Pavement Damage Segmentation Using YOLO11 for Road Maintenance Planning in Mempawah
Abstract
Purpose: Road-maintenance planning by regional governments depends on timely information about the type and location of pavement distress, yet routine field surveys are labour-intensive.
Design/methodology/approach: This study evaluates YOLO11 instance segmentation as an applied technology for reviewing vehicle-camera footage from Mempawah Regency, West Kalimantan, Indonesia, and linking visual observations to road-survey records.
Findings: The model was trained on a public four-class distress dataset of 1,988 training, 555 validation and 278 test images. After 80 epochs, it achieved a mask mAP50 of
0.439 and mAP50–95 of 0.163 on the test split. A preliminary check of 14 local images produced six candidates on five images, with image-level pothole agreement in three. Screening 414 frames from 32 recordings yielded 105 candidates on 83 frames, each linked to its source image, timestamp, estimated camera position and survey segment. The workflow supports traceable candidate review that can help road agencies target field verification. Once validated against local annotations and physical measurements, these observations could inform maintenance prioritisation within the Provincial/Kabupaten Road Management System (PKRMS). Future work should quantify verification accuracy and the survey time saved.
Paper type: Research Paper
Downloads
Copyright (c) 2026 IJEBD (International Journal of Entrepreneurship and Business Development)

This work is licensed under a Creative Commons Attribution-ShareAlike 4.0 International License.
