FDRMFL:Multi-modal Federated Feature Extraction Model Based on Information Maximization and Contrastive Learning