Extraction Of The Vegetation Information Based On Temporal And Spectrum Information

The vegetation plays an important part in the earth system. It influences the energy balance of earth climate system and has the important function on climate, water and biology circle. It is also the sensitive index of the infection which climate and human impact on environment. The dynamic surveillance on vegetation can reflect the trend of climate change.Vegetation classification with remote sensing images to achieve the extraction of vegetation cover information is a significant topic among remote sensing application. Because of"the same objects with different spectrum","the same spectrum in different objects"and"mixed pixel", remote sensing images classification with computer based on the traditional mathematical statistics restricts high-precision extraction of vegetation information. Vegetation grows in a cycle of a year. Difference types of vegetation respectively have its developmental rhythm or special growing law in this cycle. All of these can reflect different spectrum values. Therefore the study of vegetation classification based on temporal and spectrum information has theoretical basis, practical meaning and also applied potential.This paper chooses 85 Ground Truths which include different types of features in MODIS data, which spatial resolution is 1 kilometer, in 2003 about Shanxi province. The ETM data in 2002 the spatial resolution of that is 30 meters, digital Chinese vegetation type's map with the scale of 1:4000, 000, Shaanxi province vegetation partition map with the scale of 1:3000, 000 and land use database with the scale of 1:1000, 000 are the references. In this paper, decision tree classification method combined with NDVI time sequence curves of Ground Truths realized the extrzction of vegetation which based on temporal and spectrum information in Shanxi province.We use C4.5 decision tree algorithm for purposes of mining classification rules from training samples. Decision tree model of land information extraction was built from these samples through C4.5 algorithm, which integrates spectral and so on. And precision evaluation was carried out for the classified image by confusion matrix which confirmed the feasibility of this method. And compared with the traditional maximum likelihood classification method, the results show that the decision tree method can reflect the regional distribution of vegetation objectively and extract the vegetation information of study area accurately.This method depends on the characteristics of remote sensing images in different resolution. And time sequence curves are the basis for the classification of low resolution images to participate in the process of classification. This method provides a useful means for vegetation monitor in large area.