Using Deep Neural Nets for Flower Classification

Abstract

With the growing influence of Machine Learning, Image classification is a topic of interest for many. Flower Classification remains to be a hurdle which is yet to be crossed in this topic. The nature of flower image datasets, with indifferentiable backgrounds, high inter-class commonalities and intra-class dissimilarities makes the process complex and demanding. Our research employs the transfer learning approach by training a pre-trained ImageNet image classifier, VGG-16 using TPU. The TPU, due to its cores, speeds up the process of training and is a part of various large-scale classifiers. The epoch size and learning rate have been chosen to favour the process. The results achieved were gratifying with an accuracy rate of 94.1%. This work holds a scope in the future which include automation using an application. Our study represents a significant step forward in flower classification using machine learning, with implications for fields such as agriculture, botany, and ecology.