Abstarct: Accurate road extraction from satellite imagery is one of the important research topics in the fields of computer vision and remote sensing, playing a key role in Geographic Information Systems (GIS) and navigation applications. In this study, a lightweight and efficient method called Improved Mobile-Unet is proposed for road region detection from satellite images. The proposed model is baxsed on the standard U-Net architecture; however, to reduce computational complexity, the encoder part is replaced with laxyers from the MobileNetV2 network. This modification enables faster model execution on resource-constrained devices such as mobile phones and embedded systems. In the decoder section, transposed convolution laxyers and skip connections are employed to reconstruct the segmentation map and effectively fuse low-level and high-level features. Furthermore, to improve the detection accuracy of narrow road structures, a modified loss function baxsed on Dice Loss is introduced. The performance of the proposed model is evaluated on the DeepGlobe dataset. Experimental results demonstrate that the proposed approach achieves an IoU of 94.89% and an F-Score of 97.38%, outperforming baxseline models while maintaining lower computational complexity.