Putra, Daniswara Aditya (2025) Klasifikasi Spesies Burung Berdasarkan Suara Menggunakan Deep Learning Dengan Fitur Mel-Spektogram Dan Mel Frequency Ceptral Coeffiicients (MFCC). Other thesis, Institut Teknologi Sepuluh Nopember.
![]() |
Text
5003211132-Undergraduate _Thesis.pdf Restricted to Repository staff only Download (4MB) | Request a copy |
Abstract
Burung merupakan indikator penting dalam menilai kesehatan ekosistem dan biodiversitas suatu wilayah. Perkembangan teknologi pengenalan suara berbasis deep learning memungkinkan pemantauan spesies burung dilakukan secara otomatis melalui data audio, tanpa perlu pengamatan langsung yang berisiko mengganggu habitat alami. Pada penelitian ini dilakukan klasifikasi suara enam spesies burung yang umum ditemukan di Indonesia, yaitu Kutilang, Gereja, Perkutut, Tekukur, Trucukan, dan Cendet, menggunakan lima arsitektur model deep learning, yaitu CNN (Mel-Spektrogram), CNN (MFCC), MobileNet (Mel-Spektrogram), MobileNet (MFCC), dan VGGish (waveform). Data yang digunakan merupakan rekaman suara dari situs xeno-canto.org yang telah melalui proses ekstraksi fitur. Model dibangun menggunakan framework TensorFlow dan Keras, kemudian dilatih dengan kombinasi hyperparameter yang bervariasi untuk memperoleh performa terbaik. Berdasarkan hasil evaluasi menggunakan metrik akurasi, precision, recall, dan F1-score pada data test, model CNN dengan fitur MFCC menunjukkan performa terbaik dengan akurasi mencapai 97,97%, precision 97,78%, recall 98,27%, dan F1-score 97,96%. Model ini tidak hanya mampu mengenali pola suara burung secara akurat, tetapi juga menunjukkan kemampuan generalisasi yang sangat baik terhadap data baru. Hasil penelitian ini menegaskan bahwa CNN dengan fitur MFCC merupakan pendekatan yang efektif dan andal untuk tugas klasifikasi suara burung, serta memiliki potensi besar untuk mendukung upaya konservasi berbasis teknologi di Indonesia.
=============================================================================================================================================
Birds are important indicators for assessing the health of ecosystems and biodiversity in a region. Advances in deep learning-based sound recognition technology enable automatic monitoring of bird species through audio data, eliminating the need for direct observation that may disturb natural habitats. This study focuses on classifying the calls of six common bird species found in Indonesia—Kutilang, Gereja, Perkutut, Tekukur, Trucukan, and Cendet—using five deep learning architectures: CNN (Mel-Spectrogram), CNN (MFCC), MobileNet (Mel-Spectrogram), MobileNet (MFCC), and VGGish (waveform). The dataset consists of bird call recordings sourced from xeno-canto.org and processed through feature extraction. Models were developed using the TensorFlow and Keras frameworks and trained with various hyperparameter combinations to achieve optimal performance. Based on evaluation metrics including accuracy, precision, recall, and F1-score on the test data, the CNN model with MFCC features demonstrated the best performance, achieving an accuracy of 97.97%, precision of 97.78%, recall of 98.27%, and F1-score of 97.96%. This model not only accurately recognizes bird call patterns but also exhibits excellent generalization capability to unseen data. The results confirm that CNN with MFCC features is an effective and reliable approach for bird sound classification, with strong potential to support technology-based conservation efforts in Indonesia.
Item Type: | Thesis (Other) |
---|---|
Uncontrolled Keywords: | Burung, CNN, Mel-Spektogram, MFCC, MobileNet, VGGish, Bird, CNN, Mel-Spektogram, MFCC, MobileNet, VGGish |
Subjects: | Q Science > Q Science (General) Q Science > QA Mathematics > QA76.87 Neural networks (Computer Science) |
Divisions: | Faculty of Science and Data Analytics (SCIENTICS) > Statistics > 49201-(S1) Undergraduate Thesis |
Depositing User: | Daniswara Aditya Putra |
Date Deposited: | 01 Aug 2025 07:11 |
Last Modified: | 01 Aug 2025 07:11 |
URI: | http://repository.its.ac.id/id/eprint/125794 |
Actions (login required)
![]() |
View Item |