Image Classification using AlexNet with SVM Classifier and Transfer Learning
Abstract
Since the last decade, Deep Learning has paved the direction for demanding and wide-ranging applications in almost everyday life and attained quite good performance on a diversity of complex problems such as face recognition, motion detection, health informatics and many more. This paper discusses the performance of state-of-the-art pre-trained Convolutional Neural Networks (CNNs) model. We have analyzed the accuracy and training time of AlexNet (already trained on millions of images) while using it with Support Vector Machines (SVM) classifier and under Transfer Learning (TL). Since training CNN from scratch is time-consuming in case of real problems and is compute-intensive so the image features extraction using already trained AlexNet network reduces effort and computation resources. The learned features are used to train a supervised SVM classifier and transfer learning (fine-tuning) of layers is achieved by retraining in the last few layers of pretrained network for image classification. The classification accuracy while Transfer Learning is obtained greater and more stable than SVM classifier.
Copyright (c) 2019 Journal of Information Communication Technologies and Robotic Applications
This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License.