Abstract: Objectives To address the conflict between lightweight design and low-noise performance in composite unmanned surface vehicle (USV) structures and to enhance underwater acoustic stealth, this study investigates the acoustic radiation characteristics and thickness optimization of composite sandwich USV structures. Methods Based on vibro-acoustic coupling theory, a finite element acoustic model of a composite sandwich USV structure composed of glass fiber reinforced polymer (GFRP) skins and a PVC foam core is established to numerically predict underwater acoustic radiation in the 10-250 Hz frequency range. Parametric analyses are conducted to systematically examine the effects of GFRP skin thickness and PVC core thickness on structural acoustic radiation. On this basis, a dual-objective optimization model with structural weight and radiated sound pressure total level as objectives is constructed. A Gaussian process regression surrogate model is employed to represent the thickness-acoustic response relationship, and a multi-objective optimization is performed using the non-dominated sorting genetic algorithm to obtain the Pareto-optimal solutions in the continuous thickness design space. Results The results indicate that the acoustic radiation performance of the sandwich structure is sensitive to variations in both skin and core thicknesses, with the PVC core thickness playing a dominant role in acoustic performance improvement, while the GFRP skin thickness varies within a relatively narrow range once structural requirements are satisfied. The obtained Pareto front exhibits a pronounced diminishing-returns characteristic. Based on the analysis of the Pareto solution set, three representative thickness-matching schemes—lightweight, balanced, and enhanced noise-reduction—are identified. For example, the balanced scheme, with a skin thickness of approximately 3-4 mm and a core thickness of approximately 10-18 mm, achieves a reduction of about 3-5 dB in the radiated sound pressure total level while maintaining acceptable structural weight. Conclusions The proposed analysis and optimization framework provides quantitative engineering guidance for the low-noise design of composite sandwich structures in unmanned surface vehicles.