Confronted with the challenges of hierarchical decision-making and dynamic strategy adaptation in UAV swarm defense, this paper introduces HO-SAP, a novel algorithm that synergizes hierarchical and opponent modeling. Our approach structures swarm decisions into tactical and execution layers and employs a learning-based strategy estimation mechanism to infer opponent behavioral patterns from observation sequences, embedding this awareness into the value function for superior generalization. We train a BiLSTM-based classifier to identify opponent strategy types through supervised learning, enabling defenders to adapt their tactics accordingly. Results across diverse adversarial scenarios reveal that HO-SAP significantly outperforms existing methods in critical metrics like defense success rate and interception efficiency, exhibiting exceptional adaptability and tactical planning. The hierarchical model ensures rational and scalable decisions, while opponent modeling enables dynamic responses, establishing an effective learning framework for swarm adversarial tasks.