Maturity level prediction and classification of lemon fruit (Citrus limon cv. Montaji Agrihorti) using combined reflectance fluorescence computer vision and machine learning models

Abstract

Montaji Agrihorti is an Indonesia lemon variety known for minimal seeds and abundant fruit juice. Conventional methods for estimating fruit maturity rely on physical observation and flowering time calculations, which can be subjective and inaccurate. Computer vision technology offers a non-destructive approach to determining fruit maturity. The purpose of this research was to build a prediction model of physicochemical parameters on lemon fruit maturity based on reflectance and fluorescence digital image analysis using color and texture features. The experimental laboratory method was carried out in two stages, namely destructive tests and non-destructive tests. Destructive tests were conducted by measuring fruit firmness, total soluble solids (brix) and total acidity. The non-destructive test was conducted by taking images of the fruit using fluorescence reflectance-based computer vision. The combination of fluorescence reflectance light sources using color and texture features in image capture can predict the level of fruit maturity. The data that has been obtained is then processed and processed using machine learning algorithm methods including K-NN (K-Nearest Neighbor), SVM (Support Vector Machine), and Random Forest. The best machine learning algorithm modeling based on visual lemon fruit recommendation on Random Forest model RGB image data with min-max scaling obtained a training test result of 1.00 and a test accuracy of 0.884.