To address the limited utilization of waveform physical features and the restricted interpretability and generalization ability of existing deep-learning methods for airborne oceanic LiDAR (AOL) full-waveform classification, this study proposes a dual-driven waveform classification network, DDWC-Net, that integrates AOL full waveforms with their waveform features. The full-waveform branch is used to extract deep temporal features, while the waveform-feature branch selects highly discriminative waveform features, such as intensity, rising-edge area, and rising-edge slope. These features are then nonlinearly projected to a higher-dimensional space and fused with the full-waveform features to achieve waveform classification and raft waveform identification. Experimental results based on field-measured data from the Optech CZMIL system show that, compared with a single full-waveform-driven model, DDWC-Net improves overall accuracy, the Kappa coefficient, macro-averaged F1-score, and macro-averaged IoU by 0.35%, 0.94%, 0.40%, and 0.69%, respectively. The dual-driven strategy based on full waveforms and their waveform features can effectively improve the classification accuracy and physical interpretability of AOL full waveforms, thereby providing new methodological support for shallow-water raft detection.