FECT

Abstract: As a developing country Timor-Leste needs to conduct regular monitoring for road conditions and its furniture, particularly the existence of several road furniture such as gabions, road lines, and retaining walls. This monitoring system always needs to employ people as labor for direct monitoring to the location. In this research, we proposed a method for road and retaining wall monitoring system by applying semantic segmentation using U-Net model. A total of 1,179 road images was used for this research. We examined this method using different image sizes and altering model parameters such as batch size and learning rate. The experiment result shows U-Net model achieve 98.1% of accuracy and 0.94 mean intersection over union (mIoU).

Keywords: road, retaining wall, deep learning, segmentation, u-net