| Forests are basis of human beings and various biological survival and development, which has a complex internal structure and significant ecological processes. Meanwhile, Forest ecosystem is the largest terrestrial ecosystem system. Therefore, It is current important task to investigate and dynamic monitor the forest resource, and establish forest vegetation classification system is the premise demand. In this study, Based on MODIS image data in the year of 2002 and taken Sichuan Province as the test area, according to the existing forest classification system, combined with the needs of forest ecosystem service function, we establish the forest vegetation classification system which faces the needs of evaluation of forest ecosystem services function. In the Classification process, we fully consider the forest's physical appearance and process characteristics.This paper mainly take six kinds of vegetation types in Sichuan Province include evergreen broadleaved forest, deciduous broadleaved forest, mixed conifer and broadleaf forest, low mountains evergreen coniferous forest, middle-mountains evergreen coniferous and mountainous evergreen coniferous forest as stuied object. Firstly built the Spectrum procedural knowledge of the typically land class including agricultural land, grassland, woodland and typical tree species including alders, cyclobalanopsis, oak class, pinus yunnanensis, fir, spruce on the 1-7 bands composed the MODIS data and the features of NDVI value. And we compared the three kinds of classification methods which respectively is time series of main component NDVI decision-making tree classification, the decision tree classification method of multi-temporal NDVI Images, and the decision tree category based on the spectral procedural knowledge. The results show that the overall accuracy is percentage of 88.5; the Kappa coefficient is 0.87 by the decision tree classification method based on spectrum process knowledge. Nevertheless, the other classification methods'overall accuracy are both about percentage of 78, and the Kappa coefficient are all about 0.75. It proved that the classification method based on spectral procedural knowledge is a best method,and it is more improved on the user accuracy and map-maker accuracy of the forest types. Finally, this method is applied in the classify the forest vegetation in Sichuan Province of 2008. And though dynamic analysis in several of forest vegetation in Sichuan Province from 2002 to 2008, it's revealed that forest vegetation types turn piece, turn to area, and the distribution in the spatial patternThe innovations in this paper are showed in the following aspects: First, in respect of moderate or low resolution of remote sense images ,based on the pseudo invariant features relative radiation correction,It is established that improved arithmetic relative to multidate radiation correction ,and built a realization module. Second, discover the spectral procedural knowledge of the different land types and typical tree were analyzed, establish forest characteristics (B1-B3-100*NDVI) and Brown cover characteristics (NDVI-B4/B6) .put forward forest remote sense image classification method .At last, apply this method on the classy of forest vegetation,and reveal it's dynamic variation process. |
