The most significant issue in remote sensing is the extraction of roads from very high-resolution satellite (VHR) photos. The authors of this research provide an effective Densenet121 block that can understand relationships between global features. As a result, road segmentation is more precise due to the ability of each geographical version available as a pointer to other data that is collected. In the specific, our single model outperformed every other contemporary aggregation model that has been presented in the official Densenet121, our suggested model offers a shorter training convergence time, fewer parameters, and fewer Giga floating-point operations per second (GFLOPs). The authors also provide empirical evaluations on how non-local blocks should be used appropriately for the base model. Theoretically, the applied methodology provided a DenseNet-121 and an unpublished best alternative identification method DenseNet-121, a really sizable dataset provided by the CNN model's training phase. Additionally, because the implemented architecture takes advantage of data augmentation, no custom data extraction method is necessary.