Transfer learning with convolutional neural networks for moving target classification with micro-Doppler radar spectrograms

In this work, we propose a transfer learning approach with Convolutional Neural Networks (CNNs) for radar Automatic Target Recognition (ATR). Radar echo signals of moving targets introduce micro-Doppler signatures that are widely used in classifying moving targets. Spectrograms have the advantage of expressing the distinctive micro-Doppler signatures of different targets, and thus fed as 2D images to a CNN model. A pre-trained CNN model namely AlexNet is employed as a feature extractor in which feature maps can be extracted from any of the layers to train a classical classifier. SoftMax classifier have been used in this approach. The efficiency of the presented framework is demonstrated on the public RadEch database of 8 ground moving target classes, in which the experimental results indicate that our methodology significantly outperforms other competitive state-of-the-art methods with an accuracy of 99.9%.