One of my favorite paper is from ISPRS Laserscanning '09.....very useful and innovative ideas, we need that kind of work, and not the discussions of RMS faults....how can I correct another mm-s....
A DECIDIOUS-CONIFEROUS SINGLE TREE CLASSIFICATION AND INTERNAL
STRUCTURE DERIVATION USING AIRBORNE LIDAR DATA
C. Ko, G. Sohn, T. K. Remmel
"ABSTRACT:
This project has two main purposes; the first is to perform deciduous-coniferous classification for 65 trees by using the leaf-on single flight LiDAR data. It was done by looking at the geometrical properties of the crown shapes (spherical, conical or cylindrical), these shapes were developed by a rule-driven method Lindenmayer Systems (L systems). Two more parameters that are data driven (convex hull analysis and buffer analysis) were developed to further capture the geometrical differences between deciduous and coniferous trees. Proposed methods are scale independent and arithmetically simple, they were developed simply by looking at the geometrical differences between the two types of trees. The classification rate was cross-validated and trees are 85% - 88% correctly classified. The second part of the project is to derive the internal structures of the LiDAR tree according to the results obtained from the first part. Internal structures include bole and branches; the location and orientation of the bole was done by connecting the geographic centres of horizontal slices of the tree. The branches were derived by k-means clustering techniques, different types of trees will yield a different type of branching structures for better visualization.
DISCUSSION / CONCLUSION
There are two major types of LiDAR systems for research and commercial use, full waveform and discrete returns. This paper has used only the discrete returns of the range data, and by studying the geometry of the crown shape properties, we classified the different crowns into two major classes, deciduous and coniferous. From the classified results, bole and branching structures were reconstructed according to the type of tree. The shape of the tree crown is inherited in the gene (adaptation) and therefore a certain species will have the similar crown shape and branching structures. The other factor affecting crown shape and branching structure is the growing strategies, which is adopted by the growing neighbour environment and those are more difficult to model (Horn, 1971). As a result, crown geometry is believed to be an important piece of information for species classification. By using just the three geometrical shapes (sphere, cone and cylinder), results were improved from 65% to 67% when the outliers were removed. If other parameters are included (area to volume ratio of convex hull and point to polygon buffering analysis), results were improved from 85% to 88%. Using crown shapes to classify trees is an intuitive method, but in this study it did not show promising results. By looking at the other geometrical properties, the results for classification increased considerably. Although different from what was expected, it is still believed crown shape and internal structure are good indicators for classifying trees, and future studies should be conducted in this direction. This method of classification is quite simple to produced and arithmetically easy. Tree bole and branches reconstructions are for visualization, but can also be used to study growth behaviour and to provide insight regarding why trees grow in a particular directions. These results are useful in many types of studies. For example, it can be used to study the potential hazards of a tree growing into structures, by classifying trees into deciduous and coniferous provide a better growth estimates."
No comments:
Post a Comment