Classification of Javanese Eagle Tweet Based on Improved Mel-Frequency Cepstral Coefficients and Deep Convolutional Neural Network

Permana, Silvester Dian Handy (2024) Classification of Javanese Eagle Tweet Based on Improved Mel-Frequency Cepstral Coefficients and Deep Convolutional Neural Network. Doctoral thesis, Asia e University.

	Text Thesis Silvester Dian Handy Permana.pdf Restricted to Repository staff only Download (9MB)
	Text Thesis Silvester Dian Handy Permana-1-24.pdf Download (707kB)

Abstract

The Javanese Eagle is a rare and protected animal in Indonesia, threatened with extinction due to its limited population. Conservation efforts in zoos and nature reserves are essential to prevent their extinction. One critical aspect of conserving the Javanese Eagle is understanding their communication through tweets, which can provide insights into their needs and behaviours. This study addresses the problem of effectively classifying the Javanese Eagle's vocalizations to aid in their conservation. The primary technique involves the use of Improved Mel Frequency Cepstral Coefficients (IMFCC) and Deep Convolutional Neural Networks (DCNN), combined to create a robust classification system. Data were collected from zoos and nature reserves in Indonesia, used to train and test the models, and then validated by experts. Experts validate after the best model is obtained and use new data to test its validity. The classification system aimed to distinguish between tweets indicating lack of food or drink, normal tweets, and those related to finding a partner. The study compared various CNN architectures, including AlexNet and VGGNet, and different combinations of training, validation, and test data. The best-performing model, VGGNet, was trained with a dataset split into 80% training, 10% validation, and 10% testing. During training, the VGGNet model achieved a peak accuracy of 100%, and during testing, it attained an accuracy of 99%. The Receiver Operating Characteristic (ROC) Curve analysis showed that the 'Normal' category had an area under the curve of 0.996, the 'Looking for Partner' category had an area under the curve of 1.000, and the 'Looking for Food' category had an area under the curve of 0.996. These results demonstrate the effectiveness of the proposed classification system in accurately identifying the Javanese Eagle's primary needs. The significance of this study lies in its potential to enhance conservation efforts by providing a reliable tool for monitoring the Javanese Eagle's well-being. By accurately classifying their vocalizations, conservation site managers can better understand and address the eagles' needs, improving their chances of survival and preventing extinction. This research also contributes to the broader field of bioacoustics and wildlife conservation, offering a methodology that can be adapted for other endangered species.

Item Type:	Thesis (Doctoral)
Uncontrolled Keywords:	Improved MFCC, deep convolutional neural network, Javanese eagle sound, sound classification
Divisions:	School of Graduate Studies
Depositing User:	Siti Nor Fairuz Rosaidee
Date Deposited:	07 Jan 2025 02:09
Last Modified:	07 Jan 2025 02:09
URI:	http://ur.aeu.edu.my/id/eprint/1282

Actions (login required)

View Item