Xie Donghui

Hierarchical support vector machines

The speed and accuracy of a hierarchical SVM (H-SVM) depend on its tree structure. To achieve high performance, more separable classes should be separated at the upper nodes of a decision tree. Because SVM separates classes at feature space determined by the kernel function, separability in feature space should be considered. In this paper, a separability measure in feature space based on support vector data description is proposed. Based on this measure, we present two kinds of H-SVM, binary tree SVM and k-tree SVM, the decision trees of which are constructed with two bottom-up agglomerative clustering algorithms respectively. Results of experimentation with remotely sensed data validate the effectiveness of the two proposed H-SVM.

A new vegetation index and its principle and application

Aiming at shortages of current vegetation indices, we put forward Three-band Gradient Difference Vegetation In- dex(TGDVI), and established algorithms to inverse crown cover fraction and Leaf Area Index(LAI) from it. Theoretical analysis and model simulation show that TGDVI has high saturation point and the ability to remove the influence of background, and explicit functional relation with crown cover fraction and LAI can be established. We also theoretically analyzed why NDVI has low saturation point and indicate that relationship between Sim- ple Ratio Vegetation Index(SR) and LAI closes to piecewise linear instead of linear.

3D reconstruction of a single tree from terrestrial LiDAR data

Terrestrial LiDAR systems have received lots of attention on three-dimensional (3D) structure reconstruction for trees, especially on the branches skeleton generation. On this basis, a method is proposed to add leaves structures based on point density by dividing small cube in the canopy to reduce the influence of uneven distribution of point cloud, combining gap fraction model to retrieve leaf area of a tree using terrestrial LiDAR data. It is successfully applied to reconstruct 3D trees using points data simulated by ray tracing algorithm as well as field measured points data. The relative error of leaf area between reconstructed and real structure is less than 0.9%. Meanwhile, the most relative error of directional gap fraction is also less than 4.1%. The experimental results prove that the method has gotten a satisfied consistency on visual sense and quantitative evaluation between the 3D structure reconstructed and real structure.

BRF of the scene of corn simulated by radiosity-graphic combined model

With the development of remote sensing and the technology of computer, computer simulation models are paid more and more attention to research Bidirectional Reflectance Distribution Function (BRDF) of the Earth surface, which can describe vegetations with much more detailed structures and simulate the interaction between light and vegetations on the Earth more reality. As known to all, according to the principles of the BRDF models, physical models of vegetation in the field of remote sensing can be divided into three categories: Geometry Optical models (GO), Radiance Transfer models (RT) and computer simulation models. To understand the process and mechanism of the interaction between light and vegetations better, some computer simulation models are provided, for example, DIANA which is based on the method of Radiosity, RAYTRAN, based on the method of ray tracing and Monte Carlo and so on. In this paper, a model based on Radiosity method is used to simulate the bidirectional reflectance factor (BRF) of summer corn field. It includes three parts: modeling the 3D scene; calculating the radiostiy of the components in the scene; and then accounting BRF of the scene. In the paper, at first 3D structures of corns are reconstructed by an extended L-system based on the measurement in the site of Luancheng Hebei Province, China in 2000. Then the Radiosity- Graphic combined model is applied to simulate the light of the scene of corn, and BRF of the scene of corn will be computed and compared with the measured BRF in the station of Luancheng. Simulated and measured results are fitted very well. Then the hemispherical BRF of the scene are calculated. At last, after analyzing the process of simulation and the results, several advices to advance the Radiosity model are put forward.

Validation of BRF data based on computer simulation model

At present, modeling bi-direction reflectance of vegetation canopy scene, produced by computer simulation and a radiosity-graphics combined method (RGM), has already made preliminary progress. In order to qualify the simulated data, in this paper, taking winter wheat as an example, besides comparing computer simulated data with measured data, we also use anther method: regarding simulated data as the prior-data and condition data of SAIL model to inverse the structure parameters of vegetation canopy ,such as LAI、 ALA and component spectrum, which were the input parameters of computer simulation. Next, comparing these SAIL modeled data with the input data of computer simulation. Thus we can get the precision of LAI, and validate the reliability of BRF data based on computer simulation. On the other hand, evaluating the quality of algorithms and models used by computer simulation. Some conclusions are drawn from the investigation:(1) LAI is the important structure parameter for vegetation canopy scene, and we can control the whole scene and its radiation regime only using LAI;(2)BRF data based on computer simulation model is reliable and its precision can meet the actual demands. So these BRF data can be considered as measured data to carry on research under certain conditions. Keyword: BRF, LAI , SAIL model, Computer Simulation

Extracting city information in tm image using mixed decision tree method

As traditional classification methods use spectrum of objects only, they cannot distinct the same objects with different spectrum, and sometimes they will misclass the different objects into one class with the similar spectrum. In order to classify the different objects correctly, Mixed Decision Tree (MDT) method and Minimum Distance Texture Feature Vector (MDTFV) are presented in this paper. As a case study, TM image in Beijing City, including some parts of northern suburban in China, is selected. Considering the particularity of big city, lots of mixed pixels exist, we recognize not only pure pixels but also mixed pixels as a class for the result. Eight classes, including forest, grass, farm, water, building, useless, building and vegetation, useless and vegetation, are gotten in the research region. At last all the classes are overlaid into an image to get the classification map. The classification accuracy is up to 97.25% and Kappa coefficient reaches 0.9612, which is improved greatly than that of using spectral method only.

Validation of scale effect based on computer simulation model

With the developing of satellite technology, more and more sensors, based on various resolutions and functions, are launched to the sky to perform their missions. Enormous data are sent back to the Earth stations. In deriving surface parameters using these remotely sensed data, the transportability of algorithms from one resolution to another often cause the scale effect because of the surface heterogeneity on the Earth which can induce the change of reflectance. The problem, that the change of reflectance data affected by discontinuity as part of surface heterogeneity impacts the retrieval of vegetation leaf area index (LAI), is addressed in This work. Two cases, inducing the scaling issue in deriving surface parameters of interest, are considered here. One is the discontinuity between contrasting cover types within a mixed scene, the other is the nonlinear relationship of NDVI and LAI. Therefore, it is necessary to apply the correction based on NDVI-LAI relationships to modify scaling problem. In the processing, considering the field of wheat, firstly a series of 3D scenes with wheat and soil mixed are made based on the field measurement; secondly, computer simulation model

A large scale LAI inversion algorithm

the method of radiosity, which can calculate the balance of light energy in the simulated scenes, is used to model BRF (bi-directional reflectance factor) of these scenes. If the NDVI-LAI relationship from homogeneous scenes can be taken as standard, the relationship from heterogeneous scenes will be modified according to contextural parameter. Some conclusions are drawn from the investigation: (1) different distributional contextures of vegetation even with the same LAI affect the reflectance heavily; (2) we compare the reflectance simulated by the method of radiosity with the mean reflectance calculated using the area-weighted linear relationship of reflectance from components, and find that the accuracy of the mean reflectance can be accepted so that the linear equation to calculate the reflectance of mixed pixels is reasonable to relate images with high and low resolution; (3) using contextural parameter for quantifying the scale effect can get promising results.

Application of BEPS model in estimating winter wheat yield in North China Plain.

In this paper, a new LAI retrieval method is developed. The algorithm borrows ideas from the principles and methods of ground LAI measurements, and adopts a new frame which differs from traditional remote sensing LAI inversion methods. The ground data acquired from two field experiments are used to validate the algorithm. In order to resolve the scale exchange problem between high resolution ground observation and low resolution remote sensing data, two high resolution remote sensing images almost having the same resolutions with ground measurements are used as transitions