Accurate and fast are two requirements difficult to achieve at the same time if we consider automatic road extraction.

There are some innovative studies about applying Neural Networks approaches, however usually they need a proper training and are seldom used on SAR images (see a paper here in researchgate at the first link attached).

I suggest you to start reading this review: "State of the art on automatic road extraction for GIS update: a novel classification" by J.B. Mena (2003) (second link attached). Then you can give a read to: Road network extraction from SAR imagery supported by context information (B. Wessel) (third link).

http://www2.uah.es/juan_mena/Publicaciones/State_of_the_art_PATREC.pdf

http://www.isprs.org/proceedings/XXXV/congress/comm3/papers/295.pdf

Article Automated road network extraction using artificial neural network

hi,

I agree with Giulio Ceriola that its very difficult to have at the same time accurate and fast method for road extraction and specialy in SAR images, It depends of the resolution and the noise of the SAR images. You can see some of our papers (Baadeche et al) and  the research of Florence Tupin, see her homepage link: http://perso.telecom-paristech.fr/~tupin/

Very good and comprehensive reply from @Giulio Ceriola. I agree to it. Speed with accuracy is still a challenge.

The question is ill posed. SAR image doesn't mean too much. A set of different SAR modes are now operated, different SAR sensor with very different characteristics....Once that you define your data set you can select the most appropriate procedure. On a general point of view I invite to consider SAR polarimetric modes.

I'm also working in SAR field and interested in road detection problem. Can anyone suggest me a good dataset of SAR images for testing of road detection methodology?

IFSAR data is a good dataset that you can use for road extraction. on the other hand, object based image analysis is suitable for such applications.