Extract Forest Vegetation Types Based On The Different Date Of TM Data

Have many extraction of forest vegetation types, and get a better accuracy. In this paper, Landsat-5TM data of growing season (July5.2006) and non-growing season (October25,2006) were used, by image pre-processing, study of best band combination, extract forest land and non-forest land in the study area and using the maximum likelihood method and vegetation index (NDVI and SRI) of extraction studies forest vegetation types (evergreen forest, deciduous forest), the results showed that:(1) In this study, the elimination hill shade of the best ratio band operation is Band4/Band3; the best band combination of different date Landsat-5TM images are234,245,345, visually best band combination is4(R)3(G)2(B).(2) SRI extracted the forest land and non-forest land of the growing season image overall accuracy was97.10%, kappa coefficient was0.9343, forest land1749108.42hm2, accounting for96.754%; non-forest land58685.58hm2, accounting for3.246%. The most similar with forest land1751535hm2, accounting for96.888%; non-forest land56259hm2, accounting for3.112%of Inventory Data.(3) When using the maximum likelihood method to extracted the evergreen forest and deciduous forest based on the different date TM images. Overall accuracy of non-growing season image was96.35%, kappa coefficient was0.8764, evergreen forest183414.96hm2, accounting for10.503%of forest land; deciduous forest1562974.94hm2, accounting for89.497%of forest land, It is the highest classification accuracy.(4) When using the NDVI and SRI extracted the evergreen forest and deciduous forest of non-growing season image, SRI extracted overall accuracy was95.46%, kappa coefficient was0.8500, evergreen forest182982.40hm2, accounting for10.478%of forest lands; deciduous forest1563407.5hm2, accounting for89.522%of forest land, It is the highest classification accuracy.