Abstract:
This study presents a robust and economically efficient method for the discrimination of four MungBeans varieties on the basis of quantitative parameters, which is otherwise a challenging task due to their similar physical and morphological features, such as color, shape and size etc. Digital images of the bulk samples, used as input data, are acquired by using a digital camera, in an absolute natural environment without any complex laboratory arrangement. A total number of 230 first-order and second-order, sometimes known as Gray Level Co-occurrence Matrix (GLCM) textural parameters are extracted from different sizes of Regions of Interest (ROIs), in all radial and axial directions up to 5 pixel distance by using Mazda software. The most relevant 10 features are selected by Fisher Coefficient and classification/clustering capability of the selected features data is verified with the implementation of two multivariate approaches, Principal Component Analysis (PCA) and Linear Discriminant Analysis (LDA), whereas, a feed-forward Artificial Neural Network (ANN) classifier has been employed for training and testing purpose. The best results are achieved with an average accuracy of 98.17% and 94.35% during training and testing phase respectively, when the data of 10 selected features from ROI (64×64) is deployed to the classifier.
Page(s):
79-87
DOI:
DOI not available
Published:
Journal: Pakistan Journal of Engineering and Applied Sciences, Volume: 24, Issue: 1, Year: 2019