Dynamic aerial target recognition by using radar range profiles

This article considers the problem of aerial target recognition based on analysis characteristics of radar range profiles and also information base structure of recognition system. We will compare static and dynamic recognition schemes. Firstly, static pattern of target is built, this pattern is data received at a moment. Secondly dynamic patterns of target are built, they are tracking data of target in a specific period of time.

There is an important feature of radar range profiles: they have complex configuration and their range profiles can be greatly changed when change observation angle. Moreover, some aerial targets of different types in different course angles may have similar range profiles and therefore generate indistinguishable pattern in the signature space. Thus, the static recognition schemes may give wrong answers regardless of the choice of informative signatures and make it difficult or even exclude the possibility of their practical use.

Studies show that required reliability and quality of moving target recognition can be obtained by applying dynamic recognition scheme. In this scheme, dynamic pattern is built by data aggregation at different moments of radar observation.

Radar observation occurs in noise and interference conditions, which limits resolution and accuracy of radar, uncertainties and variability of reflection properties of targets. Therefore, only robust recognition algorithms can have practical value, they are designed to work in conditions of uncontrolled disturbing factors. These dynamic recognition schemes allow to minimizing the impact of uncertainty factors by expanding data base of recognition algorithms.

The first stage of radar recognition procedure is forming informative signatures, which is considered as an input pattern tuple of target observation, then it solves the problem pattern classification. There are two ways of dynamic aerial target recognition, which differ in the processing order of observation data. The first method uses aggregation of input data, which is supplied to the input of the classifier. In the second method, aggregation procedure applies with respect to the output data of the classifier.