Three-dimensional Full-resolution Imaging And Root Parameters Estimation Using Ground Penetrating Radar

Ground Penetrating Radar(GPR) is an important geophysical technique for probing subsurface objects non-invasively using electromagnetic(EM) wave. As the fields of GPR application extending, GPR has been used more frequently in three-dimensional(3D) reconstruction and characteristics analyzing of subsurface targets, which is urgent to develop the advanced 3D GPR imaging technique and enhance the ability of data interpretation. Root biomass estimation belongs to the field of ecological studies, but GPR has shown its great potentials on tree root detection and root biomass estimation. The research object of this dissertation aims to subsurface root system, and the further studies of full-resolution 3D imaging and root biomass estimation based on GPR are selected as the core topics of this dissertation.First of all, the full-resolution 3D imaging method based on rectangular grids sampled GPR data is investigated. The core of full-resolution imagine is that the imaging result can recover the undistorted 3D construction of subsurface targets. Accordingly, Nyquist spatial sampling criteria of rectangular grids sampling is proposed in advance, and two practical estimating methods about Nyquist sampling intervals are also provided subsequently. As we know, the conventional 2D rectangular grids sampling needs a long time, and has a low positioning accuracy. In order to overcome this problem, an advanced 3D-GPR system with high positioning accuracy was used in 2D GPR measurement. Due to the depolarized effect of cylindrical-shape root in EM probing, the 3D imaging results based on rectangular grids sampled GPR data suffers from a serious degradation because the commercial GPR system usually employs the linearly-polarized antenna. For solving this problem, this dissertation proposes a new circular survey using GPR which effectively removes the depolarized effect of root in the 3D imaging results.Secondly, a new radial sampling based on GPR circular survey is investigated, the geometric model of which has defined in detail, and the Nyquist spatial sampling criteria related with this new sampling method has also been derived in theory. Based on the radially sampled GPR data, there are four different routines to implement full-resolution 3D imaging which has been derived theoretically one by one in this dissertation. According to the imaging results using synthetic GPR data, the one of the proposed imaging methods called the filtered-backprojection(FBP) method in wavenumber domain has both the high efficiency and the high accuracy of 3D imaging. Additionally, a rotatable GPR system was developed for obtaining the radially sampled GPR data. The field measurement of tree root using the rotatable GPR system testifies the effectiveness of the proposed radial sampling method and the FBP method.In the studies of root biomass estimation using GPR, due to the positive correlation between the coarse root diameter and root biomass, the hyperbolas reflected from the coarse root are feasible to estimate the root diameter which means estimate the biomass indirectly. The proposed root diameter estimation is built on the hyperbola fitting model with the separate transmitting/receiving(T/R) antennas. According to the least squares method, the improved Levenberg-Marquardt method is used to estimate the four unknown parameters successively in the hyperbola fitting model, and the estimated root diameter is finally obtained. Besides, an inverse Q filtering method is used in the A-scan processing which can correct the distorted waveform in the loss and dispersive soil. This method enhances the accuracy during the samples extraction on the hyperbola, and reduce the errors between the estimated root diameter and the true value.At last, the 3D full-resolution GPR data of larch tree root acquired by 3D-GPR system in the field is investigated. According to the spatial distribution of root system in the field, a coarse root detection based on the iterative searching is proposed in this dissertation. This proposed detection method starts from the location of the taproot, and then searches and detects the root candidate pixels towards the surrounding space of the taproot in the migrated GPR data cube. Besides, the linear and slender shape of root is considered as an important feature in the detection, so this detection method can avoid the interference of the strong and large-area pixels and find out the true coarse root distribution. Based on the correct detection results of coarse roots, a linear regression model is built using a GPR index(pixels within the threshold range), and it can also obtain a rough estimated biomass in total according to the accumulating pixels within the threshold range. Furthermore, a new index called magnitude width is proposed and defined in the GPR B-scan data, which can estimate the coarse root diameter directly. The extracted GPR samples of the detected roots testify the effectiveness of this new index and list their estimation errors.