Classifying Calamansi (Citrofortunella microcarpa) using Convolutional Neural Network

Abstract

Calamansi also known as the Philippine lime is among the top four agricultural produce of the Philippines that is being exported to other countries. Postharvest grading of Calamansi is an arduous task when manually done. Thus an efficient way of grading it according to size, color, maturity, and condition of the fruit is needed to help calamansi farmers, grade and classify them non-destructively with precision and accuracy. The goal of this study is to create a model for classifying Calamansi fruit. The study used ImageJ to measure the size of the fruit before it is run and trained using Convolutional Neural Network. Additionally, two datasets were used during the training of the model. The original dataset contains images with reference objects while the second dataset is a duplicate of the original dataset with the reference object removed from the images. The reference object used is a United States (US) quarter coin with a diameter of 2.1426 cm. The result revealed a 96.67% average accuracy using the original dataset that contains the reference object. Whereas, using the dataset without the reference object yielded a 70.24% accuracy rate. It shows that the size of the fruit could be best measured and produced the highest accuracy rate when a reference object is used. Based on the result of the study conducted, it could be established that a convolutional neural network could be used in the classification of Calamansi.