Algorithm for Recognition of Small Air Targets by Trajectory Features in Passive Bistatic Radar

Abstract

   Introduction. In the past few years, the rapid development and widespread use of unmanned aerial vehicles (UAVs) for solving a variety of tasks has created new threats. The problem of ensuring the safety of protected objects, especially in the area of critically important objects or in places with difficult ornithological conditions (airports, wind power facilities), is of particular importance. In this regard, the issue of detecting small air targets and recognizing their type and degree of danger is acquiring greater importance. This paper presents an algorithm for recognizing air targets based onartificial intelligence technology. The results of a comparative analysis of decision-making methods for recognizing small UAVs and birds based on their trajectory features are presented. The results obtained can be used in the development of systems for recognizing classes of small airborne targets in existing and future radar stations.   Aim. Development of an algorithm for recognizing small air targets by trajectory features based on machine learning. Implementation and evaluation of the quality of decision-making methods in a given recognition problem.   Materials and methods. Experimental data on the trajectories of UAVs and birds obtained in a passive bistatic radar system are used. The trajectory parameters of the targets and their statistical characteristics are calculated; a set of features for recognition is formed. Using the MATLAB software package, a program for implementing the recognition algorithm and analyzing the quality of decision-making methods was developed.   Results. An algorithm for recognizing air targets based on artificial intelligence technology is presented. A comparative analysis of the six most common recognition methods based on machine learning (Naïve Bayes, decision trees, k-nearest neighbors, neural network recognition algorithm, support vector machine, random forests) was carried out, which showed that, under the conditions of this problem, the most effective are k-nearest neighbor method and support vector machine.   Conclusion. The presented methods can be used to directly determine the class of targets from a set of marks of their trajectories. Further research will be aimed at developing and implementing a real-time recognition algorithm.