Pembangkitan Keterangan Citra Pelanggaran Lalu Lintas Berbasis X-Linear Attention Networks (X-LAN)

Nor, Muhammad Naufal Hawari (2024) Pembangkitan Keterangan Citra Pelanggaran Lalu Lintas Berbasis X-Linear Attention Networks (X-LAN). Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5002201067-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only until 1 October 2026. Download (3MB) | Request a copy

Abstract

Maraknya pelanggaran lalu lintas yang terjadi di Indonesia, khususnya oleh pengendara sepeda motor, mendapat tanggapan serius dari pihak kepolisian dengan diluncurkannya sistem Electronic Traffic Law Enforcement (ETLE). ETLE mampu mendeteksi dan mengklasifikasikan pelanggaran lalu lintas yang terjadi melalui kamera CCTV. Namun, di balik keunggulannya, identifikasi pelanggaran lalu lintas yang ditangkap oleh ETLE masih dilakukan secara manual. Hal ini mendorong perlunya solusi yang dapat membantu sistem dalam meningkatkan kevalidan dari identifikasi pelanggaran melalui pembangkit keterangan citra otomatis atau image captioning. Image captioning merupakan salah satu bidang dalam Artificial Intelligence (AI) yang menggunakan framework encoder-decoder untuk memahami citra dan menghasilkan deskripsi dari citra tersebut. Keterlibatan attention dalam model image captioning memiliki peranan penting karena mampu membantu model untuk memperhatikan bagian penting dari citra. Meskipun demikian, attention konvensional ternyata masih terbatas pada interaksi fitur orde pertama antara fitur-fiturnya. Sebagai solusi, X-Linear attention hadir dengan teknik bilinear pooling yang memanfaatkan interaksi fitur orde lebih tinggi bahkan tak terhingga. Tidak hanya pada mekanisme attention saja, penelitian tersebut juga memperkenalkan model image captioning yang disebut X-Linear Attention Networks (X-LAN). Oleh karena itu, dalam Tugas Akhir ini telah dilakukan penelitian terkait pembangkitan keterangan citra pelanggaran lalu lintas dengan memanfaatkan X-Linear Attention Networks (X-LAN). Hasil skor metrik terbaik yang diperoleh dari eksperimen dalam penelitian ini adalah 68.78 BLEU@1, 59.21 BLEU@2, 50.07 BLEU@3, 41.99 BLEU@4, 55.02 Average BLEU, dan 68.37 ROUGE-L dengan variasi model Orde 8.

=================================================================================================================================

The increasing number of traffic violations in Indonesia, especially by motorcyclists, has prompted a serious response from the police with the launch of the Electronic Traffic Law Enforcement (ETLE) system. ETLE can detect and classify traffic violations through CCTV cameras. However, despite its advantages, the identification of traffic violations captured by ETLE is still carried out manually. This has led to the need for a solution that can help the system in enhancing the validity of violation identification through image captioning. Image captioning is an area within Artificial Intelligence (AI) that utilizes the encoder-decoder framework to understand images and generate descriptions based on them. The involvement of attention mechanisms in image captioning models plays a crucial role as it helps the model focus on important parts of the image. However, conventional attention is still limited to first-order feature interactions. To address this limitation, X-Linear attention introduces with bilinear pooling techniques that utilize higher-order feature interactions, potentially even infinite. Not only does this mechanism enhance attention, but the research also introduces an image captioning model called X-Linear Attention Networks (X-LAN). Therefore, this Final Project has conducted research on image captioning for traffic violations using X-Linear Attention Networks(X-LAN). The best metric scores obtained from experiments in this research are 68.78 BLEU@1, 59.21 BLEU@2, 50.07 BLEU@3, 41.99 BLEU@4, 55.02 Average BLEU, and 68.37 ROUGE-L with an 8th-order model variant.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	image captioning, x-linear attention, orde interaksi fitur, feature interaction orders
Subjects:	Q Science > QA Mathematics > QA336 Artificial Intelligence Q Science > QA Mathematics > QA76.6 Computer programming. Q Science > QA Mathematics > QA76.87 Neural networks (Computer Science)
Divisions:	Faculty of Science and Data Analytics (SCIENTICS) > Mathematics > 44201-(S1) Undergraduate Thesis
Depositing User:	Muhammad Naufal Hawari Nor
Date Deposited:	06 Aug 2024 18:40
Last Modified:	06 Aug 2024 18:40
URI:	http://repository.its.ac.id/id/eprint/114214

Actions (login required)

View Item