Large-scale three dimensional vehicle aerodynamics prediction poses critical computational challenges in modern automotive design, where traditional CFD methods require prohibitive simulation times that conflict with rapid design iteration demands. While recent neural operator approaches show promise, existing methods struggle with computational complexity in dense meshes and fail to preserve essential topological information when processing large-scale point clouds. We propose FCMO, a physics-aware neural operator that integrates fluid mechanics principles with selective state space modeling for efficient large-scale vehicle aerodynamics. FCMO introduces four synergistic components: FlowCurv Anchor Sampling that intelligently selects mesh nodes based on normalized local curvature and windward sensitivity. Additionally, dual-scale physics-aware position encoding with adaptive k-NN construction transforms 3D irregular meshes into causality-preserving sequences through feature-guided serpentine scanning. The model integrates a flow-aware Mamba processor incorporating selective mechanisms that dynamically modulate state transitions based on wall distance and flow characteristics. Finally, a physics-constrained decoder enforces conservation laws through mixed weighted interpolation. Extensive experiments on Ahmed-Body and DrivAerNet benchmarks demonstrate that FCMO achieves consistent state-of-the-art performance with 5.2% improvement in surface pressure prediction, 9.3% enhancement in wall shear stress estimation, and 11.4% boost in drag coefficient accuracy, while maintaining superior computational efficiency with 9.4% fewer FLOPs and 9.9% reduced memory usage compared to existing methods.

Finger vein recognition offers significant advantages in biometric authentication, while federated learning addresses data silo challenges in distributed environments. However, label noise issues severely impact recognition performance due to variations in data acquisition environments, fluctuations in user registration quality, and privacy constraints preventing centralized annotation review. Existing label noise research typically focuses on sample-level processing, overlooking quality variations between authentication systems and noise distribution characteristics across multiple source devices. This paper proposes FedRDA, a federated optimization framework that achieves precise identification and adaptive correction of noisy samples through a three-tier progressive mechanism. We first construct a hierarchical noise detection system that identifies label noise from both noisy client and noisy sample perspectives. Then, we design a dynamic pseudo-label learning module with an improved adaptive label ambiguation loss function that dynamically adjusts sample learning difficulty parameters and incorporates momentum update mechanisms, significantly enhancing model adaptability to label noise of varying difficulty, while integrating predictive uncertainty entropy with unsupervised consistency constraints for more accurate label correction. Finally, we propose an adaptive aggregation strategy based on distance awareness and gradient consistency metrics to address data isolation and label noise issues in distributed environments. Experiments on SDUMLA, MMCBNU_6000, FV-USM, and combined datasets demonstrate that FedRDA maintains high model accuracy even under high noise rate conditions, with approximately 14% accuracy improvement over existing methods. The proposed framework effectively mitigates the negative impact of label noise on model training, ensuring robust operation of finger vein recognition systems in practical distributed environments while protecting user privacy.