Radar Waveform Recognition With ConvNeXt and Focal Loss

Abstract

A method of automatic recognition of radar waves based on time-frequency analysis (TFA) and ConvNeXt model is proposed in the paper. The method aims to address the challenges of feature extraction difficulty and low recognition correctness in complex multi-class radar waveform recognition under the conditions of low signal-to-noise ratio (SNR) and sample imbalance. The first step is to convert the signal’s time domain waveform into a two-dimensional time-frequency image (TFI) using TFA. This allows for the essential characteristics of the signal to be reflected. Next, TFI preprocessing is carried out to retain complete information, which is then adapted to the input of the deep learning (DL) model. Finally, ConvNeXt model was constructed to automatically extract signal TFI features to realize radar waveform classification and recognition, and focal loss was embedded to increase the weight of classes with small sample numbers in training and solve the problem of sample imbalance. Therefore, ConvNeXt network has stronger feature learning and representation ability compared with other algorithms based on deep learning, and effectively improves the overall recognition rate of 16 classes signals. The recognition accuracy of 6 types of signals with similar time-frequency characteristics (Frank, LFM, P1, P2, P3, P4) is higher, and the recognition rate is close to normal samples under the condition of unbalanced samples. Experimental results verify the effectiveness of the proposed algorithm.

Keywords

• Radar waveform recognition
• time-frequency analysis (TFA)
• deep learning (DL)
• ConvNeXt
• focal loss