Design of FIR frequency-space wavefield extrapolation filters using back-propagation neural networks with application to pre-stack imaging of seismic data

We propose a novel method to design complex-valued Finite Impulse Response (FIR) frequency-space seismic wavefield extrapolation digital filters using Back-propagation Neural Networks (BPNN). Convergence condition is also derived using the Lyapunov function. The proposed design algorithm does not require any matrix inversion. The filters are more accurate than other existing filters such as the Li-norm algorithm. They come with a running time savings of 19.82%. In addition, the BPNN filters lead to stable pre-stacked seismic images with a better quality resolution than that of the L₁ -norm pre-stacked ones.