Reidentifikasi Orang Pada Data Visible-Infrared Yang Terkorupsi Menggunakan Klasifier Vision Transformer

Pusaka, Nathanael Tenno Phileo Wong (2024) Reidentifikasi Orang Pada Data Visible-Infrared Yang Terkorupsi Menggunakan Klasifier Vision Transformer. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5024201056_Nathanael Tenno Phileo Wong Pusaka_Tugas_Akhir.pdf - Accepted Version Restricted to Repository staff only Download (14MB) | Request a copy

Abstract

Teknik Reidentifikasi orang, yaitu mengidentifikasi kembali individu berdasarkan data gambar atau video yang telah ada. Dalam era modern, Reidentifikasi menjadi penting di bidang keamanan dan pemantauan. Dataset RegDB-C berisi gambar tampak dan infrared dari berba gai individu. Vision Transformer (ViT) merupakan arsitektur jaringan saraf yang efektif dalam memproses data visual-infrared, terutama saat data tersebut terkorupsi.Penilitian ini berfokus untuk mengidentifikasi orang pada berbagai kondisi pencahayaan, perubahan cahaya yang ek strim, perubahan suhu, data yang terkena noise pada data Visible-infrred dengan menggunakan dataset RegDB-C yang terdiri dari 2.060 citra visible dan 2.060 cintra infrared.Meskipun terda pat permasalahan berupa dataset yang mengalami korupsi, Vision Transformer dapat mengiden tifikasi individu secara akurat.Penilitian ini menggunakan berbagai konfigurasi model Vision Transformer, dengan pengujian menggunakan batchsize 16, 32, 64, dan modifikasi hyperpa rameter seperti learning rate dan loss function.Hasil yang didapatkan tertinggi pada Vision Transformer yang telah dimodifikasi hyperparameter dengan nilai MAP (mean Average Pre cision) yang didapatkan yaitu senilai 0,436899, dengan Rank@1 senilai :0.396440 , Rank@5 senilai 0.601942,dan Rank@10 senilai 0.702265.Penilitian ini diharapkan dapat berkontribusi dalam mengembangkan sistem reidentifikasi pada data visible-infrared yang terkorupsi meng gunakan klasifer Vision Transformer.
====================================================================================================
Person Reidentification techniques involve re-identifying individuals based on existing image or video data. In the modern era, reidentification has become crucial in the fields of security and monitoring. The RegDB-C dataset contains visible and infrared images of various individ uals. Vision Transformer (ViT) is a neural network architecture that is effective in processing visual-infrared data, especially when the data is corrupted. This research focuses on identifying people under various lighting conditions, extreme light changes, temperature variations, and noise in the visible-infrared data using the RegDB-C dataset, which consists of 2,060 visible images and 2,060 infrared images. Despite issues such as data corruption, Vision Transformer can accurately identify individuals. This study employs various configurations of the Vision Transformer model, testing with batch sizes of 16, 32, and 64, and modifying hyperparameters such as learning rate and loss function. The highest results were obtained with the Vision Trans former model that had modified hyperparameters, achieving a mean Average Precision (MAP) of 0.436899, with Rank@1 of 0.396440, Rank@5 of 0.601942, and Rank@10 of 0.702265. This research is expected to contribute to the development of reidentification systems for corrupted visible-infrared data using Vision Transformer classifiers.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Reidentifikasi, Vision Transformer (ViT), Inframerah. Reidentification, Vision Transformer (ViT), Infrared.
Subjects:	T Technology > T Technology (General) > T57.5 Data Processing T Technology > T Technology (General) > T57.8 Nonlinear programming. Support vector machine. Wavelets. Hidden Markov models. T Technology > T Technology (General) > T59.7 Human-machine systems.
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Computer Engineering > 90243-(S1) Undergraduate Thesis
Depositing User:	Nathanael Tenno Phileo Wong Pusaka
Date Deposited:	08 Aug 2024 07:36
Last Modified:	08 Aug 2024 07:36
URI:	http://repository.its.ac.id/id/eprint/111951

Actions (login required)

View Item