Transfer Learning for Image Classification

Convolutional neural network (CNN) gained great attention for robust feature extraction and information mining. CNN had been used for variety of applications such as object recognition, image super-resolution, semantic segmentation etc. due to its robust feature extraction and learning mechanism. By keeping constant the baseline learning topology, various CNN architectures were proposed to improve the respective system performance. Among these, AlexNet, VGG16 and VGG19 are the famous CNN architecture introduced for object recognition task. In this paper, we make use of transfer learning to fine-tune the pre-trained network (VGG19) parameters for image classification task. Further, performance of the VGG 19 architecture is compared with AlexNet and VGG16. Along with the CNN architectures, we have compared the hybrid learning approach which is comprised of robust feature extraction from CNN architecture followed by support vector machine (SVM) classifier. We have used two state-of-the-art databases namely: GHIM10K and CalTech256 to study the effect of CNN architecture for robust feature extraction. Performance evaluation has been carried out using average recall, precision and F-score. Performance analysis shows that fine-tuned VGG19 architecture outperforms the other CNN and hybrid learning approach for image classification task.