An Improved Image Classification Based In Feature Extraction From Convolutional Neural Network: Application To Flower Classification

Abstract

Nowadays, deep learning techniques are increasingly growing in machine vision for object recognition, segmentation, classification, and so on, in a wide variety of applications. In this study, we apply the convolutional neural network (CNN) to flower classification. For this purpose, we firstly increase the data with the augmentation techniques and use them in the pre-trained CNN models in which classification part is removed and instead of it, we use global average pooling (GAP) in the last layer for extracting their features. The features obtained from these models are concatenated, and then we use a support vector machine (SVM) as classifier for the flower classification. We use the Oxford 102 flower and the Oxford 17 flower datasets in our experiments. By applying this method, we achieve 96.47% classification accuracy for the Oxford 102 flower and 97.64% classification accuracy for the Oxford 17 flower. The results show the effectiveness of the proposed strategy and perform more accurate classification than the traditional methods.