Classification of Apricot Leaves with Extreme Learning Machines Using Deep Features

Abstract

Machine learning and image processing-based classification of automated plant species is significant for plant experts/herbalists. Many studies on the subject have been gained to the literature. Today, researchers have applied deep learning to various image-based object recognition tasks. In this study, the classification of automatic apricot species based on Deep Convolutional Neural Networks (DCNN) has been made. The proposed method used the VGG19 model, a pre-trained DCNN model. Seven different feature vectors were obtained by combining the features obtained from three different fully connected layers in different combinations. These feature vectors were given to the input of Excessive Learning Machines and seven different apricot types were classified. The highest performance rate was obtained from the fc8 layer as 98.8%, and the lowest performance rate was obtained from the feature vector obtained from the combination of fc6 and fc7 layers as 95.2%. © 2019 IEEE.