SEN2ROAD: Roads are also visible using Sentinel

Today there is an abundance of digitized cartographic information on different artificially created elements, especially roads and highways.

In the most developed countries roads of cities of considerable size are very well mapped through companies’ global services, or initiatives such as OSM (Open Street Map). The quality and even existence of this road mapping and its layers, however, declines as the size of the city decreases, and when they are located in developing countries with limited resources.

To all this it must be added that these layers of information are updated at very different rates, depending on their geolocation. On continents like Africa, for example, the cartographic information is very poor, or outdated, which generates a lack of knowledge and control in areas such as mobility and public services.


At Tracasa we have tackled the challenge of extracting highways and roads from images drawn from the constellations of Sentinel satellites. As is known, the spatial resolution of Sentinel-2A images is some 10 meters, in many cases is insufficient for proper segmentation of the terrain. This is why we had to rely on our SENX4 super-resolution model to enhance images and generate more precise segmentation.

Thanks to the advances in our SEN2ROAD model for the segmentation of roads and highways, a wide range of possible uses opens up:

-Rapid Mapping. Access to quick road maps

-Logistical and mobility planning

-Disaster responses of all kinds

-Analysis of urban development and carbon footprints

-The creation and updating of cartography


Satellite images are an invaluable resource for all remote sensing and territorial segmentation activities. Today, the availability of road segmentation models functioning at an adequate level is highly dependent on the spatial resolution of the satellite images used. Generally, in these cases images featuring a resolution of 2.5 meters or less are used, and they entail a high cost.

Tracasa, with the help of its super-resolution SENX4 model, is making it possible to use free- access images such as those from the Sentinel missions for this type of work, thereby greatly reducing the cost and making it possible for more companies, institutions and individuals to access these services.

SEN2ROAD is our model for the segmentation of roads and highways using Sentinel images, generally urban, and of highways, at any location worldwide. It is a model that employs the most advanced AI technology to get results, given the difficulty of this type of segmentation through Sentinel images.

Road segmentation in the city of Barcelona, with the Avenida Diagonal in the center of the image.
An aerial view of Barcelona, including its entire municipal area, featuring segmentation of its roads.
A segmentation of roads and highways in the city of Huesca.


To correctly interpret the results of our model, we employed the most widely used metrics for the assessment of segmentation models in the field of Deep Learning. As can be seen on the following table, SEN2ROAD has already achieved metrics close to 0.86 in PPV and 0.72 in mIoU – until now unthinkable for road segmentation models using Sentinel images.

Image of Bilbao featuring road segmentation.

An aerial image of Seville, with a segmentation of its roads and highways.


The segmentation of highways and roads at the pixel level on site is of great value in terms of spatial representativeness, but it is insufficient if it is to be used for decision-making in the areas of planning, communications and mobility.

With a view to making the most of the segmented information through our SEN2ROAD model, we have continued to advance with our optimized vectorization of rasters.

Through our SEG2VECTOR algorithm library we are able to vectorize the raster information of roads and highways in an extremely optimized way. Thus, we have obtained a Shape format with all the structured information on the roads. This type of vector information makes possible the simulation of transport networks and subsequent studies based on mobility criteria.


Urban development study

The world’s population continues to grow, inexorably, and its concentration in large urban centers has negative environmental effects. This growth can have a very negative impact if it occurs without observing rigorous energy efficiency standards.

The generation of a hypertemporal urban development index allows us to establish, based on remote sensing techniques, the growth patterns of cities and study their energy and environmental impacts. In this way a valuable tool for studying citizens’ mobility is made available to governments.

Tesla Gigafactory in Berlin-Brandenburg (Germany).

Mobility and the calculation of alternative routes

In many cases involving reduced mobility due to the impact of natural environmental disasters, the quick mapping of roads is extremely important in order to be able to indicate alternative routes.

Route calculation within the network of nodes and vectorized segments.

Keeping maps as up to date as possible provides a very valuable layer of information when it comes to meeting mobility needs in areas that may not have the infrastructure or sufficient means to deal with the effects produced by a catastrophe.

SENX4: At the cutting edge of superresolution technology


When we talk about extracting information of images from multispectral sensors deployed on satellites, their spatial resolution is one of the main factors that will define their future uses. The ESA constellation of Sentinel 2A satellites offers to the scientific community multispectral images every five days in an open way. But one of the drawbacks of these images is their spatial resolution (10 meters per pixel), limiting their future applications.

There are a lot of methods to increase the spatial resolution of the images. Some solutions try to improve the 13 bands of each satellite, taking all of them to 10 meters using the bands with the máximum resolution. Other solutions develop models with 2X and 4X scaling factors from the 13 bands, thinking that is possible to use this factor to take the bands from 10 meters to 2,5 meters. But nobody has yet created a model to transform the 10 meters bands into 2,5 meters with a contrast of real bands.

In Tracasa, with our SENX4 model we have improved the RGB and NIR bands resolution, going from 10 meters to 2,5 meters with the 4X factor.


Until now, to increase the resolution were used the traditional technics of interpolation and resampling: linear interpolation, bilinear, cubic and bicubic. In our case, to increase the resolution in a 4X factor, we have used deep learning technics and with cutting-edge convolutional neural networks we increase the resolution of the images without including unwanted patterns, improving the traditional methods.

We have not used sharpening or color adjustments. We avoid generating very synthetic images, far from the reality of the sensor. Our model only uses artificial intelligence to learn the patterns of the data and get an image as real as possible.



To interpret the results we have used a measuring system that is frequently used in the analysis of images: the PSNR and SSIM combination. These measures are indicators of the quality of the results based on an objective image to achieve. The higher the value in the metric, the better the image will be adjusted to the real result. Taking bicubic like the baseline to improve in the traditional models of interpolation, we have achieved an important improvement in the metrics, that offers, as a result, a more defined image, clearer and more adjusted to reality.







Semantic segmentation (Streets and buildings)

With the new image SENX4 (2,5 meters) we can have a muticlass semantic segmentation more correct and accurate. We improve the segmentation metrics that we got with the data of Sentinel 2A 10 meters.

Clustering algorithms

With SENX4 resolution we optimize the results that we get with clustering algorithms. We add more precision in the spatial limits of the clusters and in their representative average.