Adaptive SNR-Driven Fusion of CNN-GRU Networks for Enhanced Vehicle Velocity Estimation

Vehicle velocity estimation in the Internet of Vehicles (IoV) domain faces challenges due to varying signal conditions and computational constraints. Traditional approaches either suffer from performance degradation in low Signal-to-Noise Ratio (SNR) environments or incur substantial computational overhead. This letter introduces an adaptive fusion framework that combines Convolutional Neural Network (CNN) and Gated Recurrent Unit (GRU) architectures through SNR-driven dynamic model selection. A novel hysteresis-based transition mechanism ensures estimation continuity during model switching. Experimental validation using multiple datasets demonstrates that our approach achieves a 27% reduction in Mean Square Error compared to state-of-the-art methods while requiring 42% less computational resources. The system maintains sub-meter-per-second accuracy across SNR ranges from 0-35 dB, making it suitable for resource-constrained IoV applications operating in dynamic environments.