Automatic reconstruction of three-dimensional root system architecture based on ground penetrating radar

Ground penetrating radar (GPR) is widely used in root detection because of its advantages of nondestructive and periodic detection. Previously, root point positioning and root diameter estimation based on GPR have been proved to be effective, but there is still little research on automatic reconstruction of root system architecture (RSA) based on GPR. Good RSA reconstruction can effectively reflect the real growth of the root and its adaptability to the environment, so the purpose of this study is to vividly reconstruct RSA from the two aspects of topological structure and geometric features. Therefore, we used 900 MHz tree radar to scan the roots of a 9-year-old ash tree in the field, extracted the spatial coordinates of the root points from the detection results, and estimated the diameter of the root points using BP neural network. Thus, RSA was automatically reconstructed by using the root topological structure and geometric features reconstruction algorithm. By comparing the RSA reconstructed by the algorithm with the simulated RSA of the real root system, the accuracy of the reconstruction algorithm was quantitatively evaluated. The results showed that for roots with a diameter greater than 1 cm, the root length similarity between the algorithm reconstructed RSA and the simulated RSA is 76.2%, and the root topology similarity is 81.4%. The high similarity between the two indicates that the reconstruction of RSA based on GPR is effective and feasible. Moreover, the proposed algorithm in this paper enriches the three-dimensional (3D) visualization method of RSA. According to the method, the induction and regulation during the growth of the root system can improve its utilization efficiency of soil nutrients and water.