Abstract：This paper proposes a new sparse deconvolu-tion method based on matching pursuit，which canbe used to automatically interpret thin-beds.Forsingle layer thin-bed，we construct the thin-bed re-flection coefficient models of equal-magnitude-same-sign and equal-magnitude-opposite-sign withdifferent position and thickness.Taking their seis-mic response as a set of atom functions，we decom-pose real seismic trace data by matching pursuit，and the thin-bed top and bottom are picked auto-matically.For multiple thin-bed layers，we con-struct all possible thin-bed reflection coefficientmodels.Then we match the models with the realseismic trace data，and calculate the optimal reflec-tion coefficient amplitude and the correspondingmatching error.Finally，we treat the reflection co-efficient model with the minimum matching erroras the thin bed interpretation result.The wedgemodel synthetic data test proves that this methodis valid.The method is also used to predict thecarboniferous stratum thickness in a survey of Si-chuan Basin，and the thickness prediction matchesvery well with real-drilled results.
