A Fuzzy Learning Algorithm for Harumanis Maturity Classification

Abstract

This paper presents an approach to classify the fruit maturity of fruit grading system when a higher level of accuracy and easy interpretability of the estimate model is desired. The proposed method automatically generates mem- bership functions (MFs) and constructs the associated fuzzy rules (FRs). The proposed approach is applied to a case study of Harumanis mango fruit grading system. The task is to classify the fruit maturity and grade using agronomic image data set acquired by digital camera. The parameters of the MFs are adjusted by the learning algorithm for the training data. This MF is then used to generate the MFbox using hyperbox-type fuzzy partition of feature space to generate FRs from training instance to deal with the features data. FRs are extracted from the flexible MFs and MFbox. As a case study, the proposed method is applied to Harumanis data set with 108 instances (input–output pairs), two real-valued inputs, and one output. Analysis results show that the proposed maturity classifier yields an accu- racy of 98 %. The developed maturity classifier can act as an instrument in determining the correct mango fruit maturity category.