Multi-class classification using deep learning

• Main Author: Cao, Haozhi
• Other Authors: Mao Kezhi
• Format: Final Year Project
• Language: English
• Published: 2019
• Subjects: DRNTU::Engineering::Electrical and electronic engineering
• Online Access: http://hdl.handle.net/10356/78474
• Institution: Nanyang Technological University

Multi-class classification is the classification task where separates samples into more than 2 classes. An image multi-class classifier is a mathematic model which can distinguish the category of pictures. One of the traditional models of image classifier is Convolutional Neural Network (CNN). However, the fully-connected layers of CNN usually contains significant number of parameters abasing the performance of CNN. As a result, in order to elevate the performance of CNN, it is necessary to reduce the parameters of fully-connected layers. In this project, inspired by previous improvement of Feedforward Neural Network, a theoretical CNN model with a binary decode output layer is proposed. To evaluate the accuracy as well as efficiency of this possible method, three different classification tasks are conducted and the test accuracy, training accuracy and training time are recorded independently. After analyzing the results above, it shows that binary decode approach can increase the test accuracy of CNN model and slightly accelerate the training process under some restricted conditions.