Deep Learning

Deep Learning for Radar

Apply artificial intelligence techniques to radar applications

Simulate radar signals to train machine and deep learning models for target and signal classification.​

With MATLAB® and Simulink®, you can:

  • Label signals collected from radar systems using the Signal Labeler app
  • Augment datasets by simulating radar waveforms and echoes from objects with a range of radar cross sections​
  • Simulate micro-Doppler signatures of hand gestures and animated objects with non-rigid bodies such as helicopters, pedestrians, and bicyclists
  • Apply identification and classification workflows to public datasets

Why Use Deep Learning for Radar?

Synthesize radar signals to train machine and deep learning models for target and signal classification and apply deep learning techniques to data collected from radar systems.

Waveform classification

Waveform Classification

Synthesize and label radar waveforms to train deep learning networks.  Extract time-frequency features from signals and perform waveform modulation classification using deep learning networks.

Using plots to show how the values change over time.

Radar Target Classification

Classify radar returns based on radar cross sections with both machine and deep learning approaches. The machine learning approach uses wavelet scattering feature extraction coupled with a support vector machine. Two common deep learning approaches are transfer learning using SqueezeNet and a Long Short-Term Memory (LSTM) recurrent neural network.

Comparing actual and predicted labels for hand gesture classification.

Hand Gesture Classification

Classify ultra-wideband (UWB) impulse radar signal data from a publicly available dataset of dynamic hand gestures. Use a multiple-input, single-output convolutional neural network (CNN) where the CNN model extracts feature information from each signal before combining it to make a final gesture label prediction.

Micro-Doppler Signature Classification​

Classify pedestrians and bicyclists based on their micro-Doppler characteristics using time-frequency analysis and a deep learning network. The movements of different parts of an object placed in front of a radar produce micro-Doppler signatures that can be used to identify the object. 

Viewing predicted boxes and labels on test image.

SAR Image Classification

Use deep learning techniques for target classification of Synthetic Aperture Radar (SAR) images. Create and train a Convolution Neural Network (CNN) to classify SAR targets from the Moving and Stationary Target Acquisition and Recognition (MSTAR) Mixed Targets dataset.

Viewing predicted boxes and labels on test image.

SAR Image Recognition

Perform target recognition of Synthetic Aperture Radar (SAR) images using a Region-based Convolutional Neural Networks (R-CNN). The R-CNN network integrates detection and recognition with efficient performance that scales to large scene SAR images.