print preview Back Homepage

Automatic classification of radar data using machine-learning methods

During the image analysis process, machine-learning (ML) methods are used as standard for automatic image classification and feature recognition. These methods attempt to group image-specific characteristics in a training data set and assign predetermined classes (e.g. «landscape», «dog» and «person») in order to generalise new images.

19.09.2019 | Dr. Roland Oechslin, WTS

How an X-band surveillance radar sees different targets
How an X-band surveillance radar sees different targets

In the area of radar technology, such methods are still not common in operational systems, but are already being used with research demonstrators. In the «Soli» project for example, hand movements are captured by a miniature radar sensor and assigned to gestures such as «scrolling with the thumb», «clicking fingers» and «pressing a button» using machine-learning methods.

At armasuisse S+T, the opportunities offered by ML methods in radar technology were highlighted in a research project carried out in the STS and STC4I specialist areas. As part of this project, a data set was recorded with various targets (persons, vehicles, etc.) and a wide range of radar settings, split into individual processing segments (images) and provided with meta data such as «target is a person» or «target is heading towards the radar».

The aim of the research project is to check how well and under which conditions targets can be detected and classified and whether ML methods can complement and even replace traditional classification methods such as micro-doppler and high-range resolution methods.