Repository logo
 

Banknote Serial Number Recognition Using Deep Learning

Files

Thesis(3.23 MB)

Date

2020

Authors

Supervisor

Yan, Wei Qi

Item type

Thesis

Degree name

Master of Computer and Information Sciences

Journal Title

Journal ISSN

Volume Title

Publisher

Auckland University of Technology

Abstract

Deep learning has been broadly applied to pattern classification, object detection, image segmentation, speech recognition, and other fields in recent years. Convolutional neural networks take dominant role in the field of deep learning, which has excellent characteristics that traditional machine learning algorithms cannot reach. The problem of character recognition has also been extensively studied in recent years, whose scope is much wide, including license plate recognition, handwriting recognition, bank check, and handwriting recognition for postcodes on envelop, etc. According to the current circulation banknote in New Zealand, this thesis applies deep learning to the character recognition of serial numbers on banknotes. The data samples used in this thesis are the images from the sixth edition of New Zealand banknote, which have been preprocessed with labelling, augmentation, scaling, and transformation, etc. The algorithms based on deep learning are proposed which have the stability for the serial number recognition in complex backgrounds. In this thesis, a pipeline of deep neural networks is constructed for character recognition of banknote serial numbers. Since high reliability is more important than accuracy in financial applications, DenseNet is proposed as the primary classifier, the scaling transformation of SegLink is employed to locate the characters, the detection rate is up to 95.80%. A convolutional neural network with residual attention model is proposed for serial number recognition, the precision reaches up to 97.09%.

Description

Keywords

Banknote, Deep Learning, Object detection, Attention Model

Source

DOI

Publisher's version

Rights statement

Permanent link

https://hdl.handle.net/10292/13337

Collections

Masters Theses