Transfer Learning for Leaf Classification with Convolutional Neural Networks

Abstract

Convolutional Neural Network (CNN) is taking a big role in image classification. B ut f ully t raining i mages by using CNN takes a plenty of time and uses a very large data set. This paper will focus on transfer learning, a technique that takes a pre-trained model e.g., Inception, Resnet or MobileNets models then retrains the model from the existing weights for a new classification p roblem. T he r etrain t echnique drastically decreases time spending in the training process and many fewer number of image data is required to yield high accuracy trained networks. This paper considers the problem of leaf image classification t hat t he e xisting a pproaches t ake m uch e ffort to choose various types of imagefeatures for classification. This also reflects p utting b iases b y c hoosing s ome f eatures a nd ignoring the other information in images. This paper will conduct the experiments in accuracy comparison between traditional leaf image classification using image processing techniques and CNN with transfer learning. The result will show that without much knowledge in image processing, the leaf image classification can be achieved with high accuracy using the transfer learning technique.