Open-Set Radar Waveform Classification: Comparison of Different Features and Classifiers

Performing open set classification of radar waveforms is a difficult problem due to issues including varying signal to noise ratio (SNR), complexity of the data, lack of separability between classes of interest, as well as the crowded nature of the spectrum. In addition, the evolving spectrum may lead to a situation where not every waveform is present in the training library. This paper addresses these challenges by the combination of obtaining machine learning features directly from the waveform, subsequently followed by a classification algorithm. The machine learning technique used in this paper is a discriminative network, specifically a convolutional neural network (CNN), for feature extraction. The classifier employed is SV-Means, a quantile one-class support vector machine-based algorithm (q-OCSVM), with the ability to reject unknown waveform classes while also providing an estimation of the likelihood of the class of interest being a member of the waveform library. A combination of these two methods results in a system of high credibility taking into account the challenges noted.