A Computer Vision Approach to Classify Local Flower using Convolutional Neural Network

Abstract

Flower is the most beautiful part of this earth. In our busy lives, many flowers can be seen all over the places. Till now, more than 352,000 flower species in the world. In our country Bangladesh, the total numbers of species are not too much and are getting away from this natural beauty and becoming addressed with city life. Most of us are even unable to tell more than 10 names of local flowers. The problem is addressed and proposed an approach to identify the local flower of Bangladesh. Our proposed approach will be valuable to a botanist as well as people of other fields. With the support of machine learning techniques, object identification from an image is now quite encouraging with some challenges. Recent research has been focused on CNN (Convolutional neural network) model to train a machine with a large dataset to get more accurate results. A model is proposed, where CNN has used to classify the local flower dataset. The "ReLu" acti vation function "Adam optimizer" and the "Softmax" function are used to build the network layer. Our experiments are conducted on eight types of local flowers and considered a total of 5120 training images and 1280 test images to present eight types of flower categories and then applied eight augmentation methods to increase data volume. Finally, our proposed CNN structure provided 85% classification accuracy.