expanded u-net model for road extraction from satellite images

Reliable extraction of information from aerial images is a challenging issue with many practical applications. one of the specific challenges within this problem is the automatic detection of roads. due to the presence of shadows, obstructions, and a wide variety of non-road objects, this task is considered as a complex problem in computer vision. despite the previous efforts in the field of automatic road detection, there is still room for improving in this area. this paper aims to enhance detection accuracy by proposing a model for road segmentation in satellite images based on image segmentation techniques. to this end, we introduce a novel model, namely expanded u-net (eu-net) by embedding the vgg19 layers to the base u-net model. evaluation results on the deepglobe road extraction dataset indicate enhancements in results compared to a base u-net model.