Citra Digital untuk Deteksi Paku di Jalan Menggunakan YOLOv4-Tiny

IMMANUELLA SENJA DWI FEBRIANI, . (2024) Citra Digital untuk Deteksi Paku di Jalan Menggunakan YOLOv4-Tiny. Sarjana thesis, UNIVERSITAS NEGERI JAKARTA.

	Text Cover.pdf Download (3MB)
	Text BAB I.pdf Download (1MB)
	Text BAB II.pdf Restricted to Registered users only Download (3MB) | Request a copy
	Text BAB III.pdf Restricted to Registered users only Download (2MB) | Request a copy
	Text BAB IV.pdf Restricted to Registered users only Download (2MB) | Request a copy
	Text BAB V.pdf Restricted to Registered users only Download (502kB) | Request a copy
	Text DAFTAR PUSTAKA.pdf Download (1MB)
	Text LAMPIRAN DAN DAFTAR RIWAYAT HIDUP.pdf Restricted to Registered users only Download (2MB) | Request a copy

Abstract

Dalam era transportasi modern, keamanan jalan raya dan pemeliharaan infrastruktur menjadi sangat penting untuk mendukung mobilitas dan keselamatan pengguna jalan. Salah satu tantangan utama adalah keberadaan paku di jalan yang dapat menyebabkan kerusakan ban, kecelakaan, dan gangguan lalu lintas. Penelitian ini bertujuan untuk mengembangkan sistem deteksi paku di jalan menggunakan teknologi pengolahan citra dengan YOLOv4-tiny. Model ini menunjukkan performa yang baik dalam mendeteksi paku, ditandai dengan penurunan nilai loss hingga 0,2876 dan peningkatan mAP hingga 70% pada iterasi ke-5400. Meskipun terdapat penurunan mAP setelah iterasi ke-5400 yang menunjukkan potensi overfitting, model secara keseluruhan mampu mengenali objek paku dengan cukup baik dalam data pelatihan. Evaluasi performa menunjukkan bahwa model memiliki AP sebesar 90,87% untuk kelas "paku," dengan TP sebanyak 394 dan FP sebanyak 32, menunjukkan kemampuan deteksi yang tinggi untuk objek paku. Namun, untuk kelas "bukan paku," AP hanya mencapai 49,83%, dengan TP sebanyak 65 dan FP sebanyak 51, menunjukkan bahwa model masih sering salah mendeteksi objek bukan paku. Statistik performa lainnya, seperti presisi 85%, recall 82%, F1-score 83%, dan rata-rata IoU sebesar 67,17%, menunjukkan bahwa meskipun model cukup baik, masih ada ruang untuk peningkatan dalam akurasi deteksi objek bukan paku. Implementasi teknologi ini diharapkan dapat membantu mencegah kebocoran ban dan meningkatkan keselamatan berkendara. Kata Kunci. Paku, YOLOv4-Tiny, deteksi (*****) In the era of modern transportation, road safety and infrastructure maintenance have become crucial for supporting mobility and ensuring the safety of road users. One of the primary challenges is the presence of nails on the road, which can cause tire damage, accidents, and traffic disruptions. This study aims to develop a nail detection system using image processing technology with YOLOv4-tiny. The YOLOv4-Tiny model demonstrated good performance in detecting nails, marked by a decrease in loss value to 0.2876 and an increase in mAP to 70% at the 5400th iteration. Although there was a decrease in mAP after the 5400th iteration, indicating potential overfitting, the model overall successfully recognized nail objects in the training data. Performance evaluation showed that the model achieved an Average Precision (AP) of 90.87% for the "nail" class, with 394 true positives (TP) and 32 false positives (FP), indicating a high detection capability for nail objects. However, for the "non-nail" class, the AP was only 49.83%, with 65 TPs and 51 FPs, indicating that the model often misclassified non-nail objects. Other performance statistics, such as precision at 85%, recall at 82%, F1-score at 83%, and an average Intersection over Union (IoU) of 67.17%, suggest that while the model is quite good, there is still room for improvement in the accuracy of non-nail object detection. The implementation of this technology is expected to help prevent tire punctures and enhance driving safety. Keywords. Nail, YOLOv4-Tiny, detection

Item Type:	Thesis (Sarjana)
Additional Information:	1). Dr. Mutia Delina, M.Si. 2). Haris Suhendar, S.Si., M.Sc.
Subjects:	Sains > Fisika
Divisions:	FMIPA > S1 Fisika
Depositing User:	Immanuella Senja Dwi Febriani .
Date Deposited:	20 Aug 2024 07:08
Last Modified:	20 Aug 2024 07:08
URI:	http://repository.unj.ac.id/id/eprint/50171

Actions (login required)

View Item