Accurate estimation of raster points of an 2D-satellite data and identifying heat-spots and figuring out possible causes with correlation of historical data acquired from different satellites seems to be difficult practically. On this context, this research focuses on developing sophisticated and desired automated pipelines for identifying and extracting these parameters from satellite imagery using deep learning-based image segmentation. Deep learning techniques were implemented such as CNN-based architectures; like U-Net and Seg-Net to segment high-spectral resolution images to segment based on different classes effectively, enabling region-based analysis. Firstly, this research includes different interpolation techniques for identifying different DEM patterns from vector data to raster data and to enhance spatial resolution and filling missing raster points in 2D image scale. Secondly, the proposed system addresses this challenge by leveraging the capabilities of U-Net’s performance, known for their effectiveness in modeling and segmenting images based on ground-truth image (known as mask image). By using encoder-decoder capabilities along with Bottleneck. The U-Net architecture will then be trained on multispectral image data using techniques which involve raster data and vector data. Evaluation of this model system is conducted using various measuring metrics including m-IoU, accuracy, precision, Loss and R2 value. Additionally, the system's predictive capabilities will be compared against baseline models and existing approaches to assess its effectiveness in real-world scenarios contributing to Sustainable Development Goal (SDG-13). Overall, this paper aims to observe and classify different key component classes and align with given multispectral images to visualize different UHI components on maps using different interpolation techniques.