Autoencoder assisted radar for target identification

1. An autoencoder assisted radar for target identification, comprising:

an Intelligent Metamaterial (“iMTM”) antenna module configured to radiate a transmission signal with an iMTM antenna structure and generate radar data capturing a surrounding environment;

a data pre-processing module having an autoencoder configured to encode the radar data into an information-dense representation; and

an iMTM perception module configured to detect and identify a target in the surrounding environment based on the information-dense representation and to control the iMTM antenna module.

2. The autoencoder assisted radar of claim 1 , wherein the autoencoder includes an encoder network and a decoder network, and wherein the encoder network includes a feed-forward neural network that learns to compress input data into the information-dense representation that is used to reconstruct the input by the decoder network.

3. The autoencoder assisted radar of claim 2 , wherein:

the encoder network is configured to receive the radar data and compress the radar data into the information-dense representation, and

the decoder network is configured to obtain the information-dense representation and reconstruct the radar data into an output that represents a multi-dimensional radar Range-Doppler (“RD”) map from the information-dense representation.

4. The autoencoder assisted radar of claim 1 , wherein the autoencoder is trained via backpropagation with raw input radar data.

5. The autoencoder assisted radar of claim 4 , wherein weights and biases of the trained autoencoder are fed into a convolutional neural network for target detection and identification.

6. The autoencoder assisted radar of claim 5 , wherein the convolutional neural network obtains RD map encoding data and provides output data indicating identifying information about a detected target, wherein the identifying information includes one or more of a type of the target, a location of the target, or a velocity of the target.

7. The autoencoder assisted radar of claim 6 , wherein the iMTM perception module comprises a deep neural network configured to select a next radar action based at least on the output data from the convolutional neural network.

8. The autoencoder assisted radar of claim 1 , wherein the iMTM perception module is further configured to provide the information-dense representation to one or more neural networks to classify one or more objects for target detection and identification.

9. The autoencoder assisted radar of claim 1 , wherein the data pre-processing module is further configured to aggregate a plurality of RD maps in the radar data from radar scans across a field of view at a rate at which the iMTM antenna module generates the radar data.

10. An autoencoder for assisting a radar system, comprising:

an encoder network configured to receive an input and compress the input into an information-dense representation; and

a decoder network configured to obtain the information-dense representation and reconstruct the input into an output that represents a multi-dimensional radar Range-Doppler (“RD”) map from the information-dense representation.

11. The autoencoder of claim 10 , wherein the encoder network and the decoder network are fully-connected feed-forward neural networks having a plurality of layers that are substantially mirror images from each other.

12. The autoencoder of claim 10 , wherein the plurality of layers of the encoder network and decoder network is programmable to a predetermined number of layers.

13. The autoencoder of claim 10 , wherein the encoder network is trained to generate the information-dense representation from one or more RD maps.

14. The autoencoder of claim 13 , wherein weights and biases of the trained encoder network are fed into a deep convolutional neural network for target detection and identification.

15. The autoencoder of claim 10 , wherein the decoder network generates a radar point cloud that identifies one or more points of interest from the information-dense representation.

16. A method for identifying a target with an autoencoder assisted radar in a surrounding environment, the method comprising:

directing an Intelligent Metamaterial (“iMTM”) antenna structure to radiate Radio Frequency (“RF”) beams with determined parameters;

receiving reflections from the RF beams to generate radar data about the surrounding environment;

encoding the radar data into an information-dense representation;

identifying a target in the surrounding environment from the information-dense representation; and

determining a next radar control action for the iMTM antenna structure.

17. The method of claim 16 , further comprising training an autoencoder with a data set of unlabeled radar Range-Doppler (“RD”) map data.

18. The method of claim 17 , further comprising:

sending weights and biases of the trained autoencoder to initialize a convolutional neural network (CNN);

training the CNN with labeled RD map data;

capturing radar data including RD map data with the trained CNN;

encoding the captured radar data and RD map data with the autoencoder to generate an information-dense representation of the captured radar data and RD map data; and

feeding the encoded RD map data to the CNN to identify a target from the captured radar data.

19. The method of claim 18 , wherein determining the next radar control action comprises selecting the next radar control action with a deep neural network based at least on results from the CNN.

20. The method of claim 16 , wherein the next radar control action includes one or more actions that cause steering of one or more RF beams to a same or different direction.