A Hybrid YOLO–Mamba Deep Learning Framework for Real-time Ballistic Limit Velocity Prediction with Multiphysics-coupled Feature Fusion

Abstract To support dynamic penetration decision-making, the millisecond-level real-time response requirement of missile attitude control systems requires efficient ballistic limit velocity (BLV) prediction models. This study proposes a deep learning model based on a YOLO–Mamba hybrid architecture, which achieves the adaptive modeling of multiphysical field coupling effects through feature cross-modules and polynomial expansion. The global feature extraction capability of YOLO and the local temporal modeling of Mamba synergistically enhance multiscale feature capture. In experiments, the model’s inference speed is 1.3 times greater than that of traditional methods, and its prediction error on ballistic datasets is reduced by 32.5–47.8% compared to those of SVM/random forests while maintaining a generalization accuracy of over 92% in data-scarce scenarios. The proposed model serves as a high-precision tool for the optimization of protective materials, and the YOLO–Mamba hybrid architecture offers a novel approach to data-driven modeling of complex impact dynamics problems.