近年来,探地雷达(Ground Penetrating Radar,GPR)凭借其非破坏性检测与高效率优势,已成为植物根系研究的重要技术手段.然而,从GPR数据中快速准确识别根点双曲线反射信号并反演波速信息仍面临多重挑战,包括噪声干扰、双曲线形态复杂及实...近年来,探地雷达(Ground Penetrating Radar,GPR)凭借其非破坏性检测与高效率优势,已成为植物根系研究的重要技术手段.然而,从GPR数据中快速准确识别根点双曲线反射信号并反演波速信息仍面临多重挑战,包括噪声干扰、双曲线形态复杂及实测数据样本稀缺等挑战.针对上述瓶颈,本文提出一种改进型深度学习模型:Yolov4-HPV(Hyperbolic Position and Velocity),基于Yolov4目标检测框架创新性集成关键点定位算法,通过解析双曲线特征中的五个关键几何点,实现波速参数的辅助高精度计算.为解决实测训练数据不足的难题,本研究设计了一种基于gprMax正演模拟软件的仿真根系GPR数据生成框架.首先,采用Merge策略优化仿真图像批量生成流程;其次,利用Multi-CycleGAN策略对图像进行风格转换,显著提升仿真数据的多样性和模型的泛化能力.结果表明,Yolov4-HPV模型能够准确识别根点双曲线反射信号并估算波速,关键点法可进一步提高波速估算精度.在测试集中,Yolov4-HPV和关键点法的波速估算相对均方根误差(RRMSE)分别为4.76%和3.43%;在控制实验中,根埋深定位的平均绝对误差分别为4 cm和3 cm,平均相对误差分别为15%和11%,充分说明了上述方法在根系定位和波速反演中的高精度和鲁棒性.本研究提高了GPR自动化根系识别和波速估算的效率,显著降低了仿真图像生成的时间成本,为拓展GPR根系研究提供了新途径.展开更多
Roots play a key role in ecosystem functioning as they transfer water and nutrients from soil to plants. Traditional methods for measuring roots are labor-intensive and destructive in nature, which limits quantitative...Roots play a key role in ecosystem functioning as they transfer water and nutrients from soil to plants. Traditional methods for measuring roots are labor-intensive and destructive in nature, which limits quantitative and repeatable assessments in long- term research. Ground-penetrating radar （GPR） provides a non-destructive method to measure plant roots. Based on the superiority of GPR with 2 GHz frequency, we developed a new, practical method to estimate root biomass. First, average root matter density was measured by collecting a small number of root samples. Second, under controlled, experimental conditions in a sandy area, a root diameter estimation model base on GPR was developed from which root diameter was estimated. Third, root volume was calculated using the estimated root diameter and assuming the shape of roots to be cylindrical. Finally, root biomass was estimated by averaging root matter density and root volume. Results of this study suggest the following： （1） the density of coarse roots with diameters greater than 0.5 cm is relatively uniform; （2） a new wave shape parameter, AT, extracted from profile data of 2 GHz frequency antenna is independent of root depth, thus enabling the construction of a root diameter estimation model with high accuracy; and （3） results of a field experiment demonstrated the GPR-based method to be feasible and effective in estimating biomass of coarse roots. These findings are helpful for improving GPR-based root diameter and biomass estimation models and suggest the potential of GPR data in studying root systems.展开更多
文摘近年来,探地雷达(Ground Penetrating Radar,GPR)凭借其非破坏性检测与高效率优势,已成为植物根系研究的重要技术手段.然而,从GPR数据中快速准确识别根点双曲线反射信号并反演波速信息仍面临多重挑战,包括噪声干扰、双曲线形态复杂及实测数据样本稀缺等挑战.针对上述瓶颈,本文提出一种改进型深度学习模型:Yolov4-HPV(Hyperbolic Position and Velocity),基于Yolov4目标检测框架创新性集成关键点定位算法,通过解析双曲线特征中的五个关键几何点,实现波速参数的辅助高精度计算.为解决实测训练数据不足的难题,本研究设计了一种基于gprMax正演模拟软件的仿真根系GPR数据生成框架.首先,采用Merge策略优化仿真图像批量生成流程;其次,利用Multi-CycleGAN策略对图像进行风格转换,显著提升仿真数据的多样性和模型的泛化能力.结果表明,Yolov4-HPV模型能够准确识别根点双曲线反射信号并估算波速,关键点法可进一步提高波速估算精度.在测试集中,Yolov4-HPV和关键点法的波速估算相对均方根误差(RRMSE)分别为4.76%和3.43%;在控制实验中,根埋深定位的平均绝对误差分别为4 cm和3 cm,平均相对误差分别为15%和11%,充分说明了上述方法在根系定位和波速反演中的高精度和鲁棒性.本研究提高了GPR自动化根系识别和波速估算的效率,显著降低了仿真图像生成的时间成本,为拓展GPR根系研究提供了新途径.
基金supported by National Natural Science Foundation of China (Grant No. 41001239)the Program for New Century Excellent Talents in University,Ministry of Education of China
文摘Roots play a key role in ecosystem functioning as they transfer water and nutrients from soil to plants. Traditional methods for measuring roots are labor-intensive and destructive in nature, which limits quantitative and repeatable assessments in long- term research. Ground-penetrating radar （GPR） provides a non-destructive method to measure plant roots. Based on the superiority of GPR with 2 GHz frequency, we developed a new, practical method to estimate root biomass. First, average root matter density was measured by collecting a small number of root samples. Second, under controlled, experimental conditions in a sandy area, a root diameter estimation model base on GPR was developed from which root diameter was estimated. Third, root volume was calculated using the estimated root diameter and assuming the shape of roots to be cylindrical. Finally, root biomass was estimated by averaging root matter density and root volume. Results of this study suggest the following： （1） the density of coarse roots with diameters greater than 0.5 cm is relatively uniform; （2） a new wave shape parameter, AT, extracted from profile data of 2 GHz frequency antenna is independent of root depth, thus enabling the construction of a root diameter estimation model with high accuracy; and （3） results of a field experiment demonstrated the GPR-based method to be feasible and effective in estimating biomass of coarse roots. These findings are helpful for improving GPR-based root diameter and biomass estimation models and suggest the potential of GPR data in studying root systems.
