Adaptive Fuzzy Network based Transfer Learning for Image Classification

With the introduction of Convolutional Neural Networks (CNN) the computer vision domain has witnessed a tremendous increase in novel architectures achieving results on vision tasks that exceed human performance. Neuro-fuzzy hybrid systems are a great avenue for enhancing the interpretability of neural networks. A lot of research in recent times has explored the technique of transfer learning applied to CNNs for computer vision applications. However, a pre-trained deep convolutional network with a subsequent adaptive fuzzy based network is yet to be explored. Hence in this paper, a novel adaptive fuzzy network based convolutional network is proposed. The paper focuses on using non-hybrid learning based adaptive fuzzy networks in conjunction with pre-trained convolutional networks for the task of image classification. The results illustrate the proposed approach eclipses over existing architectures used for image classification.