|
|
1.INTRODUCTIONForest canopy height, as an attribute of forest vertical structure, is an important parameter used to estimate stand stock, stand productivity and biomass1. Polarized Interferometric Synthetic Aperture Radar (PolInSAR) is a new remote sensing technique that has the properties of both polarized SAR sensitive to the direction and canopy of the vegetation scatters and PolInSAR sensitive to the height and distribution of the vegetation bodies, which can not only improve the accuracy of topographic measurements but also obtain the relevant physical parameters of scattering mechanisms, especially for tree height2. It is of great significance in forestry resource management and forestry investigation and research, especially the forest parameters provide the possibility of high precision inversion of tree height and various other parameters of ground features3. The forest height estimation methods can be divided into two types: high-dimensional data nonlinear optimization4 and polarization mechanisms-based algorithm5. The disadvantages for high-dimensional data nonlinear optimization methods are also obvious: firstly, the computational effort is huge; secondly, the dependence of the algorithm performance on the initial value of the iteration is too great, which is easy to fall into the trap of local optimal solutions6. Then there are problems such as inaccurate estimation of polarization parameters for those polarization mechanisms-based algorithm7. Therefore, in this paper, a PSO-improved algorithm based on the Cloude three-stage strategy, is introduced to improve its coherent region estimation process to improve the accuracy of forest height inversion. The experimental results are partially carried out using TerraSAR-X data for the inversion experiments and combined with the field measured heights for statistical analysis. 2.BASIC THEORYRandom Volume over Ground (RVoG) model is one of the most widely used models in polarization interference. The model contains two layers of scatter and ground structures, and the model describes a randomly distributed forest cover model with a thickness of HV polarization in a two-layer medium model8. Under the RVoG model, the forest height can be understood as the thickness of the random body. The relevant equation in the RVoG model is as follows. where The three-stage forest height inversion algorithm is a classical forest height inversion algorithm proposed by Cloude8, which is completed by three steps: linear fitting, surface phase solving, and forest height inversion. As shown in Figure 1, the observations of the six polarization channels can be arranged in the complex plane, according to the variational form of the RVoG model. The line represents the phase of the surface reflection when the complex coherence coefficient varies between 0 and 1. Therefore, the farthest intersection point in the complex plane can be consider is the complex coherence of the ground layer Φ0. After estimating the line of the complex coherence coefficient, the feature can be used to complete the forest height inversion that the complex coherence coefficient γv is determinated by both the forest height hv and the extinction coefficient σ5. 3.FOREST HEIGHT INVERSION ALGORITHM IMPORVED BY PSO FOR POLINSAR DATADue to the existence of temporal decorrelation and some other unfavorable conditions, the inaccurate estimation of the “pure body decorrelation coefficient” is the biggest problem faced by the three-phase method9. Therefore, to improve the accuracy of the surface phase, the maximum coherence magnitude difference criterion and the maximum interference phase difference criterion are introduced in this paper, i.e., the polarization state corresponding to the maximum coherence magnitude difference is selected as the optimal polarization state, as in equation (2): where |γ(ωi)|, |γ(ωj)|are the amplitude of the corresponding complex coherence coefficient of polarized interference vectors ωi, ωj respectively. Meanwhile, to improve the efficiency of parameter estimation, the PSO algorithm was introduced to the coherent region estimation process in this paper10. By integrating equations (1) and (2), the objective function could be designed for the particles which is called “fitness”. And the solution element which is named “position” in the feature space was combined with the coherence coefficient γv, extinction coefficient σ and the forest height hv 11. The flow of the PSO-improved forest height inversion method is as follows:
4.EXPERIMENT RESULTIn order to verify the performance of the forest height inversion algorithm on the real data, the dual-star fully PolInSAR images of TerraSAR-X were used for the experiment which provided by the German Aerospace Center (DLR). The Pauli color composites of the master and salve images were generated after reading the single-view complex data of the primary and secondary images separately as shown in Figure 2. The forest height inversion results of the three-stage algorithm and the PSO improved algorithm were compared with the measured data. After removing the outliers, tree height inversion scatter plots were produced and the following results were obtained as shown in Figure 3. The statistical analysis of the height inversion results of the two methods is shown in Table 1. Table 1.The precision of inversion result.
