AnimNet: An Animal Classification Network using Deep Learning

Summary

Image classification is a combination of technologies: Image Processing (IP), Machine Learning (ML), and Computer Vision (CV). The classification of animals has been done in this work that are commonly found in the Indian scenario using two approaches: transfer learning and a custom‐built classification network, i.e., AnimNet. For transfer learning, we have used VGG16, VGG19, and Xception network that are existing pre‐trained networks and compared the results of the custom‐built AnimNet network with these existing networks. The comparison was done on the basis of the accuracy and size of the models as the size of network is as important as the accuracy of network in this era of mobile computing. A lightweight network with good performance is the most optimal choice nowadays. The accuracy was observed to be highest for the Xception network whereas the AnimNet network is lightweight, i.e., 5X smaller than the Xception model with second‐highest accuracy.