ANALISIS CLUSTERING POLUSI CAHAYA DI OBSERVATORIUM DUNIA BERBASIS DATA CITRA DENGAN DEEP LEARNING

M. Cahyana Bintang Fajar, - and Lala Septem Riza, - and Judhistira Aria Utama, - (2025) ANALISIS CLUSTERING POLUSI CAHAYA DI OBSERVATORIUM DUNIA BERBASIS DATA CITRA DENGAN DEEP LEARNING. S1 thesis, Universitas Pendidikan Indonesia.

Abstract

Observatorium merupakan situs penting dalam pengembangan ilmu astronomi, sehingga kelestarian langit malam di sekitarnya menjadi aspek yang perlu dijaga. Namun, pertumbuhan penduduk, ekspansi wilayah perkotaan, dan peningkatan aktivitas ekonomi telah memicu peningkatan kecerahan langit malam yang berujung pada polusi cahaya. Penelitian ini bertujuan untuk mengidentifikasi pola dan karakteristik kelompok (cluster) polusi cahaya di sekitar observatorium dunia melalui pendekatan berbasis deep learning dan analisis clustering. Data yang digunakan berupa citra satelit Visible Infrared Imaging Radiometer Suite-Day/Night Band (VIIRS-DNB) dari tahun 2012 hingga 2024 dalam radius 50 km dari masing-masing observatorium. Ekstraksi fitur spasial dari data citra dilakukan menggunakan arsitektur Convolutional Neural Network (ConvNeXt), kemudian dimensi temporal dianalisis menggunakan model LSTM Autoencoder untuk memahami representasi temporal data. Hasil representasi tersebut selanjutnya dianalisis menggunakan dua algoritma clustering, yakni K-means dan HDBSCAN. Evaluasi hasil clustering dilakukan menggunakan metrik Silhouette Score, Calinski-Harabasz Index, dan Davies Bouldin Index , dengan nilai tertinggi sebesar 0,72, yang diperoleh dari algoritma HDBSCAN. Nilai ini menunjukkan kualitas pemisahan klaster yang cukup baik. Penelitian ini berhasil mengidentifikasi perbedaan dan kesamaan pola polusi cahaya antar observatorium, serta menunjukkan potensi pendekatan deep learning dalam analisis spasio-temporal citra satelit. Hasil dari studi ini diharapkan dapat memberikan kontribusi ilmiah dalam upaya konservasi langit malam dan mendukung kebijakan mitigasi polusi cahaya di area sekitar observatorium. ; Observatories play a crucial role in the advancement of astronomical science, making the preservation of the surrounding night sky an essential aspect to be maintained. However, population growth, urban expansion, and increased economic activity have led to a rise in night sky brightness, resulting in light pollution. This study aims to identify patterns and cluster characteristics of light pollution around observatories worldwide using a deep learning-based approach combined with clustering analysis. The dataset used consists of satellite imagery from the Visible Infrared Imaging Radiometer Suite–Day/Night Band (VIIRS-DNB) spanning from 2012 to 2024, within a 50 km radius of each observatory. Spatial features were extracted using the ConvNeXt architecture, a modern Convolutional Neural Network (CNN), while temporal dimensions were analyzed using an LSTM Autoencoder to capture the temporal representation of the data. The resulting representations were then clustered using 2 algorithms, namely K-means and HDBSCAN. Clustering performance was evaluated using the Silhouette Score metric, with the highest score of 0.72 achieved by both HDBSCAN indicating a reasonably good cluster separation quality. This study successfully identifies both differences and similarities in light pollution patterns across observatories and demonstrates the potential of deep learning in spatio-temporal satellite image analysis. The findings are expected to contribute to scientific efforts in night sky conservation and support light pollution mitigation policies around observatory areas.

	Text S_KOM_2102665_Title.pdf Download (718kB)
	Text S_KOM_2102665_Chapter1.pdf Download (301kB)
	Text S_KOM_2102665_Chapter2.pdf Restricted to Staf Perpustakaan Download (1MB)
	Text S_KOM_2102665_Chapter3.pdf Download (303kB)
	Text S_KOM_2102665_Chapter4.pdf Restricted to Staf Perpustakaan Download (3MB)
	Text S_KOM_2102665_Chapter5.pdf Download (265kB)
	Text S_KOM_2102665_Appendix.pdf Restricted to Staf Perpustakaan Download (318kB)

Item Type:	Thesis (S1)
Additional Information:	https://scholar.google.com/citations?view_op=list_works&hl=en&user=dEQjm0kAAAAJ ID SINTA Dosen Pembimbing: Lala Septem Riza: 5975668 Judhistira Aria Utama: 6125861
Uncontrolled Keywords:	Polusi Cahaya, ConvNext, LSTM Autoencoder, Observatorium, VIIRS-DNB, Clustering
Subjects:	Q Science > Q Science (General) Q Science > QA Mathematics > QA75 Electronic computers. Computer science Q Science > QB Astronomy Q Science > QC Physics T Technology > T Technology (General)
Divisions:	Fakultas Pendidikan Matematika dan Ilmu Pengetahuan Alam > Program Studi Ilmu Komputer
Depositing User:	Muhammad Cahyana Bintang Fajar
Date Deposited:	11 Sep 2025 14:20
Last Modified:	11 Sep 2025 14:20
URI:	http://repository.upi.edu/id/eprint/138328

Actions (login required)

View Item