Mango (Mangifera indica L.) Classification Using Convolutional Neural Network and Linear Classifiers

Abstract

Identifying fruit quality of a mango is a vital aspect for farmers and consumers; additionally, fruit classification is an imperative stage of fruit grading. Automation has been a boon in classification and grading of a mango (Mangifera indica L.). In this paper, we picked up various categories of mangoes such as Aafush, Kesar, Jamadar, Rajapuri, Totapuri, langdo and Dasheri. This set of mangoes was used for classification process which includes dataset preparation and feature extraction using pre-trained convolutional neural network (CNN) models. Four linear classifiers, namely support vector machine (SVM), logistic regression (LR), naïve Bayes (NB) and random forest (RF), are used for classification and compared. The paper also addresses techniques and issues of classification of mangoes on the basis of nondestructive method in particular advancement in deep learning and CNN. However, we have also discussed five CNN models, namely Inception v3, Xception, ResNet, DenseNet and MobileNet. Several experiments were carried out through these models, and the highest Rank-1 accuracy of 91.43% and lowest 22.86% accuracy was achieved. The model MobileNet is fastest, while DenseNet was found to be the slowest. In CNN models, Xception and MobileNet while in linear classifiers SVM and LR performed well.