Deteksi Retak Pada Jalan Beton Menggunakan Transformer Video Object Detection (TransVOD) dari Data Video

Akbar, Mohammad Fisal Aly (2023) Deteksi Retak Pada Jalan Beton Menggunakan Transformer Video Object Detection (TransVOD) dari Data Video. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 06111940000057-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only until 1 October 2025. Download (1MB) | Request a copy

Abstract

Pengembangan infrastruktur terutama infrastruktur jalan di Indonesia terus mengalami peningkatan. Perkerasan jalan berbahan beton dianggap sebagai solusi pembangunan infrastruktur jalan agar lebih kuat dan tahan lama. Walaupun perkerasan jalan beton memiliki ketahanan yang tinggi, hal tersebut tidak bisa terlepas dari kerusakan akibat faktor alam maupun faktor manusia. Retak jalan beton menjadi kerusakan awal yang apabila tidak segera ditangani akan menyebabkan kerusakan lebih parah. Sistem deteksi retak secara otomatis perlu dikembangkan untuk mempermudah proses inspeksi. Penelitian Tugas Akhir ini berfokus pada eksplorasi salah satu model berbasis arsitektur Transformer Network untuk mendeteksi retak pada jalan beton berdasarkan data video. Model yang digunakan adalah TransVOD Lite yang merupakan pengembangan dari model TransVOD yakni arsitektur yang menggabungkan antara CNN sebagai ekstraktor fitur dari suatu citra dan Spatial-Temporal Transformer sebagai detektor objek. Pada Tugas Akhir ini telah dilakukan deteksi retak pada jalan beton menggunakan model TransVOD Lite dengan data video. Terdapat tiga tahapan utama dalam penelitian Tugas akhir ini. Tahap pertama yakni pengumpulan data berupa rekaman video jalan beton yang memiliki retak beserta anotasinya, tahapan kedua yakni pelatihan model sehingga model dapat mengenali retak jalan beton, dan tahapan ketiga adalah uji coba untuk mengetahui performa dari model TransVOD Lite. Dari data validasi model dapat mencapai performa mean average percision (mAP) sebesar 47,5%. Dari data uji model dapat memperoleh performa terbaik pada skenario tingkat pencahayaan rendah dan kecepatan kendaraan saat pengambilan video berada pada 20-30 km/h dengan accuracy sebesar 90,46%, precision sebesar 80,77%, recall sebesar 94,23% dan F1-score sebesar 86,98%.
===============================================================================================================================
Infrastructure development, especially road infrastructure in Indonesia, continues to increase. Road pavement made of concrete is considered as a solution to the development of road infrastructure to be stronger and more durable. Although concrete road pavement has durability, it cannot be separated from damage due to natural and human factors. and human factors. Cracking of concrete roads is the initial damage that if not addressed immediately will cause more severe damage. not addressed immediately will cause more severe damage. An automatic crack detection system automatic crack detection system needs to be developed to simplify the inspection process. Final Project Research This research focuses on exploring one of the models based on the Transformer Network architecture to detect cracks in concrete roads based on video data. The model used is TransVOD Lite which is a development of the TransVOD model, an architecture that combines CNN as a feature extractor from an image and the architecture that combines CNN as a feature extractor from an image and Spatial-Temporal Transformer as an object detector. In this Final Project, crack detection has been carried out on concrete road using TransVOD Lite model with video data. There are three main stages in this final project research. The first stage is data collection in the form of recorded video recordings of concrete roads that have cracks and their annotations, the second stage is model training so that the model can recognize cracks in concrete roads, and the third stage is data collection. so that the model can recognize concrete road cracks, and the third stage is a test to determine the performance of the TransVOD Lite model. the performance of the TransVOD model. From the validation data, the model can achieve mean average percision (mAP) of 47,5%. From the model test data, it can get the best performance performance in the scenario of low lighting levels and vehicle speeds when taking in 20-30 km/h with accuracy of 90,46%, precision of 80,77%, recall of 94,23% and F1-score of 86.98%.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Deteksi, Retak, Jalan Beton, TransVOD, Detection, Crack, Concrete Road, TransVOD
Subjects:	Q Science > Q Science (General) > Q325.5 Machine learning. Support vector machines. Q Science > QA Mathematics > QA336 Artificial Intelligence Q Science > QA Mathematics > QA76.87 Neural networks (Computer Science) T Technology > TA Engineering (General). Civil engineering (General) > TA440 Concrete--Cracking.
Divisions:	Faculty of Science and Data Analytics (SCIENTICS) > Mathematics > 44201-(S1) Undergraduate Thesis
Depositing User:	Mohammad Fisal Aly Akbar
Date Deposited:	29 Aug 2023 01:37
Last Modified:	29 Aug 2023 01:37
URI:	http://repository.its.ac.id/id/eprint/103652

Actions (login required)

View Item