Sistem Deteksi dan Klasifikasi Pothole pada Jalanan Menggunakan Algoritma YOLOv11 dan Faster R-CNN

Abstract

Road damage, especially potholes, poses significant risks to both road users and a country's infrastructure. Conventional detection methods are often ineffective in dealing with the visual patterns and environmental conditions present on roads. This research aims to develop a pothole detection system using deep learning, specifically YOLOv11 and Faster R-CNN. These two algorithms will be compared to determine which model is more effective in handling road surface patterns. The dataset was collected directly from several road sections in Medan City and classified into four categories: small-shallow, small-deep, large-shallow, and large-deep. The testing results show that YOLOv11 is more effective in detecting potholes with higher mAP 0,894, precision 0,92, recall 0,95, and F1-score 0,93. However, misclassifications still occurred, indicating the need for a more diverse dataset to improve the model's performance in detecting potholes on roads.