Extraction Information Technology Of Forest Vegetation Types Based On Object-based Method

Using high spatial resolution of the Quickbird remote sensing data and object-oriented method to extract the study area’s forest tree type niformation in Jiangle national forest farm of fujian province. When extracting, we fully consider the geometric,texture and context relationship information and join the custom soil adjustment vegetation index of diffierent parameters. Then we choose the multi-segmentation method and use membership function and the decison tree classification method to extract features. Finally, we realize the tree species extraction. The focus of the research and discuss are mainly the parameters of the multi-scale segmentation, feature information filtering and the object-oriented classification method. The following four conclusions:(1) According to the principle and algorithm of multi-scale segmentation, analysis different appropriate scale at different terrain types. Based on the multiple set scale, color factor, shape factor, firmness and smoothnesssets tests and visual discrimination, select appropriate segmentation parameter in each level. At last, contribute to feature information extraction and classification accuracy.(2) Through in-depth study and research, understanding the most common researchers usually combine the spectral information and texture information, or spectral information and vegetation types informationn classification method. Due to the research area of vegetation spectral information and texture information have similar type and recognize difficulty, the study base on the multi-scale segmentation and combine with a variety of vegetation index, spectral features and texture features information.Finally solve the problem of secondary forest vegetation classification better. The study has joined the different coefficient of different vegetation types in SAVI to improve the precison.At the same time, SAVI accuracy depends on the value L, in the experimental zone choose L value are 0.5,2, and 5 better. SAVI determine decreased confusion phenomenon of similar features, such as the farm land, broad-leaved tree and shrub-herbaceous, then improve the classification accuracy.(3) Adopts fuzzy classification’s membership function in the object-oriented method to recognize the vegetation and the non-vegetation and adopts decision tree classification algorithm to recognize the vegetation types. Then use multi-level classification rules method to extract different features, avoid some features confusion phenomenon effectively, realize and obtain the better classificaiton results to the vegetation type recognition.(4) According to the above methods and tests, combine the method of the various vegetation index, spectral characteristics and the methods of texture feature compare with only combine texture and spectral characteristics of the methods. The former classification accuracy is 91.3% increase greatly berrer than the latter 84.6%.