Ground Penetrating Radar (GPR) has become a pivotal tool in plant root studies owing to its non-destructive nature and operational efficiency. However, rapid and precise identification of root's hyperbolic reflections and wave velocity estimation from GPR data remain constrained by challenges including noise interference, complex hyperbola morphologies, and limited field-measured datasets. To address these limitations, this study introduces Yolov4-HPV (Hyperbolic Position and Velocity), an enhanced deep learning model built upon the Yolov4 framework. The proposed methodology integrates a key-point detection algorithm that identifies five characteristic points of hyperbolic signatures, enabling supplementary wave velocity calculations with improved accuracy. To mitigate training data scarcity, a synthetic data generation framework was developed using gprMax forward simulation software. The framework employs two strategies: (1) a Merge protocol to streamline simulated image synthesis, and (2) a Multi-CycleGAN approach for style transfer, substantially augmenting dataset diversity and model generalizability. The results show that Yolov4-HPV's capability to detect hyperbolas and estimate wave velocities with high precision. The key-points method further improves the accuracy of wave velocity estimation. The key-points method further reduced RRMSE of wave velocity estimation to 3.43%, outperforming Yolov4-HPV's 4.76% for the testing datasets. In control experiments, the average absolute errors of root depth positioning were 4 cm and 3 cm, with average relative errors of 15% and 11%, respectively, confirming the model's high accuracy and robustness. This work advances GPR-based root investigation by enhancing automatic target identification and wave velocity quantification while optimizing computational cost, offering significant methodological improvements for ecological and hydrological applications.