Least-squares imaging and deconvolution using the hybrid norm conjugate-direction solver |
There are two ways to linearize this model. The first one is to use model perturbation and neglect the non-linear higher order terms in the following:
in which are the initial model and source wavelet respectively. are the pertubation of them, the linearized inversion will output . The other way of linearization is a two-stage linear least squares formulation; i.e. alternately fixing one term (m or s) and inverting for the other one. First use an initial wavelet , keep unchanged and invert for model m
As is in all non-linear inversion problems, the difficulty in these methods is to find a good starting model. Another issue is to add proper constrain on the wavelet , for example, the wavelet should have constant energy during inversion, but this constrain does not fit the linear inversion framework.
Least-squares imaging and deconvolution using the hybrid norm conjugate-direction solver |