The application of blended acquisition along with irregular acquisition geometries contributes to the economic perspective of a seismic survey. The joint migration inversion scheme is capable of directly processing the data acquired in this way, i.e., without deblending or data reconstruction, and of subsequently estimating both reflectively and velocity models. The workflow proposed in this study aims to design the source blending operator as well as detector and source sampling operators. The approach iteratively computes these parameters in such a way that the quality of reflectivity and velocity models, which are directly estimated from blended and irregularly-sampled data, is adequate. The workflow integrates a genetic algorithm and a convolutional neural network to derive optimum parameters. Bio-inspired operators enable the simultaneous update of the blending and sampling operators. To relate the choice of survey parameters to the performance of a joint migration inversion, we utilize a convolutional neural network. The applied network architecture discards suboptimal solutions among newly generated ones. Conversely, it passes optimal ones to the subsequent step, which successfully enhances the efficiency of the proposed approach. The resultant acquisition scenario yields a notable enhancement in both reflectivity and velocity estimates attributed solely to the choice of survey parameters.

@en

We introduce a workflow to derive survey parameters responsible for source blending as well as spatial sampling of detectors and sources. The proposed workflow iteratively performs the following three steps. The first step is
application of blending and sub-sampling to an unblended and well-sampled data. We then apply a closed-loop deblending and data reconstruction enabling a robust estimate of a deblended and reconstructed data. The residue
for a given design from this step is evaluated, and subsequently used by genetic algorithms (GAs) to simultaneously update the survey parameters related to both blending and spatial sampling. The updated parameters are fed into a next iteration till they satisfy given stopping criteria. We also propose repeated encoding sequence (RES) used to form a parameter sequence in GAs, making the proposed designing workflow computationally affordable. We demonstrate the results of the workflow using numerically simulated examples that represent blended dispersed source array data. Difference attributable only to a way to design parameters is easily recognizable. The optimized parameters yield clear improvement of deblending and data reconstruction quality and subsequently provide optimal acquisition scenarios. Additionally, comparison among different optimization schemes illustrates ability of GAs along with RES to efficiently find better solutions.@en

We introduce a generalized concept of blending and deblending, and establish the generalized-blending and - deblending models. Accordingly, we establish a method of deblending, or deblended-data reconstruction, using
these models. The generalized blending can handle real-life situations; this includes random encoding both in the space and time domain, both at the source and receiver side, thus all incoherent and inhomogeneous shooting, signature stamping, non-uniform and under sampling. Similarly, the generalized deblending includes data reconstruction that works all for shot-generated-wavefields separation, spectrum recovery and balancing, designature, regularization and interpolation, again both at the source and receiver side. However, we do face a challenging question: how to fully reconstruct deblended data from the fully generalized blended data. To address this, we consider an iterative optimization scheme using a so-called closed-loop approach with the generalized-blending and -deblending models, in which the former works for the forward modelling and the latter for the inverse modelling in the closed
loop. We established and applied this method to synthetic datasets. The results show that our method succeeded to fully reconstruct deblended data from the fully generalized blended data.@en