Research On Classification Of Vegetation From Remote Sensing Imagery

As the main body of the ecosystem and important part of human survival environment, thevegetation is closely related to water, air, soil, climate, species diversity, and it is one of theimportant factors of relief map in basic scale. This paper in-depth analysed the characteristics ofvegetation represented in remote sensing Imagery, researched a vegetation classification methodbased on spectral and texture knowledge rules, and maked object-oriented idea be applied tovegetation classification.Finally,this paper obtained a better classification results than thecommonly used classification methods by experiments. The major works completed in thisarticle:1．Status of the vegetation extraction in mapping relief maps using remote sensing imagerywas analyzed; the importance and urgency of the vegetation classification extraction techniqueresearch were discussed; status of vegetation classification technology research was summed up.2．According as what’s follow-up experiments required in this paper and characteristics ofimage used in experiments, piecewise linear stretch enhancement and the image fusion wereanalyzed and summed up, and analyzed by experiments.3．By in-depth analysis in character information of vegetation and other surface featuresand different vegetation types in image used in experiments,differences in characteristicsbetween them were detected, and the differences characteristics of different vegetation typeswere pulled by using certain enhanced method; on the principle of “from simple to complex,fromspectral to texture”, vegetation classification method based on spectral and texture knowledgerules was established, test results showed that this method was easy to be implemented,compared to the commonly used method,accuracy of this method had improved greatly.4．The related theorys of object-oriented classification were introduced; By experiments,multi-scale segmentation parameters setting rules were explored; The commonly used featuresselection method of image objects was summed up,and the most excellent combination ofcharacteristics was selected; Training classification of samples by the optimal characteristiccombination, image objects segmented in the optimal segmentation parameters were classified,the result showed that: object-oriented technology could improve the classification accuracy,andmade the classification results be interpreted easier.5．To test of the applicability of object-oriented methods, several different spatialresolution images were classified based on object and pixel, classification results accuracywere compared and analysed in "vertical" and "horizontal", the trial conclusion: the higher imagespatial resolution, the more markedly it can show the advantages of object-oriented approach.