Seafloor sediment acoustic property inversion from reflection coefficients with a PSO-BP neural network approach

This study presents an innovative approach for marine sediment parameter inversion based on the Biot theory and the Biot-Stoll model to generate training datasets for a Particle Swarm Optimization-Backpropagation (PSO-BP) neural network. The developed inversion network was validated using surface data collected from in situ measurements and laboratory samples in the northwestern South China Sea. The experimental results demonstrated high accuracy in retrieving sediment properties such as porosity, density, and sound speed across multiple frequencies. Specifically, the average relative error was 2.06% for porosity when utilizing laboratory sample data at 100 kHz, and 3.79% for porosity when applied to in situ measurement data at 8 kHz. Comparison of high-frequency data (100 kHz) with mid-frequency in situ data (8 kHz) confirmed the robustness and adaptability of the method under different frequency conditions. The validation results underscore the effectiveness of the proposed inversion framework for marine sediment characterization, indicating its potential for integration into marine observation systems for enhanced seabed monitoring and resource assessment.