02-05-2023 дата публикации
Номер: CN116047448A
Принадлежит:
The invention discloses a method for predicting conductor target RCS (radar cross section), which comprises the following steps: fusing a physical optical method (PO) as prior information into Gaussian process regression, constructing a new GPR covariance function by adopting spectrum asymptotic representation of a linear covariance function and a steady-state covariance function, designing an initialization method based on empirical spectral density, and predicting the RCS of a conductor target. And finally, a universal and high-precision RCS prediction model is obtained. By introducing spectrum asymptotic representation of a stationary covariance function, the limitation of adopting a period hypothesis can be eliminated, so that GPR can adaptively represent RCS features of targets with simple and complex shapes, and higher prediction precision is provided; the method is initialized, convergence performance in the training process can be ensured, and prediction precision is improved.
Подробнее