Sound target strength (TS) is one of the important indicators for measuring the acoustic stealth performance of underwater weapon platforms such as submarines and large unmanned underwater vehicles. The application of sound-absorbing structures is one of the key technologies for controlling the TS of underwater structures. The research on sound-absorbing metamaterials has emerged owing to the rapid development of acoustic metamaterials. Most sound absorption structures are implemented using local resonance models, which include thin-film metamaterials, curled space metamaterials, Helmholtz resonant cavities, local resonance scatterers, and so on. When these complex microstructures are applied to the surfaces of large underwater equipment, the finite element model for calculating TS has countless small mesh sizes due to the consideration of the fine features of the complex structures, resulting in a large number of meshes that are difficult to calculate or optimize. To solve this problem, the neural network deep learning model is utilized to extract the elastic equivalent parameters of complex sound-absorbing structures within a single period range based on the transfer matrix method of the elastic layer for the first time. This paper validates the typical homogeneous structures, composite sandwich structures, and composite structures with cavities. The errors in the TS values calculated from the original structures and the equivalent parameters are all within 2[Formula: see text]dB, taking the application of structures on a cylinder with a 0.5-m radius as an example. It demonstrates that these equivalent parameters can be used to accurately and quickly calculate the TS of complex sound-absorbing structures. The calculation method based on equivalent parameters proposed in this paper provides convenience and efficiency for the optimization design of structures pursuing lower sound target strength.