Forest Type Precise Identification Based On Hyperion Data

Tree species survey is an important part of forest resources inventory. Some forest resources parameters are estimated from the tree species. Therefore, identification of forest tree species is key to acquire forest resources information.Development of remote sensing techniques provides a new technical means for forest resources information acquisiton, particularly, make it possible for that in large areas. The commonly used multi-spectral remote sensing can not provide more detailed forest species classification due to its low spectral resolution. Hyperspectral remote sensing, started in the 80s of the 20st century, and broke through the spectral resolution bottleneck. In imaging process, the continuous spectrums of the targets are obtained, and the influence of other interference factors is suppressed in the spectrum space, Various land cover types with subtle spectral differences can then be accurately detected and separated. Thus, making it possible for the accurate identification of different forest species.Taking Zulai forest farm in Shandong province as test site, study was carried out on the forest species identification using spaceborne hyperspectral Hyperion images. Two processing steps were included for forest species identification. In step 1, a method combining the SVM and MLC was used for general land cover classification. In step 2, the Kernel Canonical Correlation Discriminant Analysis (KCCDA) method was used for the tree species identification within the general forest land cover from step 1. The main results and conclusion of this dissertation are as follows:1. A multi-classifier procedure was proposed for multi-spectral image classification, combining the SVM and MLC. The accuracy for general land cover classification using the proposed method reached 99.9% with Kappa coefficient 0.999. This means that the multi-classifier makes the classification more stable and improves the accuracy. The results indicate that this method has good application prospect in hyperspectral remote sensing image classification.2. A classification method based on the Kernel Canonical Correlation Discriminant Analysis (KCCDA) was proposed and used for tree species identification using the visual and near infrared bands of Hyperion data. The overall accuracy is 89.80% and the Kappa coefficient is 0.86. Results show that the kernel canonical correlation discriminant analysis method can identify tree species precisely and it has good capability to distinguish spectral similar forest trees.3. Software module was developed for the proposed classification methods, whicn laid a sound foundation for the application of the developed methods.