RESEARCH ON EVALUATING THE PERFORMANCE OF DEEP LEARNING NETWORKS IN RADAR SIGNAL RECOGNITION

Abstract

Artificial intelligence (AI) is widely used in civil applications such as: target recognition, object recognition, automatic flight control or natural language processing. In military applications, AI is also used for classifying and recognizing radar signals. Nowadays, there are different deep learning networks used for solving these problems such as: SqueezeNet, AlexNet, DarkNet-53 and DenseNet-201. This paper compares the performance of these networks by simulated radar signals in MATLAB environments. The simulated results showed that all network required the same training time. However, the AlexNet provided the highest accuracy (P = 87.5%), following
by DarkNet-53 (P = 85.62%), DenseNet-201 (P = 84.32%) and the lowest accuracy was provided
by SqueezeNet with P = 75.0%.