A deep convolutional neural network architecture for image classification
Year: 2020
Type: dissertation
Abstract: ENGLISH SUMMARY : Convolutional Neural Networks (CNNs), a specialised form of Neural Networks (NNs), are wellknown for their state-of-the-art results obtained in Computer Vision (CV) and Deep Learning (DL) tasks throughout the past few years. Some of the exciting application areas of CNNs include \nimage classi cation, object detection, video processing, natural language processing, and speech \nrecognition. \nThe powerful learning ability of deep CNNs is primarily owed to the use of multiple feature \nextraction stages that can automatically learn representations from the data. The availability of a \nlarge amount of data and improvement in hardware technology has accelerated the research done \nin CNNs, and recently interesting deep CNN architectures have been reported. Several inspiring \nideas to bring advancements in CNNs have been explored, such as the use of di erent activation \nand loss functions, parameter optimisation, regularisation, and architectural innovations by using \ndi erent layer structures. \nTherefore, the objective of this study is based on image classi cation and object detection \ntasks, that is, creating custom-designed CNN architectures for deployment on real-world datasets \nwhile comparing these custom-designed architectures to those state-of-the-art architectures found \nin literature while comparing di erent optimisation procedures and activations functions. All major \ndevelopments of CNNs are discussed and critically considered, with a view to improve CNNs in the \ncontext of the number of parameters used to obtain satisfactory results and additionally to obtain \na better understanding of the term known as a `black box' which is usually associated with CNNs \nsuch that they are complex models with little understanding in the way how their classi cations \nare done. \nThe most promising modern CNN architectures with associated hyperparameters are further \nexplored by means of empirical work. Evaluation is done on the validity of ndings reported in the \nliterature and comments are made on the e ectiveness of recent proposals through the use of ve \ndi erent real-world datasets. The empirical work done will be complemented by additional coded \nnotebooks that could be used to implement state-of-the-art techniques, as well as for comparative \nand model assessment experiments.
Author: Willem Lodewikus Pretorius
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W3135789880