This numerical approach uses pixel ratio and resolution to estimate the final result. We also introduce a simple mathematical formulation to solve the second step of calculating road area from the satellite image. In this study, we propose an extensive post-processing method that outperforms the previous methods combined with the deep learning architectures. The pixel map should be subjected to post-processing to obtain a clean binary mask. Semantic models such as U-net generate a noisy pixel map of the roadway mask. We also highlight the significance of using aggregated residual transformations, such as ResNeXt, in place of regular residual networks.

This paper aims to analyze the performance of this architecture on the roadway extraction problem. proposed an improved variant of U-net, U-net++, which outperformed U-net in medical image segmentation.

These approaches involve using U-net family models, which are state-of-the-art architectures for roadway extraction problems. The first step is the roadway extraction from the given satellite images, while the second step involves calculating the roadway area using pixel resolution.Д®xisting approaches to solving the first step use deep learning architectures to generate a semantic map of the road line, which is then post-processed to obtain a final road mask. We formulate this task as a two-step problem. The task uses open-source satellite images with a specified resolution to calculate the entire road area covered in the image view.

The calculation of the roadway area from openly available satellite images aims to automate the analysis of the total road area covered by satellite images.

Spacenet satellite