Abstract: A depth-domain seismic wavelet is the response function of subsurface velocity and exhibits significant depth-variant waveform. It does not satisfy the linear and depth-invariant assumption of the conventional convolution model, making time-domain seismic wavelet extraction methods directly inapplicable to the depth domain. To address this problem, this paper proposes a depth-domain seismic wavelet extraction method that integrates weighted stacking. First, based on the depth-domain weighted stacking method, piecewise cubic Hermite interpolation is introduced to achieve vertical continuity of wavelet morphology. Second, combined with non-stationary convolution theory, depth-variant seismic wavelet clusters and their Circulant matrices are constructed to generate depth-domain synthetic seismic records directly. Finally, a depth-domain generalized seismic wavelet (DGSW) is introduced within the iterative separation parameter estimation (ISPE) framework. Using the Pearson correlation coefficient as a quantitative metric for waveform similarity, this approach reduces the parameter dimension while enabling precise control of the iteration termination condition, thereby achieving accurate extraction of depth-domain seismic wavelets. Model tests and field data applications demonstrate that even under strong noise and complex structures with limited seismic and log data, the proposed method can still extract reliable depth-domain seismic wavelets.