Certain investigations on implementations of convolutional neural network architecture for fundus image classification and fabric defects detection

Abstract

Artificial Intelligence (AI) belongs to the field of computer science. Recently there has been much advancement in artificial intelligence domain. Artificial intelligence has reduced the gap between man and machine. The main objective of artificial intelligence is to enable a machine to view its surrounding with a human perspective. Machine learning is a subsection of artificial intelligence that facilitates machines to perform a variety of tasks like natural language processing, image analytics and classification, signal processing, video recognition, and so on. Image classification is one such process that involves computers to assign a predefined label to the image based on information extracted from the image. Image classification finds its application in almost every field like quality inspection, disease diagnosis, face recognition, image, video recognition, etc. The evolution of convolutional neural network is an important innovation in the field of machine learning. Most of the popular social networks, to name a few, Twitter, Instagram, Facebook, etc., employ convolutional neural networks for learning their search information. Generally a classifier accepts features of the image as input attributes and trains the network to classify the images. On the other hand, a convolutional neural network accepts the image itself as the input and classifies the image based on the probability scores computed. A convolutional neural network can be designed according to the requirement. One of the noteworthy features of a convolutional neural network system is its capacity to handle large volume of data. A convolutional neural network is a special type of neural network that is formed by stacking several nonlinear layers one after the other. The input image is converted into semantic features and fed to successive layers and finally gets converted to class score. The objective of the present research is to design a convolutional neural network for applying it in problem domains like classification of fundus image according to the severity of diabetic retinopathy and classification of fabric images according to the defects newline