Feature Extraction for Moving Targets Classification Using Radar Doppler Echoes

  • Esra Al Hadhrami

Student thesis: Master's Thesis


Automatic target recognition (ATR) is one of the most demanding tasks in modern radar systems, which is one of the most challenging to implement. The efficiency of knowledge-based recognition or classification depends highly on the training dataset, the superiority of the extracted features as well as the classifier's performance. Most of the efforts in the radar's target classification focused on the Micro-Doppler signatures as they show distinctive characteristics of different targets. Micro-Doppler signatures are created by the frequency shifts introduced by the target's movement and captured by Doppler radar. Micro-Doppler signatures can be observed in Spectrogram, a time-frequency analysis that illustrates the time varying Micro-Doppler shifts. Convolutional Neural Networks (CNNs) are the state-of-the-art when it comes to image vision tasks as they proved to achieve an outstanding performance. In this thesis, we are utilizing CNNs in a transfer learning approach to classify ground moving targets of Pulsed-Doppler radar. Pre-trained CNNs were used as feature extractors to extract feature automatically from spectrograms, rather than depending on hand-crafted features. Three different pre-trained CNNs modules were exploited, namely, AlexNet, VGG16 and VGG19, as feature extractors whereas the output features were used to train a classifier. Two classifiers were experimented: a multiclass support vector machine (SVM) and a softmax. To validate our approach, we used RadEch database of 8 ground moving target classes. Our approach outperformed the state-of-the-art methods, using the same database, with an accuracy of 99.9%. Indexing Terms: Micro-Doppler, Radar classification, Automatic target recognition, Convolutional Neural Network, Transfer learning, AlexNet, VGG.
Date of AwardApr 2018
Original languageAmerican English
SupervisorNaoufel Werghi (Supervisor)


  • Micro-Doppler
  • Radar classification
  • Automatic target recognition
  • Convolutional Neural Network
  • Transfer learning
  • AlexNet
  • VGG.

Cite this