Objective In order to achieve the quantitative estimation of the relative permittivity of tree root systems, an estimation method based on radar wave signals and tree root system parameters was proposed.

Method Firstly, the propagation paths of radar waves in the root systems of trees with different radii and relative permittivity were simulated, and the A-scan curves at the hyperbolic vertices in the ground-penetrating radar images were obtained through the orthogonal analysis. Then, the target amplitude parameter ΔF, which is related to the relative permittivity of the root system, was extracted from the A-scan curves. Finally, the dataset for relative permittivity estimation was established by combining the relative permittivity of the soil, the radius of the root system, and the depth of the root system. The estimation models were established based on partial least squares regression (PLSR), back propagation (BP) neural network, and particle swarm optimization-back propagation (PSO-BP) neural network, respectively, and the estimation accuracies of the three models were compared and analysed.

Result (1) In the simulation experiment, the root mean square error and average absolute error of PSO-BP neural network estimation model were 0.701, 0.255, and the R² was 0.990, and all indexes were better than PLSR and BP neural network estimation model. (2) In the field pre-embedding experiment, the estimation accuracies of PSO-BP neural network estimation model were all better than those of PLSR and BP neural network estimation models, with a maximum absolute error of 3.16 and the whole average relative error of 10.88%.

Conclusion Using the dataset established by the target amplitude parameter ΔF, soil relative permittivity, root radius and root depth extracted in this study, combined with PSO-BP neural network estimation model, an accurate estimation of the relative permittivity of the root system of trees can be achieved, which is of great significance for assessing the growth status and health of the tree root systems.