Trajectory similarity measurement is critical for smart cities, pandemic prevention, and public safety, but existing methods suffer from inefficiency and inaccuracy under dynamic sampling frequencies and large-scale data. This paper proposes a dynamic frequency-aware framework leveraging temporal proximity constraints and co-occurrence time ratio calculations, alongside encoder-based trajectory feature learning and acceleration mechanisms, to enhance robustness and computational efficiency. The framework mitigates sampling uncertainty by weighting trajectory points via temporal proximity, generating spatiotemporally consistent features. An encoder extracts high-dimensional embeddings optimized through spatial-aware loss functions and supervised decoder. Evaluations on simulated and real-world datasets demonstrate 70–100 × faster computation and superior accuracy (Rank 1, Rank 5, Rank 10), offering a scalable solution for large-scale trajectory analysis.