Convolutional Neural Network Models for Image Classification and Object Detection

Abstract

Computer vision is one of the most frontier and revolutionary fields of computer science. It aims at understanding and managing the enormous visual content available in the present day. Vision begins with eyes but really takes place in the brain; hence the concept and functioning of ANNs are studied, which enable the computers to understand images by examining its features. The study focuses on a category of deep neural networks typically used to deal with visual data, known as convolutional neural networks (CNNs). CNNs eliminate the need to hand-engineer the features of the object class, thus making the process more like a child learning to see and interpret things. The thesis presents the implementation of image classification and object detection which are two fundamentals to build any computer vision system. Certain well-known algorithms are applied, including Alexnet for classification and Faster R-CNNs and Mobilenet-SSD for detection. Their performances are analyzed and compared. The outcome of the study shows that Mobilene-SSD takes less detection time as compared to Faster R-CNNs, on the other hand, Faster R-CNNs can detect more object instances in an image, with higher class scores.