Flower Species Detection Over Large Dataset Using Convolution Neural Networks and Image Processing Techniques

Abstract

We investigate flower species detection on a large number of classes. In this paper, we try to classify flower species using 102 flower species dataset offered by Oxford. Modern search engines provide methods to visually search for a query image that contains a flower, but it lacks robustness because of the intra-class variation among millions of flower species around the world. So, we use a Deep learning approach using Convolutional Neural Networks (CNN) to recognize flower species with high accuracy. We use the Oxford dataset which was made by the use of electronic items like a built-in camera in mobile phones and also a digital camera. Feature extraction of flower images is performed using a Transfer Learning approach (i.e. extraction of complex features from a pre-trained network). We also use image augmentation and image processing techniques to extract the flower images more efficiently. After the experimental analysis and using different pre-trained models, we achieve an accuracy of 85%. Further advancements can be made by using optimization parameters in the neural nets.