As shown by the tree height inversion scatter plot and the accuracy table of inversion results. Compared to the three-stage algorithm, the PSO-improved algorithm has a strong correlation with a correlation coefficient R of 0.643 and a root mean square error (RMSE) of 5.53, which has a significant improvement. 5.CONCLUSIONA novel intelligent algorithm for the forest height inversion problems with PolInSAR data was proposed in this paper, which effectively combines the RVoG model of PolInSAR data and the advantages of the PSO algorithm. In this method, the initial positions of the particles are calculated by estimating the line that represents the phase of the surface reflection with six polarization channels. Then the height inversion results are optimized with the PSO processing, to improve the correlation coefficient. The experimental results show that the method is effective in forest canopy height inversion with PolInSAR data. ACKNOWLEDGEMENTSYoung innovative talents project of general colleges and universities in Guangdong Province 2022 “Research on rice fine recognition method based on multi temporal sentinel-1 data”; The basic and applied basic research project of Guangzhou basic research plan, “Research on three-dimensional tracking and registration technology of multi person collaborative virtual simulation experiment platform” (20210201298), the special project for high-level talents of Guangdong industry and Trade Vocational and technical College “Research on UAV surveying and mapping virtual simulation training system”(2021-gc-07), and Guangdong Polytechnic of Industry and Commerce 2021 school level scientific research project “Research on forest height retrieval algorithm based on PolInSAR data”(2021-ZK-16) REFERENCESZhao, L., Chen, E., Li, Z., et al.,
“A new approach for forest height inversion using X-band single-pass InSAR coherence data,”
IEEE Transactions on Geoscience and Remote Sensing, 60
(99), 1
–18
(2021). https://doi.org/10.1109/TGRS.36 Google Scholar
Tang, F., Xie, J., Suo, Z., et al,
“X-band polinsar vegetation canopy height inversion strategy based on frequency segmentation,”
in 2020 IEEE Inter. Geoscience and Remote Sensing Symp,
180
–183
(2020). Google Scholar
Liu, F. L., Wu, C., Tang, X. Y., et al,
“Research on forest height inversion algorithm of three-stage full-polarization SAR data based on improved coherence coefficient,”
in 2019 Inter. Conf. on Informatics, Control and Robotics,
182
–186
(2019). Google Scholar
Li, T. W., Liang, D. N., Huang, H. F. and Zhu, J. B.,
“A BP neural-network based method for vegetation height inversion of the polarimetric interferometric SAR,”
Journal of National University of Defense Technology, 32
(3), 60
–64
(2010). Google Scholar
Managhebi, T., Maghsoudi, Y. and Zoej, M.,
“An improved three-stage inversion algorithm in forest height estimation using single-baseline polarimetric SAR interferometry data,”
IEEE Geoscience & Remote Sensing Letters, 15 1214
–1218
(2018). https://doi.org/10.1109/LGRS.2018.2830744 Google Scholar
Li, X. W.,
“Research on Polarization Interference SAR Information Extraction Method and Its Application,”
Chinese Academy of Sciences, Doctor’s Thesis,
(2002). Google Scholar
Pardini, M., Qi, W., Dubayah, R., et al.,
“Exploiting TanDEM-X Pol-InSAR data for forest structure observation and potential synergies with NASA’s global ecosystem dynamics investigation lidar,”
in Proc. of 11th European Conf. on Synthetic Aperture Radar,
1
–6
(2016). Google Scholar
Papathanassiou, K. and Cloude, S. R.,
“Single-baseline polarimetric SAR interferometry,”
IEEE Transactions on Geoence & Remote Sensing, 39
(11), 2352
–2363
(2001). https://doi.org/10.1109/36.964971 Google Scholar
Long, J. P., Liu, F. and Duan, Z. G.,
“A new method of coherence optimization based on the phase and coherence magnitude in polarimetric SAR interferometry,”
Acta Geodaetica et Cartographica Sinica, 46
(1), 74
–82
(2017). Google Scholar
Kennedy, J. and Eberhart, R. C.,
“Particle swarm optimization,”
in Proc. of Inter. Conf. on Neural Networks,
1942
–1948
(1995). Google Scholar
Zhu, T., Gao, Z., Huang, T., et al.,
“Applying particle swarm intelligence in PolSAR image clustering,”
in Journal of Physics: Conference Series,
012064
(2021). Google Scholar
|
