Design and Development of Mango Ripeness Classification Tool using CNN Android-based Platform

Abstract

Artificial ripening methods use calcium carbide (carbide) which often leaves harmful residues on the mango fruit. This research designs a classification tool for carbite and non-carbite mango fruit using the Android-based InceptionV3 Convolutional Neural Network method. The mango fruit image dataset consists of 1622 images (881 images of carbite mangoes and 811 images of non-carbite mangoes) used to train and test the model. The testing process is done by implementing the model on a Raspberry Pi B+ connected to a camera pi to take pictures of mangoes at a distance of 30 cm. The results showed that the CNN model developed achieved an average accuracy of 94.4% in classifying carbitan and non-carbitan mangoes. This result shows that the classification tool designed can provide significant benefits for farmers, traders, and consumers in ensuring marketed quality.