PENERAPAN TEKNOLOGI DEEP LEARNING DENGAN YOLOV4 DAN ALGORITMA SEGMENTASI SEMANTIK UNTUK MENDETEKSI JALAN BERLUBANG DALAM PENINGKATAN INFRASTRUKTUR TRANSPORTASI
Abstract
Abstract: Potholes are a serious problem in transportation infrastructure that directly impacts driver safety and travel efficiency. This study proposes a deep learning-based method by integrating the YOLOv4 algorithm for rapid object detection and using semantic segmentation to produce more precise pixel image analysis of various forms of potholes. The YOLOv4 method (You Only Look Once version 4) is used as the main algorithm in the object detection process because of its ability to identify objects in real-time with high speed and optimal accuracy. This study lies in a hybrid approach between object detection using YOLOv4 and semantic segmentation with DeepLab, which enables the system to recognize potholes. Testing was carried out using two methods: by inputting images (PNG/JPG) and using real-time video input via webcam. The dataset used was 1,561 images with various variants of road damage, which were annotated and labeled using Roboflow. The test results of the YOLOv4 model showed a detection accuracy ranging from 90–92%, with the highest performance of 95% for large potholes at close range, and a minimum of 59% for small potholes at long distances. These results confirm the effectiveness of YOLOv4 in quickly and accurately detecting potholes across various road conditions. This research is expected to contribute to the automation of road infrastructure maintenance systems, reduce the risk of traffic accidents, and improve the comfort and safety of road users.
Keywords: pothole detection; YOLOv4; deep learning; semantic segmentation; transportation infrastructure
Abstrak: Kerusakan jalan berlubang menjadi permasalahan serius dalam infrastruktur transportasi yang berdampak langsung terhadap keselamatan pengendara dan efisiensi perjalanan. Penelitian ini mengusulkan metode berbasis deep learning dengan mengintegrasikan algoritma YOLOv4 untuk deteksi objek secara cepat dan menggunakan segmentasi semantik guna menghasilkan analisis gambar piksel yang lebih presisi terhadap berbagai bentuk jalan berlubang. Metode YOLOv4 (You Only Look Once versi 4) digunakan sebagai algoritma utama dalam proses deteksi objek karena kemampuannya dalam melakukan identifikasi objek secara real-time dengan kecepatan tinggi dan akurasi yang optimal. Penelitian ini terletak pada pendekatan hibrida antara deteksi objek menggunakan YOLOv4 dan segmentasi semantik dengan DeepLab, yang memungkinkan sistem dapat mengenali jalan berlubang. Pengujian dilakukan dengan dua metode yaitu dengan cara input menggunakan citra (PNG/JPG) dan menggunakan input video real-time melalui webcam. Dataset yang digunakan berjumlah 1.561 citra dengan berbagai varian kerusakan jalan, yang dianotasi dan dilabeli menggunakan Roboflow. Hasil pengujian model YOLOv4 menunjukkan akurasi deteksi berkisar antara 90–92%, dengan performa tertinggi sebesar 95% untuk lubang besar pada jarak dekat, dan minimum 59% untuk lubang kecil pada jarak jauh. Hasil ini menegaskan efektivitas YOLOv4 dalam mendeteksi jalan berlubang secara cepat dan akurat di berbagai kondisi jalan. Penelitian ini diharapkan dapat memberikan kontribusi dalam otomatisasi sistem pemeliharaan infrastruktur jalan, mengurangi risiko kecelakaan lalu lintas, serta meningkatkan kenyamanan dan keselamatan pengguna jalan.
Kata kunci: deteksi jalan berlubang; YOLOv4; deep learning; segmentasi semantik; infrastruktur transportasi
Full Text:
PDFReferences
Dib, J., Sirlantzis, K., & Howells, G. (2023). An annotated water-filled, and dry potholes dataset for deep learning applications. Data in Brief, 48. https://doi.org/10.1016/j.dib.2023.109206
Fortin, L. V., Delos Santos, A. R. V., Cagud, P. J. B., Castor, P. R., & Llantos, O. E. (2024). Eyeway: An Artificial Intelligence Of Things Pothole Detection System With Map Visualization. Procedia Computer Science, 251, 216–223. https://doi.org/10.1016/j.procs.2024.11.103
Ihsan, M., Amrizal, M. A., & Harjoko, A. (2024). A pothole video dataset for semantic segmentation. Data in Brief, 53, 110131. https://doi.org/10.1016/j.dib.2024.110131
Jadhav, R., Thite, S., Pawar, S., Patil, K., & Chumchu, P. (2024). Exploring the natural pothole dataset generated by the abrasion and cavitation effects of river water on rocks. Data in Brief, 57, 110873. https://doi.org/10.1016/j.dib.2024.110873
Lee, S. Y., Le, T. H. M., & Kim, Y. M. (2023). Prediction and detection of potholes in urban roads: Machine learning and deep learning based image segmentation approaches. Developments in the Built Environment, 13(October 2022), 100109. https://doi.org/10.1016/j.dibe.2022.100109
Nissimagoudar, P. C., Miskin, S. R., Sali, V. N., Ashwini, J., Rohit, S. K., Darshan, S. K., Gireesha, H. M., Hongal, R. S., Katwe, S. V., Basawaraj, & Nalini, C. I. (2024). Detection of Potholes and Speed Breaker for Autonomous Vehicles. Procedia Computer Science, 237(2022), 675–682. https://doi.org/10.1016/j.procs.2024.05.153
Palwe, S., Gunjal, A., Jindal, S., Shrivastava, A., Deshmukh, A., & Navalakha, M. (2024). An Intelligent and Deep Learning Approach for Pothole Surveillance Smart Application. Procedia Computer Science, 235(2022), 3271–3282. https://doi.org/10.1016/j.procs.2024.04.309
Paramarthalingam, A., Sivaraman, J., Theerthagiri, P., Vijayakumar, B., & Baskaran, V. (2024). A deep learning model to assist visually impaired in pothole detection using computer vision. Decision Analytics Journal, 12(July), 100507. https://doi.org/10.1016/j.dajour.2024.100507
Roman-Garay, M., Rodriguez-Rangel, H., Hernandez-Beltran, C. B., Lepej, P., Arreygue-Rocha, J. E., & Morales-Rosales, L. A. (2025). Architecture for pavement pothole evaluation using deep learning, machine vision, and fuzzy logic. Case Studies in Construction Materials, 22(February), 1–26. https://doi.org/10.1016/j.cscm.2025.e04440
Ruseruka, C., Mwakalonge, J., Comert, G., Siuhi, S., Ngeni, F., & Anderson, Q. (2024). Augmenting roadway safety with machine learning and deep learning: Pothole detection and dimension estimation using in-vehicle technologies. Machine Learning with Applications, 16(March), 100547. https://doi.org/10.1016/j.mlwa.2024.100547
Saisree, C., & Kumaran, U. (2022). Pothole Detection Using Deep Learning Classification Method. Procedia Computer Science, 218, 2143–2152. https://doi.org/10.1016/j.procs.2023.01.190
DOI: https://doi.org/10.54314/jssr.v8i3.4070
Article Metrics
Abstract view : 111 timesPDF - 30 times
Copyright (c) 2025 Muhammad Amin, Irianto Irianto, Feby Ariska