Research On The Large Area Remote Sensing Discrimination Of Main Non-Timber Forest Surrounding Tarim Basin In Southern Xinjiang Based On Multi-Source Data

Since the 10 th Five-Year Plan, the Tarim Basin in Southern Xinjiang has been rapidly developed to be a main production place of characteristic forest fruit industry possessing variety adaptability, regional scale,environmental adaptability and economic advantages with its wide geographical space, the superior water,soil, light, heat conditions and rich labor resources. Therefore, to understand and grasp the present situation or variable information of the large area non-timber forest resources has an important significance to accelerate the forest fruit industry informatization. However, the traditional forest resource survey and monitoring methods are difficult to meet the urgent need that rapidly and accurately grasp the layout and scale of large scale characteristics forest fruit resources in present construction process of forest fruit industry informatization. Given that, based on different remote sensing images, this paper had an exploratory research on large area remote sensing discrimination method of the main non-timber forest surrounding Tarim Basin in southern Xinjiang, in order to provide a technical reference for the investigation division, resource management and other production practices of non-timber forest resources.The conclusions are as follows:Through different ways of dimension reduction, the paper screened the Hyperion image for sensitive band combinations, and evaluated the band combinations and dimension reduction methods by Euclidean distance. The results show that four kinds of band combinations(96-193-213, 33-106-193, 55-33-25 and55-158-33) can distinguish Ziziphus jujuba, Malus pumila and Pyrus bretshneideri trees, and both the separability of 55-33-25 and 55-158-33 combinations for three species are better than 96-193-213 and33-106-193 combinations. The larger mean Euclidean distance and smaller standard deviation among three tree species indicate that the band index method based on the corresponding bands of TM image is a sensitive band selection method of the identification of fruit trees in southern Xinjiang.Identifying the main non-timber forest species by using three kinds of supervised classification methods: neural network, maximum likelihood method and support vector machine, the results show that four band combinations has good recognition effect overall to different non-timber forest species, which classification accuracy are above 75% and Kappa coefficients are above 0.61. Considering the overall classification accuracy, Kappa coefficient and single tree species classification accuracy, the paper determines the most sensitive band combination is 55-158-33, which has a best identification among three main non-timber forest species. At the same time, support vector machine can be used as the best classification method to identify the main non-timber forest surrounding Tarim Basin in Southern Xinjiang.Using multi-resolution segmentation of object-oriented method to QuickBird remote sensing image,this paper determined the best parameters of image segmentation, which optimum scale is 122, band weight ratio is 1:3:1:3, shape factor is 0.9, and compactness is 0.1, and obtained a better boundary information extraction result of non-timber forest garden. By using GIS spatial analysis method, the automatic division of subcompartment based on information is realized, which provides a technical reference for the investigation and division, resource management and other production practices of non-timber forest resources.