Estimation of Root Diameter in GPR Images Via CycleGAN-Guided Multi-Objective Integration Neural Network

Xihong Cui,Shupeng Li,Luyun Zhang
DOI: https://doi.org/10.1109/igarss53475.2024.10640543
2024-01-01
Abstract:The diameter of roots is crucial to study the geometric characteristics of subsurface root structure. Due to the invisibility of the underground root system, its diameter parameters are also difficult to obtain directly. Therefore, estimating the diameter based on ground-penetrating radar B-Scan images is challenging. In this study, CycleGAN-guided multi-objective integration neural network (CMI-Net) was constructed to simultaneously extract root diameter and location. The CMI-Net includes two sub-networks: CycleGAN and YOLOv4-Hyperbolic Position and Diameter (YOLOv4-HPD). The former ensures that the YOLOv4-HPD model, trained on the simulated datasets, can be used in the field environment. The latter can accurately identify the root objects and estimate the root diameter. The performance of the model was evaluated using simulated test dataset and field control experimental dataset. The model’s availability in estimating root diameter was demonstrated by the experimental results.
What problem does this paper attempt to address?