Underwater wireless optical communication (UWOC) links are highly susceptible to absorption, scattering, turbulence, and receiver noise, which severely degrade the received waveform and make reliable detection difficult in low-SNR conditions. To address this problem, we present a Q-learning-based adaptive stochastic resonance (Q-SR) method for weak-signal enhancement in UWOC. The method integrates an interpretable bistable stochastic-resonance circuit with lightweight online parameter selection, using recovered SNR as the control reward. A digital-to-circuit parameter mapping enables direct transfer of the learned policy to hardware-adjustable SR parameters. Compared with fixed-parameter SR and swarm-intelligence baselines, Q-SR achieves faster convergence and markedly lower computational cost while preserving competitive BER performance. In hardware experiments, an input SNR of ?4.9 dB is improved to 6.23 dB, with a measured BER of 2.86*10-3.