This study presents a deep learning approach for semantic segmentation of urban environments using the U-Net model. The proposed U-Net architecture consists of an encoder and decoder structure. The encoder uses convolutional layers and downsampling to extract hierarchical information from input images, while the decoder reconstructs higher-resolution feature maps using upsampling layers. The model is evaluated on the Cityscapes dataset, demonstrating superior performance in terms of accuracy, mean Intersection over Union (IoU), and mean DICE scores compared to existing models. The encoder employs batch normalization and dropout layers to stabilize training and prevent overfitting, while the decoder uses skip connections to integrate high-level and low-level features. The model's performance is assessed using various metrics, including mean IOU, which quantifies the precision of pixel and area categorization. The results show that the proposed U-Net model achieves high accuracy in segmenting urban scenes, making it a valuable tool for applications such as urban planning, transportation management, and autonomous driving. The study also highlights the importance of addressing research gaps such as the reliability of models in varying environmental conditions, small object detection, and the interpretation of 3D urban environments. Future work includes improving model efficiency, enhancing real-time processing capabilities, and exploring domain adaptation techniques to ensure robustness and safety in urban environments. The U-Net model's effectiveness in semantic segmentation of cityscapes underscores its potential for advancing urban planning and smart city initiatives.This study presents a deep learning approach for semantic segmentation of urban environments using the U-Net model. The proposed U-Net architecture consists of an encoder and decoder structure. The encoder uses convolutional layers and downsampling to extract hierarchical information from input images, while the decoder reconstructs higher-resolution feature maps using upsampling layers. The model is evaluated on the Cityscapes dataset, demonstrating superior performance in terms of accuracy, mean Intersection over Union (IoU), and mean DICE scores compared to existing models. The encoder employs batch normalization and dropout layers to stabilize training and prevent overfitting, while the decoder uses skip connections to integrate high-level and low-level features. The model's performance is assessed using various metrics, including mean IOU, which quantifies the precision of pixel and area categorization. The results show that the proposed U-Net model achieves high accuracy in segmenting urban scenes, making it a valuable tool for applications such as urban planning, transportation management, and autonomous driving. The study also highlights the importance of addressing research gaps such as the reliability of models in varying environmental conditions, small object detection, and the interpretation of 3D urban environments. Future work includes improving model efficiency, enhancing real-time processing capabilities, and exploring domain adaptation techniques to ensure robustness and safety in urban environments. The U-Net model's effectiveness in semantic segmentation of cityscapes underscores its potential for advancing urban planning and smart city initiatives.