A study on identification of dog breeds through multi-class classification using Deep Learning techniques

No Thumbnail Available
Authors
Isac, Renji
Issue Date
2019
Degree
MSc in Data Analytics
Publisher
Dublin Business School
Rights
Items in eSource are protected by copyright. Previously published items are made available in accordance with the copyright policy of the publisher/copyright holder.
Abstract
Image classification, a discipline which over the years has made tremendous advancements with new and improved techniques continuously being implemented to improve the accuracy. With an enormous amount of effort being put into the field, multiclass classification has proven to be particularly challenging. Hence, in the present study the researcher focuses on achieving multiclass classification on dog breed identification using state of the art deep learning techniques. Having previously collaborated with NGO’s that managed dog adoptions, the researcher saw first-hand how wrongful breed classification affected the lives of these feeble beings by keeping them from getting the home they deserved. This in turn helped the researcher realize the urgency of such an identification system. Moreover, the researcher also focuses on a comparative study between deep learning and support vector machines which has proven to be particularly effective in binary classification. The results observed during the research inferred that ‘multiclass classification’ posed a problem even for deep learning however from the results observed from the comparative study, there was some light shed on how these problems could be tackled in the future. The results are presented.