<p>Legal judgment prediction (LJP) is a fundamental task in legal artificial intelligence. LJP is used to predict relevant law articles, charges, and terms of penalty on the basis of the factual description of legal cases. Existing approaches fail to make full use of external legal knowledge, and rarely pay attention to judges who refer to historical similar cases before they make decisions in real judicial scenarios. To simulate this process, this study proposes a dual-stage framework that integrates law article semantics and similar case retrieval (LAS-SCR). This framework generates initial predictions by deeply fusing case facts with law article semantics through cross-modal attention and hierarchical feature augmentation and, targeting easily confused charges identified from the initial output, it corrects these predictions using contrastive learning-based similar case retrieval. Experiments on two real-world legal datasets demonstrate that our method achieves better results than several baseline models do. Furthermore, our method is effective at distinguishing easily confused charges.</p>