Intelligent Classification Model for Cashew Kernel Grades Based on Colour, Texture, and Morphological Features

Abstract

In this paper, the method for building a supervised intelligent classification model for white wholes (WW) grades of cashew kernel using different images was discussed. The morphological, colour, and texture features were used to train or test different classifiers for recognition and classification. In order to achieve the best prediction accuracy, the subsets of features from feature sets were selected using a correlation-based feature selection (CFS) algorithm. In this study, the best prediction accuracy was obtained using multilayer perceptron, simple logistic, support vector machines, sequential minimal optimization, and logistic classifiers. The percentage of classification models that were correctly classified for the training/test set of WW grades ranged from 70% to 90%, and the validation set was as high as 86%. Receiver operating characteristics (ROC) were used to present the studied classifier performances in recognition and classification of WW grade cashew kernels