Segmentation of Roads in Satellite Images using specially modified U-Net CNNs

TL;DR

This paper introduces a novel modified U-Net CNN architecture with data augmentation and sliding window techniques for accurate road segmentation in satellite images, outperforming existing methods in mean F-score.

Contribution

The paper presents a specially modified U-Net CNN architecture combined with data augmentation and sliding window approach for improved road segmentation in satellite imagery.

Findings

Outperforms baseline methods in mean F-score

Effective use of data augmentation improves accuracy

Modified U-Net architecture enhances segmentation quality

Abstract

The image classification problem has been deeply investigated by the research community, with computer vision algorithms and with the help of Neural Networks. The aim of this paper is to build an image classifier for satellite images of urban scenes that identifies the portions of the images in which a road is located, separating these portions from the rest. Unlike conventional computer vision algorithms, convolutional neural networks (CNNs) provide accurate and reliable results on this task. Our novel approach uses a sliding window to extract patches out of the whole image, data augmentation for generating more training/testing data and lastly a series of specially modified U-Net CNNs. This proposed technique outperforms all other baselines tested in terms of mean F-score metric.