3-D Reconstruction Framework for Tree Roots based on Permittivity Inversion and Feature-Matching Interpolation

Abstract

Abstract Background and Aims The structure of tree root systems is crucial for their growth, health, and stability. However, traditional methods for detecting root systems commonly face challenges such as computational complexity, low precision, and inadequate imaging visualization. This study proposes a method for the 3-D reconstruction of tree root systems, utilizing ground-penetrating radar (GPR) data coupled with deep learning-based inversion of 2-D permittivity distributions and feature-matching interpolation. Methods Our approach involves the inversion of 2-D permittivity distributions from GPR scan data using deep learning techniques to obtain cross-sectional parameter information of the root systems. We enhance the imaging accuracy of root identification through cluster analysis and threshold segmentation. Furthermore, by integrating target root detection, parameter calculation, and feature-matching interpolation, we reconstruct the 3-D structure of the root systems. Results In the test of simulated data, the method proposed in this paper shows smooth results in interpolation reconstruction and matches the actual values to a high degree. In the validation of actual data, FMIR successfully reconstructed the 3D dielectric constant model of the tree root system with larger diameters in the four main regions, and the reconstructed tree root system was in good agreement with the actual excavated root system. Conclusion The effectiveness and accuracy of this method in reconstructing 3-D permittivity models of tree root systems are validated through simulated and actual testing data experiments. It offers new possibilities for research and applications in root structure analysis.

Code and data availability

The supplied preprint blocks describe GPR-based 3-D root reconstruction using MPPINet and FMIR, but contain no data availability, code deposit, or repository statements. No public dataset, images, code, or model checkpoints specific to this paper are identified.

No evidence-backed public reproduction asset is currently recorded.