Road networks are crucial to urban infrastructure and significantly affect transportation, traffic management, and emergency response. Besides, accurate mapping is essential for detecting road networks effectively, but traditional methods like manual digitization and field surveys often struggle in fast-changing urban environments. Remote sensing and deep learning techniques have emerged as effective alternatives, although initial road segmentation faced challenges such as limited image resolution. Recent advances in satellite technology have alleviated these issues by providing ultra-high-resolution (sub-meter) imagery, which is vital for accurately representing road networks. Deep learning models like U-Net have enhanced road segmentation by accurately capturing complex features. This research examines the effectiveness of multi-resolution satellite imagery for road segmentation. This study aims to analyze the accuracy assessment of road segmentation using Sentinel-2 imagery (10 m resolution) and ultra-high-resolution Pléiades Neo imagery (sub-meter resolution). Ground truth data from the Google Maps API were used for validation. Among the tested resolutions, Pléiades Neo at 30 cm achieved the highest accuracy, with an F-score of 0.87. Pléiades Neo at 15 cm resolution followed closely with an F-score of about 0.85. Pléiades Neo at 1 m resolution (upscaled data) showed a moderate decline (F-score of 0.82), while Sentinel-2 had the lowest performance (F-score of 0.78). Overall, Pléiades Neo at 30 cm resolution offers the best balance of accuracy and data efficiency for road segmentation.