KLASIFIKASI FINGER SPELLING AMERICAN SIGN LANGUAGE MENGGUNAKAN CONVOLUTIONAL NEURAL NETWORK

MARJUSALINAH, ANNA DWI and Samsuryadi, Samsuryadi (2021) KLASIFIKASI FINGER SPELLING AMERICAN SIGN LANGUAGE MENGGUNAKAN CONVOLUTIONAL NEURAL NETWORK. Master thesis, Sriwijaya University.

	Text RAMA_55101_09042621721005.pdf - Accepted Version Restricted to Repository staff only Available under License Creative Commons Public Domain Dedication. Download (113MB) | Request a copy
Preview	Text RAMA_55101_09042621721005_0004027101_01_front_ref.pdf - Accepted Version Available under License Creative Commons Public Domain Dedication. Download (113MB) | Preview
	Text RAMA_55101_09042621721005_0004027101_02.pdf - Accepted Version Restricted to Repository staff only Available under License Creative Commons Public Domain Dedication. Download (83MB) | Request a copy
	Text RAMA_55101_09042621721005_0004027101_03.pdf - Accepted Version Restricted to Repository staff only Available under License Creative Commons Public Domain Dedication. Download (83MB) | Request a copy
	Text RAMA_55101_09042621721005_0004027101_04.pdf - Accepted Version Restricted to Repository staff only Available under License Creative Commons Public Domain Dedication. Download (83MB) | Request a copy
	Text RAMA_55101_09042621721005_0004027101_05.pdf - Accepted Version Restricted to Repository staff only Available under License Creative Commons Public Domain Dedication. Download (83MB) | Request a copy
	Text RAMA_55101_09042621721005_0004027101_06_ref.pdf - Bibliography Restricted to Repository staff only Available under License Creative Commons Public Domain Dedication. Download (83MB) | Request a copy
	Text RAMA_55101_09042621721005_0004027101_07_lamp.pdf - Accepted Version Restricted to Repository staff only Available under License Creative Commons Public Domain Dedication. Download (83MB) | Request a copy
	Text RAMA_55101_09042621721005_TURNITIN.pdf - Accepted Version Restricted to Repository staff only Available under License Creative Commons Public Domain Dedication. Download (29MB) | Request a copy

Abstract

Sign language is a combination of complex hand gestures, body postures, and facial expressions. However, only a small number of people can understand and use it. Automatic finger spelling sign language recognition using a convolutional neural network (CNN) is proposed in this study. The CNN architectures i.e., Resnet50 and DenseNet121, were compared to classify a collection of American Sign Language images. This study examines the effect of the ratio of training and testing data on the classification results. In addition, several learning rate parameter values were tested to get the best model. The test results show that the Resnet50 architecture with a learning rate of 0.01 and data separation of 90:10 for training and testing data shows the best performance with a sensitivity value of 0.98814, precision 0.98766, specificity 0.99947, f1-score 0.99897, error 0.00103 and accuracy 0.99897. Furthermore, this best model was tested with unseen data and resulted in a performance of 0.86619 for sensitivity, 0.68509 precision, 0.98641 specificity, 0.97375 f1-score, 0.97375 and 0.02625 for error.

Item Type:	Thesis (Master)
Uncontrolled Keywords:	American Sign Language, Finger Spelling, Resnet50, DenseNet121
Subjects:	Q Science > Q Science (General) > Q334-342 Computer science. Artificial intelligence. Algorithms. Robotics. Automation.
Divisions:	09-Faculty of Computer Science > 55101-Informatics (S2)
Depositing User:	Mrs Anna Dwi Marjusalinah
Date Deposited:	26 Jan 2022 04:27
Last Modified:	26 Jan 2022 04:27
URI:	http://repository.unsri.ac.id/id/eprint/62828

Actions (login required)

View Item