Charge prediction is an important task in intelligent legal reasoning, aiming to automatically identify potential criminal activities from case descriptions. Existing methods often fail to capture the complex relationships in legal texts because they rely on simple feature modeling. In this paper we propose a new charge prediction method named Semantic Syntactic Fusion and Legal Constituent Elements Matching (SemSyn-LCE). It improves prediction accuracy by combining syntax and semantics features and extracting LCEs. To capturing cross-sentence dependencies which exhibit crucial contextual relationships, this method introduces a novel multi-hop syntactic dependency trees. It also identifies potential matching sentences for LCEs which contribute the most to charge prediction. Experimental results show that SemSyn-LCE outperforms existing methods, achieving higher prediction accuracy.