Increasing the Segmentation Accuracy of Aerial Images with Dilated Spatial Pyramid Pooling

This thesis addresses the environmental uncertainty in satellite images as a computer vision task using semantic image segmentation. We focus reduction of error caused by use single-environment models wireless communications. propose to and analysis segment geographical terrain order employ specific propagation model each link. Our architecture achieved segmentation accuracy 89.41%, 86.47%, 87.37% urban, suburban, rural classes, respectively. Results indicate that estimating loss with our multi-environment reduced root mean square deviation (RMSD) respect two publicly available tracing datasets.