An Image Dataset Construction for Flower Recognition Using Convolutional Neural Network

Abstract

Classifying flowers is a difficult activity because of the wide variety of flower species that have similar form. In this paper, a deep learning model for extracting features and classifying different flower types or species developed by using a popular method called Convolutional Neural Network (CNN). The identification system has been evaluated on a new dataset that has been designed in this work that collected flowers from Kurdistan. The dataset contains 1300 images of different flowers, 1040 images (%80) of which used for training purpose and 260 (%20) images used for test purpose, categorized into 26 classes. In addition, Various number of layers, activation function and pooling schemes were implemented to improve the classification rates. The recognition rate of proposed CNN model is 94.61%.