KLASIFIKASI KUALITAS PERMUKAAN JALAN RAYA MENGGUNAKAN METODE CNN BERBASIS ARSITEKTUR XCEPTION

Abstract

Abstrack-Highways are the main infrastructure for land transportation. The better the condition of a highway, the better the speed and safety for drivers. One of the main causes of accidents on highways is due to the road conditions being unsuitable for use because of damage. Therefore, monitoring and maintaining the surface condition of roads is very important. The quality check of highways is generally done manually, a method that requires significant time and effort. The vast number of roads and the manual checks that consume a lot of time and money become obstacles in road maintenance. Therefore, the system "Highway Surface Quality Classification Using CNN Method Based on Xception Architecture" was developed as an alternative to perform surface quality checks on highways. This method uses deep learning CNN with Xception transfer learning architecture. Xception was chosen because it has a complex yet efficient architecture in terms of time usage and high accuracy for image classification, producing accurate models with short training times. Furthermore, several previous studies have shown that Xception outperforms several other architectures. The use of deep learning in classifying highway surface damage is expected to speed up and simplify the process of monitoring road surface conditions. The model is created using a dataset with 4 classes based on the level of damage released by the Ministry of Public Works and Public Housing (PUPR). The highest test results showed a model accuracy of 90.11% and 90% for validation.