River Extraction from High Resolution Remote Sensing Images Worldview2
|School||Xi'an University of Architecture and Technology|
|Course||Control Theory and Control Engineering|
|Keywords||Rivers extraction Feature extraction K-mean Worldview2image Multi-feature-fusion|
With human society and the rapid development of science and technology,as themost powerful space detection of satellite remote sensing technology has already spreadall walks of life. The river network as the geographic information system to the lessimportant constituent,the identification、analysis、extraction and accurate positioningfor GIS data real-time update, surveying and mapping, remote sensing mapping andpeople’s life has important significance. In recent years, the resolution of remote sensingsatellite is the skip-type development, that is the modern is high resolution remotesensing satellite image age, and the past than low resolution, image featurecharacteristics present a breakthrough change, thus for more love based on remotesensing of feature recognition researchers provide ever more rich and powerfulbackground support.Traditional remote sensing image classification method are mainly on the basis offeatures of the spectral features，is suitable for low resolution of remote sensingimages,and the high resolution image features a rich texture information, geometricshape, space position and boundary characteristics. But at the same time spectralinformation relatively weak, so the traditional rely on spectrum method is no longeradapt to the high resolution image.Modern scientists have deep into the eye from highresolution large database search useful target information.This paper main work includes the following parts:(1) when sensor accept radiation of objects, there are a lot of interferencefactors,including the machine itself, radiation angle and so on.So the interferencefactors of remote sensing images in imaging process need processed, including radiationcorrection, geometric correction and image fusion. Introduction to various operations method of math ematical morphology. On the extracting river network, extractingtargets were processed by using all kinds of operation of mathematical morphology.Then, clear and accurate targets would be obtained.(2) Introduction to multiband index model, improved differential water index(MNDWI), improved mixed water body index model (MCIWI), principal componentanalysis (PCA), construct multiband index comprehensive vector. Will the original eightspectral band into a new spectral band for warter body extraction. Then decision treeclassification method was used to classify and extract rivers.(3) Thirdly, this paper proposed more feature synthesis cluster segmentationalgorithm based on the high resolution remote sensing satellite image. After the riverimage being preprocessed, many features joint extraction method of river, whichincluded image spectrum, texture, and geometric shape of river, was proposed. Thismethod used outstanding high resolution character of high resolution remote sensingimage. The method respectively described the spectral features, texture features andgeometry shape of river water body. Characteristic parameters were chose,comprehensive feature matrix was constructed, and mean clustering division was used.The river target was eventually got. The experiments of real high resolution remotesensing image Worldview2image proved that this method had high accuracy and fastspeed.