Specialists from the V.V. Tikhomirov Scientific Research Institute of Instrument Design have developed a method that allows a radar system to independently determine the class of a ballistic target using a neural network. The radar is not limited to detecting an object, but analyzes its movement and tries to determine what exactly is approaching.

After detection, the target is tracked. The system repeatedly measures five parameters at once — effective reflection area, range, azimuth, elevation angle, and Doppler frequency. From these data, a multidimensional time series is formed, which is then transmitted to the neural network.

The authors propose training it on mathematical models of various ballistic targets. Examples include mortar shells, artillery shells, and multiple launch rocket system ammunition.

The main difference from more complex recognition methods is that the neural network itself searches for characteristic features in the target's trajectory, without requiring a complete set of criteria to be manually specified in advance.