Seafloor soil classification is essential for marine engineering, environmental monitoring, and geological surveys. Traditional classification methods, such as physical sampling and acoustic backscatter analysis, have inherent limitations, including spatial constraints and inconsistencies in distinguishing sediments with similar acoustic properties. This study uses frequency-dependent acoustic reflection coefficients to investigate a novel spectral-based approach to seabed soil classification. Experiments were conducted in a controlled aquatic environment to isolate the spectral characteristics of two soil types: poorly graded sand (SP) and poorly graded gravel (GP). The research employed calibrated transducers to measure reflection coefficients across the 100–400 kHz frequency range, allowing for a comparative spectral analysis between the two sediments. The results demonstrate that SP and GP exhibit distinct spectral fingerprints, with SP showing higher reflectance across all measured frequencies, while GP displays a more variable spectral response. These findings suggest that frequency-dependent reflectance provides a more sensitive and accurate classification criterion than conventional backscatter intensity analysis. By eliminating environmental variability and focusing on intrinsic soil properties, this study establishes a foundation for automated, non-invasive classification methods that could be integrated into machine learning frameworks for real-time seabed characterization. The proposed methodology enhances the precision of remote sensing techniques and presents significant advantages in offshore engineering, environmental monitoring, and hydrographic surveys. Future research should extend this approach to diverse sediment types and open marine environments to refine and validate its applicability in real-world scenarios.