skip to main content

Road Damage Identification Using Deep Learning Method Convolutional Neural Networks Model

*Yoga Triardhana  -  Department of Geodetic Engineering, Universitas Diponegoro, Indonesia
Bandi Sasmito  -  Departemen Teknik Geodesi, Fakultas Teknik, Universitas Diponegoro, Indonesia
Firman Hadi  -  Departemen Teknik Geodesi, Fakultas Teknik, Universitas Diponegoro, Indonesia

Citation Format:
Abstract
Roads are an important infrastructure for daily human life in land transportation. All land transportation activities run on roads so that roads have great benefits for human life. The road network has a strategic role in development, for that it must be managed as well as possible so that its function can be felt as expected (Directorate of Public Works, 2005). All of these road uses are very beneficial for the smooth running of human activities, but there is a risk of accidents that can occur on the road. One of the reasons is because the road conditions are not suitable for use due to damaged road conditions. As many as 10 to 20 percent of road accidents are caused by road damage (Bambang Susanto to Kompas, 2016). Monitoring of road conditions has obstacles because the number of roads that have been built is so many that it can slow down the time needed to find out the current road conditions. The use of DL for object identification has been widely used, one of which is the application of the CNN model to identify road damage (Maeda, 2018). The CNN model generated in this study was used to identify road damage along Jalan Karangrejo to Jalan Lamongan Raya and Jalan Setia Budi to Jalan Perintis Kemerdekaan. The results of model identification are then analyzed for the accuracy of the results from the CNN model with validation data. The resulting output is the type of road damage and the location of the damage. The CNN model produced was able to identify as many as 205 road damage points from the research location along 11.7 kilometers along with the damage class. The results of the application of this CNN model based on the validation results have spatial accuracy with an RMSE value of 8.38 meters and have an overall accuracy value of 85.34% and a kappa of 82.36% using a confusion matrix. This shows that the resulting road damage identification model is feasible to use and is able to help efficiency in monitoring current road conditions so that road repairs can be carried out immediately.
Fulltext View|Download
Keywords: Convolutional Neural Networks;Deep Learning;Identification
Article Info
Section: Articles
Language : ID-ID
Statistics:
Share:

Last update:

No citation recorded.

Last update:

No citation recorded.