Image Classification Using Few Shot Learning

The requisite for a large number of training images for each class is a burden to Deep Learning (DL) research. As a result, Few Shot Learning (FSL) techniques have been developed that require only a few number of samples of a class to classify a query sample of that class. In this paper, ResNet-50 and Prototypical Networks (ProNet)-based FSL approach is proposed to classify images. The proposed approach uses ResNet-50 Convolutional Neural Networks (CNN) model for feature extraction and the ProNet for classifying query images. The ProNet computes a mean called prototype of all the support images of a class, which is then used to classify a query image by comparing it with the Euclidean distance. The Plant Village (PV) dataset is used to analyze the results of the proposed approach. The dataset is split into a source domain for training the technique and a target domain for testing its performance. The proposed model is trained episodically on 4000 training tasks randomly generated from the images of source domain. The target domain is split into four sets, for N = 5 (5 classes randomly selected from target domain) and K = 10 (for each of the 5 classes 10 samples are selected), the approach delivered 70.24% accuracy on split-1, 89, 84.5, and 92.26% accuracies on split-2, 3, and 4, respectively. In comparison, the proposed approach delivered better result.