This paper addresses the challenges of spatiotemporal heterogeneity and cross-modal semantic gaps in maritime ship data by proposing a spatiotemporal-aware joint hashing method for radiation signal-image association retrieval. A dual-stream network is designed to extract temporal signal features (via a multi-layer convolutional architecture with dynamic time-frequency capture) and spatial image features (using a DenseNet-based feature pyramid with adaptive pooling and cross-modal fusion). A Spatio-Temporal Perception Module (STPM) enhances feature representation through heterogeneous encoding and cross-attention mechanisms. The joint hashing framework innovatively integrates: 1) a bipolar quantization function generating compact binary codes preserving cross-modal correlations, and 2) a multi-level similarity fusion strategy dynamically weighting signal-image hash matrices and their fused representation. The hybrid loss combines contrastive, quantization, and similarity preservation objectives with staged optimization. Evaluated on a 16-class multimodal dataset (visible images and radiation signals) from Yantai Port, China, the method achieves 97.6% mAP@50, outperforming baselines by 13.1%. This work provides an effective solution for cross-modal retrieval in marine heterogeneous data environments.