DETEKSI DAN KLASIFIKASI KERUSAKAN PADA JALAN BERASPAL MENGGUNAKAN JARINGAN SYARAF TIRUAN DENGAN TRANSFER LEARNING

Nugraha, Varrel Kusuma (2025) DETEKSI DAN KLASIFIKASI KERUSAKAN PADA JALAN BERASPAL MENGGUNAKAN JARINGAN SYARAF TIRUAN DENGAN TRANSFER LEARNING. Undergraduate thesis, Universitas Pembangunan Nasional “Veteran” Jawa Timur.

	Text (COVER) 21081010198_Nugraha Varrel Kusuma_Cover.pdf Download (618kB)
	Text (BAB I) 21081010198_Nugraha Varrel Kusuma_Bab1.pdf Download (26kB)
	Text (BAB II) 21081010198_Nugraha Varrel Kusuma_Bab2.pdf Restricted to Repository staff only until 5 May 2028. Download (163kB)
	Text (BAB III) 21081010198_Nugraha Varrel Kusuma_Bab3.pdf Restricted to Repository staff only until 5 May 2028. Download (130kB)
	Text (BAB IV) 21081010198_Nugraha Varrel Kusuma_Bab4.pdf Restricted to Repository staff only until 5 May 2028. Download (1MB)
	Text (BAB V) 21081010198_Nugraha Varrel Kusuma_Bab5.pdf Download (27kB)
	Text (DAFTAR PUSTAKA) 21081010198_Nugraha Varrel Kusuma_Dafpus.pdf Download (144kB)
	Text (LAMPIRAN) 21081010198_Nugraha Varrel Kusuma_Lampiran.pdf Restricted to Repository staff only until 5 May 2028. Download (241kB)

Abstract

Road damage detection and classification play a crucial role in maintaining road infrastructure and ensuring traffic safety. This study explores the application of artificial neural networks using transfer learning, specifically leveraging the EfficientNet-B2 architecture, to detect and classify asphalt road damage into four categories: potholes, surface peeling, cracks, and edge cracks. The dataset used consists of 6,770 images, divided into training, validation, and testing sets with a 70:20:10 ratio. Various experimental scenarios were conducted to optimize model performance, including testing different optimizers, learning rates, dataset partitioning ratios, fine-tuning layers, image resolutions, and data augmentation techniques. Surprisingly, the results show that the model performs best without data augmentation, achieving an accuracy of 94.19%, precision of 94.12%, recall of 94.10%, and an F1-score of 94.09%. This finding indicates that the dataset already contains sufficient variations, making augmentation unnecessary and even potentially detrimental to model performance. Additionally, the evaluation using the ROC-AUC curve demonstrated a near-perfect classification capability, with AUC values close to 1.00 for all damage categories. The confusion matrix analysis also confirmed the model's robustness, with minimal misclassification between classes. The results of this research suggest that transfer learning with EfficientNet-B2 is highly effective for detecting and classifying asphalt road damage. Moreover, the finding that data augmentation is not always beneficial highlights the importance of dataset quality over quantity in deep learning applications. These insights can contribute to the development of automated road monitoring systems for better infrastructure management.

Item Type:

Thesis (Undergraduate)

Contributors:

Contribution	Contributors	NIDN/NIDK	Email
Thesis advisor	Diyasa, I Gede Susrama Mas	NIDN197006192021211009	igusrama.if@upnjatim.ac.id
Thesis advisor	Anggraeny, Fetty Tri	NIDN198202112021212005	fettyanggraeny.if@upnjatim.ac.id

Subjects:

Z Bibliography. Library Science. Information Resources > Z665 Library Science. Information Science

Divisions:

Faculty of Computer Science > Departemen of Informatics

Depositing User:

Nugraha Varrel Kusuma

Date Deposited:

05 May 2025 07:14

Last Modified:

05 May 2025 07:14

URI:

https://repository.upnjatim.ac.id/id/eprint/36094