used to predict the sweetness value of 50 melons.  The prediction accuracy is up to 95%, utilized for checking melon results.