Information Extraction Method Research Of High Resolution Remote Sensing Image

With the continuous development of aviation, spaceflight, sensor technology, the spacecharacteristic information, such as texture, shape and semantic relationship, of currentlywidely used high resolution remote sensing image become more outstanding，resolution ofhigh resolution remote sensing image become more and more high, Therefore, the extractionof high resolution remote sensing image information gradually become hot spot and difficultyof domestic and foreign research in recent years. Traditional remote sensing informationextraction method base on single pixel and pure spectral, with a single pixel as the researchobject, ignoring the richness texture and shape spatial information of remote sensing imagetexture, results in the waste of high resolution remote sensing image information, it stillconsider revising a lot of insufficient and abuses, and it’s difficult to accurately andeffectively extraction the macroscopic and microscopic feature information. And informationextraction method based on object-oriented rules set, with separating patch as the researchobject, taking full advantage of the high resolution remote sensing image texture, shape andspectrum of spectrum and space information, can effectively extract macroscopic featureinformation, such as water, cultivated and construction land, and the information extractionaccuracy is higher. But for micro forest information of same spectrum with different objects，such as natural forest and plantation,the information extraction results precision is low andexist a large number of "natural forest mixed with plantation in the artificial forest mixed withnatural forest" phenomenon in the extract results, That is to say that information extractionmethod based on the object-oriented rule set is difficult to effectively extract micro naturalforest and plantation vegetation information. Therefore, based on the index of RMAS(Ratio ofMean Difference to Neighbors(ABS)to Standard Deviation), this paper selected optimal splitscale, firstly, using multiresolution segmentation algorithm of definers, QuickBird highresolution remote sensing image in Fen Tai of Beijing were segmented, and establisheddifferent rules according to different macro feature object corresponding to different spectraland spatial characteristics,such as spectra, texture, shape, in order to extracting macro featureinformation, such as cultivated land, construction land. Using texture line number per unitarea TLNPUA index indicators extracts micro natural forest and plantation vegetationinformation, and respectively comparative analysis the results with information extractionresults based on a set of rules object-oriented method&based on the single pixel and purespectral traditional method and put forward object oriented and microscopic featureinformation extraction based on TLNPUA index combined with the results. The analysisresults show that (1) Fusion methods. Compared with PCA and HIS fusion method, Pan Sharpening fusionmethod is more suitable for multi-spectral and panchromatic image fusion of high resolutionremote sensing image, after fusion the texture, shape and space characteristic informationmore prominent.(2)Evaluation results of optimal segmentation scale evaluation method based on objectmatching weighted index ASFI3show that selection method of optimal segmentation scalebased on RMAS index is feasible, multi-scale segmentation algorithm based on the optimalsegmentation scale object-oriented multiresolution segmentation can effectively segment highresolution remote sensing image to extract macro, such as construction land, water, roads,farmland, and micro features,such as natural forest and plantation.(3)The method which combined object-oriented multiresolution segmentation algorithmwith TLNPUA index can effectively extract microscopic natural forest and plantationvegetation information with the same spectrum. Based on the optimal segmentation scaleobject-oriented high precision, effective segmentation, both the index of polygon number perunit area and the index of texture line number per unit area can extract microscopic naturalforest and plantation vegetation information with the same spectrum as index, but comparedwith the index of polygon number per unit area, the discrete degree of the index of texture linenumber per unit area is better, and can more effectively extract microscopic natural forest andplantation vegetation information, and0.1can be used as effective differentiate threshold.When0<TLNPUA <0.1, which is artificial forest vegetation coverage, When TLNPUA>0.1, which is natural forest vegetation coverage.(4)Different macro and micro feature information extraction method have differentinformation extraction accuracy results, overall performance is that object-orientedinformation extraction result taking into account spectral and spatial characteristics such asspectral, texture, shape, is superior to the traditional information extraction results based onsingle pixel and pure spectrum characteristic. Compared with supervised and unsupervisedinformation extraction method, the object-oriented information extraction accuracy based onspectrum, texture and shape spatial information and a set of rules is higher, and can extract themacroscopic and microscopic feature types effectively. But object-oriented informationextraction method based on a set of rules has extremely limitations, when extractionmicroscopic natural forests and plantations vegetation information with the same spectrum,the phenomenon that natural forest misclassification to plantation and artificial forestmisclassification to natural forest abound in the extraction results. Object-orientedinformation extraction method taking into account spectral and space characteristics, such asspectral, texture and shape, and base on TLNPUA index can effectively extract microscopicnatural forests and plantations vegetation information with the same spectrum, but for individual sparse natural forest, Texture line number per unit area is less, so it’s easy confusedwith artificial forest vegetation. Experiment results show that object oriented macro and microfeature information extraction method considering texture, spectral and shape spaceinformation characteristics can improve the accuracy of feature information extraction, whichoffers effective means for high precision macro feature information extraction, micro naturalforest and plantation with same spectral information extraction, forest mapping, forestmonitoring and ecological protection, and have highly signification for high resolution remotesensing information extraction.