Non-Destructive Mango (Mangifera Indica L., CV. Kesar) Grading Using Convolutional Neural Network and Support Vector Machine

Abstract

Automation in grading of mango (Mangifera Indica L., cv. Kesar) is important to reach consumer demand for quality mango. This paper addresses this issue of mango grading. In this paper, pre-trained Convolutional Neural Network (CNN) is used as feature extractor and Support Vector Machine (SVM) is used as classifier. Mango grading is performed by considering three parameters namely shape, size and maturity. Two approaches are used for feature extraction using CNN. In first approach, CNN is trained using mango samples labelled as class I, class II, class III and class IV. While in second approach, mango grading is performed in three phases. In first phase CNN is trained for shape parameter using deformed and well-formed labels; in second for size parameter using small, medium and big labels; and finally, for maturity parameter using ripe, partially ripe and unripe labels. Based on these three phases, decision of grading is taken. Four CNN architecture models namely Inception v4, Xception, ResNet and MobileNet are compared and used for experiment. In both the approaches, MobileNet performs excellent with highest accuracy and fastest execution time while ResNet performs poor in both approaches for accuracy as well as for execution. Inception and Xception both performs almost same.