Skip Connection Variant of Modified U-Net Architecture for Satellites Imagery

In this paper, we introduce an innovative model that builds upon the traditional U-Net architecture, specifically elevating the semantic segmentation performance for satellite imagery. Our architectural modification capitalizes on a concatenate block to effectively integrate the feature maps derived from each block. This strategic integration aids in mitigating the information loss from the extracted features and facilitates their equal distribution among numerous decoder blocks. The methodology underpins the capacity to augment semantic segmentation performance pertinent to satellite imagery, inherently marked by its intricate characteristics and wide-ranging features.