PENGEMBANGAN MODEL HYBRID CONVOLUTIONAL CAPSULE NETWORK UNTUK DETEKSI MULTI-OBJEK PADA CITRA TERMAL

Muhammad Gelvan Alfiandi, - and Galura Muhammad Suranegara, - and Endah Setyowati, - (2025) PENGEMBANGAN MODEL HYBRID CONVOLUTIONAL CAPSULE NETWORK UNTUK DETEKSI MULTI-OBJEK PADA CITRA TERMAL. S1 thesis, Universitas Pendidikan Indonesia.

Abstract

Pencitraan termal memainkan peran penting dalam berbagai aplikasi, termasuk pengawasan, navigasi otonom, dan pemantauan keselamatan pejalan kaki. Studi ini menyajikan implementasi Hybrid Convolutional Capsule Network (HCCN) untuk deteksi multi-objek dalam citra termal, dengan fokus pada klasifikasi objek human dan cyclist. Model dievaluasi menggunakan metode 6-Fold Cross-Validation, yang memastikan distribusi kelas objek yang seimbang di seluruh subset validasi. Pendekatan Stratified K-Fold digunakan untuk menjaga distribusi ini, sehingga penilaian model tetap adil dan tidak bias. Temuan ini menegaskan bahwa HCCN mampu melakukan generalisasi dengan baik dalam berbagai skenario pencitraan termal, menjadikannya solusi yang andal untuk tugas deteksi multi-objek di dunia nyata. Penelitian di masa depan dapat berfokus pada penanganan ketidakseimbangan kelas melalui teknik augmentasi data atau class weighting, guna lebih meningkatkan kinerja deteksi, terutama untuk kelas dengan jumlah data yang lebih sedikit. Kata Kunci— Citra Thermal, Deteksi Objek, Deteksi Multi-Objek, Machine Learning, Deep LearnIing ----- Thermal imaging plays a crucial role in various applications, including surveillance, autonomous navigation, and pedestrian safety monitoring. This study presents the implementation of a Hybrid Convolutional Capsule Network (HCCN) for multi-object detection in thermal images, focusing on the classification of human and cyclist objects. The model was evaluated using 6-Fold Cross-Validation, ensuring a balanced distribution of object classes across all validation subsets. A Stratified K-Fold approach preserved this distribution, ensuring fair and unbiased model assessment. These findings highlight HCCN’s capability to generalize well across different thermal imaging scenarios, making it a robust solution for real-world multi-object detection tasks. Future research could focus on addressing class imbalance through data augmentation or class weighting strategies to further enhance detection performance, particularly for minority classes. Keywords— Thermal Imaging, Object Detection, Multi-Object Detection, Machine Learning, Deep Learning

	Text S_SISTEL_2104425_Title.pdf Download (464kB)
	Text S_SISTEL_2104425_Chapter1.pdf Download (281kB)
	Text S_SISTEL_2104425_Chapter2.pdf Restricted to Staf Perpustakaan Download (353kB) | Request a copy
	Text S_SISTEL_2104425_Chapter3.pdf Download (797kB)
	Text S_SISTEL_2104425_Chapter4.pdf Restricted to Staf Perpustakaan Download (661kB) | Request a copy
	Text S_SISTEL_2104425_Chapter5.pdf Download (280kB)
	Text S_SISTEL_2104425_Appendix.pdf Restricted to Staf Perpustakaan Download (1MB) | Request a copy

Item Type:	Thesis (S1)
Additional Information:	https://scholar.google.com/citations?view_op=list_works&hl=id&authuser=4&user=r7W1CDkAAAAJ ID SINTA Pembimbing Galura Muhammad Suranegara: 6703764 Endah Setyowati: 6681149
Uncontrolled Keywords:	Citra Thermal, Deteksi Objek, Deteksi Multi-Objek, Machine Learning, Deep LearnIing
Subjects:	T Technology > T Technology (General)
Divisions:	UPI Kampus Purwakarta > S1 Sistem Telekomunikasi
Depositing User:	Muhammad Gelvan Alfiandi
Date Deposited:	10 Sep 2025 08:12
Last Modified:	10 Sep 2025 08:12
URI:	http://repository.upi.edu/id/eprint/136667

Actions (login required)

View Item