Rekonstruksi Superpiksel Graf Melalui Integrasi Graph Neural Networks Dan Strategi Gaussian Random Walk Untuk Klasifikasi Citra

Amaanullah, Fairuuz Nurdiaz (2026) Rekonstruksi Superpiksel Graf Melalui Integrasi Graph Neural Networks Dan Strategi Gaussian Random Walk Untuk Klasifikasi Citra. Masters thesis, Institut Teknologi Sepuluh Nopember.

Text 6002241007-Master_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (11MB) | Request a copy

Abstract

Implementasi superpiksel graf pada data citra telah banyak digunakan pada klasifikasi objek dikarenakan kemampuannya dalam merepresentasikan topologi objek citra. Model graph neural networks pada umumnya masih mempelajari superpiksel graf secara statis dan homogen. Penelitian Tesis ini membangun metode integrasi strategi Gaussian random walk dengan mekanisme rebound untuk membentuk informatif superpixel graf dan model graph neural networks untuk memprediksi kelas dari input citra yang dinamakan Adaptive Random Walk Graph Neural Networks atau ARW-GNNs. Secara umum, metode ARW-GNNs menyeleksi superpiksel node yang informatif dan diperoleh informatif superpiksel graf yang digunakan sebagai input graph neural networks di setiap iterasi secara dinamis. Graph neural networks yang digunakan adalah gated graph convolutional networks untuk mempelajari informasi spatio-temporal dari superpiksel graf untuk klasifikasi yang lebih baik. Metode ARW-GNNs diuji pada dataset MNIST dan CIFAR-10. Hasil uji diperoleh bahwa metode ARW-GNNs memprediksi kelas lebih akurat dibandingkan metode baseline. Metode ARW-GNNs juga memiliki performansi lebih efisien dibandingkan dengan metode baseline disebabkan pembentukan informatif superpiksel graf.
================================================================================================================================
Superpixel graph implementation on image data has been utilized to classify objects because of its ability to represent object topology. Graph neural networks generally learn a superpixel graph in static mode and homogenously. This Thesis builds an integration method between a Gaussian random walk strategy with rebound to form informative superpixel graphs and graph neural networks to predict the class of the input images, called Adaptive Random Walk Graph Neural Networks or ARW-GNNs. Generally, ARW-GNNs selects informative superpixel nodes and obtains informative superpixel graphs as the input graph neural networks in each iteration dynamically. In this Thesis, we employed gated graph convolutional networks to learn the spatio-temporal of superpixel graphs to have better classification. ARW-GNNs was testedon MNIST and CIFAR-10 datasets. The experimental results showed ARW- GNNs predicting more accurately than baseline methods. ARW-GNNs also had more efficient performances than baseline methods thanks to informative superpixel graph constructions.

Item Type:	Thesis (Masters)
Uncontrolled Keywords:	Superpixel Graf, Strategi random walk, Graph Neural Networks, Klasifikasi Citra Superpixel Graph, Random Walk Strategy, Graph Neural Networks, Image Classification
Subjects:	Q Science > Q Science (General) > Q325.5 Machine learning. Support vector machines. Q Science > QA Mathematics > QA166 Graph theory Q Science > QA Mathematics > QA336 Artificial Intelligence Q Science > QA Mathematics > QA76.87 Neural networks (Computer Science)
Divisions:	Faculty of Mathematics and Science > Mathematics > 44101-(S2) Master Thesis
Depositing User:	Fairuuz Nurdiaz Amaanullah
Date Deposited:	29 Jul 2026 00:42
Last Modified:	29 Jul 2026 00:42
URI:	http://repository.its.ac.id/id/eprint/138294

Actions (login required)

View Item