Deteksi dan Klasifikasi Tingkat Keparahan Lubang Jalan di Kota Medan Menggunakan YOLOv11 dan U-Net Berbasis Mobile

Abstract

Road damage, particularly potholes, poses a serious problem in the city of Medan, as it endangers public safety and disrupts traffic flow. The reporting and handling of road damage are often hindered by limited resources and delayed information. This study aims to develop a mobile application capable of automatically detecting and segmenting potholes using a deep learning approach. The YOLOv11 algorithm is employed for pothole object detection, while the U-Net model is applied for image segmentation. The application is equipped with Global Positioning System (GPS) and geotagging features, enabling accurate pothole location reporting, along with a notification system that alerts users about potholes near their location. Evaluation results show that the YOLOv11 model achieves a Mean Average Precision (MAP) of 0.9151 at an Intersection over Union (IoU) threshold of 0.5, with a precision of 0.9076 and a recall of 0.8620. The U-Net model achieves an accuracy of 0.9676, Dice coefficient of 0.8080, and Intersection over Union (IoU) of 0.7177, indicating reasonably accurate pothole area segmentation. Overall, the evaluation results demonstrate that the YOLOv11 and U-Net models effectively detect and segment potholes with high accuracy. The developed system exhibits strong performance in automatically detecting and mapping potholes, thereby improving the efficiency of damage reporting and contributing to reduced accident risk and enhanced driving comfort in the city of Medan.