The acquisition of geoacoustic parameters is of great significance in studying ocean acoustics. On the basis of deducing the seabed reflection coefficient under the layered absorbing medium, the influence of the absorption coefficient on the seabed reflection coefficient under the condition of large grazing angles is analyzed. The seabed reflection coefficient oscillates at a frequency. When it is equal to the reflection coefficient of the contact interface between seawater and sediment, the corresponding frequency point is defined as the 1/4 oscillation period frequency. At this frequency, the coupling degree between absorption coefficient of sedimentary layer and substrate geoacoustic parameters is less than those at other frequencies. In this paper, a stepwise optimization inversion method for deep water geoacoustic parameters is proposed based on the seabed reflection characteristics of large grazing angles. Firstly, the interference period of the seabed reflection coefficient is extracted by the correlation method, and the sound speed and thickness of the deposited layer are inverted by the interference period. The density is obtained from the inversion result of sound speed combined with Hamilton empirical formula. Secondly, the value of the absorption coefficient of the sedimentary layer is calculated by combining the search boundary of the substrate sound speed. The one-dimensional inversion of the substrate sound speed is realized by using the substrate reflection coefficient at 1/4 oscillation period frequency. Finally, the one-dimensional inversion of the absorption coefficient of the sedimentary layer is realized by using the seabed reflection coefficient at a half-wave layer frequency. The seabed reflection characteristics of large glancing angles are combined with stepwise inversion to reduce the coupling degree of the substrate sound speed and the absorption coefficient of the sedimentary layer. Experimental results show that the geoacoustic parameters retrieved by this method can be effectively applied to the prediction of propagation loss in a certain range under the condition of large grazing angle measurement.