Pengenalan Notasi Musik Menggunakan Convolutional Neural Network (CNN)

Mashuda, Arij Nafi'atul (2019) Pengenalan Notasi Musik Menggunakan Convolutional Neural Network (CNN). Other thesis, Institut Teknologi Sepuluh Nopember.

Text 05111540000013-Undergraduate_Thesis.pdf Restricted to Repository staff only Download (2MB) | Request a copy

Abstract

Optical Music Recognition (OMR) adalah salah satu bidang penelitian tentang pengenalan notasi musik secara komputasional. Secara garis besar langkah-langkah pengenalan objek pada OMR hampir sama dengan Optical Character Recognition (OCR). Namun, pengenalan notasi musik lebih kompleks karena notasi musik memilik simbol yang lebih beragam. Di dalam tugas akhir ini, diimplementasikan pengenalan notasi musik menggunakan Convolutional Neural Network (CNN). Data pelatihan dan uji coba diambil dari dataset “DeepScores” yang merupakan gambar partitur dengan kualitas tinggi yang terdiri dari bentuk dan ukuran simbol yang berbeda. Percobaan dilakukan pada dua arsitektur, yaitu AlexNet dan VGGNet16, kemudian diuji menggunakan beberapa optimizer dan learning rate. Dari percobaan pada kedua arsitektur, Alexnet dan VGGNet16 mendapatkan hasil akurasi tertinggi dengan menggunakan optimizer RMSProp dan learning rate 0,0001, yaitu sebesar 100%.
====================================================================================================================================
Optical Music Recognition (OMR) is one of the fields of research on computational notation music recognition. The steps to introduce objects on OMR are almost the same as Optical Character Recognition (OCR). However, the introduction of music notation is more complicated because music notation has more diverse symbols. In this undergraduate thesis, the authors implement an algorithm to Optical Music Recognition (OMR) using the Convolutional Neural Network (CNN). Training data and testing data are taken from the "DeepScores" dataset. It is a high-quality music picture consisting of different shapes and sizes of symbols. From experimenting on two architectures, AlexNet and VGGNet16, then used some optimizers and learning rates, it turned out that the highest accuracy is the experiment using RMSProp optimizer and learning rate 0,0001, it was 100%.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Optical Music Recognition, Pengenalan Simbol Musik, image classification, Convolutional Neural Networks, Deep Learning, dataset DeepScores
Subjects:	T Technology > T Technology (General) > T58.5 Information technology. IT--Auditing
Divisions:	Faculty of Information and Communication Technology > Informatics > 55201-(S1) Undergraduate Thesis
Depositing User:	Arij Nafi'atul Mashuda
Date Deposited:	23 Jul 2026 03:22
Last Modified:	23 Jul 2026 03:22
URI:	http://repository.its.ac.id/id/eprint/67840

Actions (login required)

View Item