Developing deep learning classification models when the image data has several categories is a challenging task. It becomes even more challenging when the number of images available is limited. In such situations, it may be possible to take advantage of an existing model that has been developed with the help of a much larger dataset and reuse the patterns it has learned by customizing it for another classification task. This reuse of a pretrained deep network model for a new classification task is known as transfer learning.

The Keras library provides various pretrained models for image classification tasks that are trained using over a million images, and that capture reusable features that can be applied to similar but new data. Transferring what a pretrained model has learned from a large number of samples to a model that is being built with a much smaller sample size helps to save computational resources. In addition, use of the transfer learning approach can help to outperform a model that is built from scratch using a smaller dataset. In Chapter 7, Image Classification for Small Data Using Transfer Learning, we cover transfer learning and illustrate the utilization of a pre trained deep learning image classification model using R.