<![CDATA[Road damage is an infrastructure problem that affects public safety and transportation mobility. Manual road damage identification is considered less effective because it is time-consuming and subjective. This study aimed to implement semantic segmentation using the U-Net and Fully Convolutional Network (FCN) architectures for road damage detection based on digital images. The research used the Cross Industry Standard Process for Data Mining (CRISP-DM) method with datasets consisting of secondary PotholeMix data and primary data collected in West Bekasi. Model training was conducted using 256×256 pixel images with evaluation metrics including Accuracy, Dice Score Coefficient, Intersection over Union (IoU), and Loss. The results showed that the U-Net model achieved better performance than FCN with an Accuracy of 0.9652, Dice Score Coefficient of 0.9596, IoU of 0.9267, and Loss of 0.0674. Furthermore, the model was successfully implemented into a Flutter-based mobile application for automatic road damage identification and monitoring.]]>
Copyrights © 2026
