US 12,092,732 B2
Synthetic aperture radar classifier neural network
David Payton, Calabasas, CA (US); Soheil Kolouri, Agoura Hills, CA (US); Kangyu Ni, Calabasas, CA (US); and Qin Jiang, Oak Park, CA (US)
Assigned to The Boeing Company, Arlington, VA (US)
Filed by The Boeing Company, Chicago, IL (US)
Filed on Oct. 6, 2021, as Appl. No. 17/450,161.
Prior Publication US 2023/0105700 A1, Apr. 6, 2023
Int. Cl. G01S 13/90 (2006.01); G01S 7/41 (2006.01); G01S 13/933 (2020.01)
CPC G01S 13/9027 (2019.05) [G01S 7/417 (2013.01)] 20 Claims
OG exemplary drawing
 
1. A computing system comprising:
a processor configured to train a synthetic aperture radar (SAR) classifier neural network, wherein the SAR classifier neural network includes a SAR encoder, an image encoder, a shared encoder, and a classifier that are trained at least in part by:
at the SAR encoder, receiving a plurality of training SAR range profiles that are tagged with a respective plurality of first training labels;
at the image encoder, receiving a plurality of training two-dimensional images that are tagged with a respective plurality of second training labels;
at the shared encoder, receiving a plurality of SAR encoder outputs of the SAR encoder and a plurality of image encoder outputs of the image encoder and computing a plurality of shared latent representations based at least in part on the plurality of SAR encoder outputs and the plurality of image encoder outputs;
at the classifier, receiving the shared latent representations and computing a respective plurality of classification labels based at least in part on the shared latent representations;
computing a value of a loss function based at least in part on the plurality of first training labels, the plurality of second training labels, and the plurality of classification labels; and
based at least in part on the value of the loss function, performing backpropagation through the classifier, the shared encoder, the SAR encoder, and the image encoder.