Off-road terrains are unstructured environments that emerge as one of the most challenging regions to deal with in ensuring secured Automated Guided Vehicle (AGV) navigation. Maneuvering through such amorphous topography, it is necessary to extract geometric and semantic information from the environment for executing projected and secure navigation to derive confident AGV movement. The paper addresses the significance of efficient techniques for algorithmically representing such informal terrains in accomplishing autonomous rapid movement-focusing rescue or service operations. The work also tries to highlight the proposed architecture Spatial-Attention Segmentation (SAS) from the existing end-to-end approaches in terms of the efficiency rate of concerned performances in unstructured, off-road settings. The research work majorly focuses on increasing geo-semantic precision accuracy by capturing pixelated spatial features of the objects within the environment. The constructed architecture of the segmentation decoder head, SpatiomapNet, has been emphasized here along with the mixvision transformer as the backbone of the proposed neural model. Multi-scale feature extraction through spatial attention defines the traversability of terrains. The accumulation of semantic information is done through multiview RGB inputs. Evaluation and benchmarking of the proposed architecture are established by training and testing the model on the RUGD dataset, consisting of various parameterized off-road structures. The model has also been tested on outdoor environmental custom data, collected from a real-world scenario to overcome the overfitting nature of the neural network. SAS has also been compared with State-of-the-Art (SOTA) methods where it achieves an improvement over mean Intersection over Union (mIoU) by 1.26% with an actual mIoU of 90.34%, and Average Precision Accuracy (aAcc) of 95.89% which points to an improvement over the recent existing algorithm by 0.21% on the taken datasets.