Study On Land Cover Classification Based On Intelligent Algorithm

Remote sensing data acquisition technology and the expanding applicationneeds to jointly is promoted the development of remote sensing applicationprocessing. From remote sensing image was analyses land cover classificationthat was a series of the pre-process of land information collation, classificationof land was not only from the amount of land, the spatial distribution of acorrect understanding of the status and quality, but also to be improved to helpexpand the scope of application of the land, and the rational use it. Remotesensing image into information and knowledge that can be used by remotesensing application processing was prerequisite for the satellite remote sensingapplications.It has a variety of land cover classification in the wake of thecontinuous development of scientific information and remote sensing technique.However, due to various factors of time, space, multi-scale has not been able toexplore the classification of the "universal ". Land cover classification withintelligent classification methods based on several commonly used for differenttypical regional, and then the classification accuracy of the comparison ofseveral methods to filter out more applicable to the different attributes of theregional classification in this paper.The natural properties was more significant in Dalinuoer Nature Reserveand socio-economic attributes was the more significant Hohhot used land coverclassification by several intelligent classification, as a result as following:(1) The experiment of the result used method of genetic and artificialneural network(BP algorithm and SOFM network)on the natural properties wasmore significant in Dalinuoer Nature Reserve and socio-economic attributes wasthe more significant Hohhot, the classification accuracy of SOFM network thanthe classification accuracy of the genetic algorithm and BP algorithm was higherthat analysis and comparison of experimental results obtained for grassland, arable land classification of forest lands and waters; the BP algorithm and theclassification accuracy of SOFM network were relatively close to unused land;which the classification accuracy of the three methods were lower in residentialand industrial land; by contrast, the classification results of the genetic betterthan BP algorithm and SOFM network effect.The results of classificationshowed that the overall classification accuracy of the genetic, BP algorithm andSOFM network was85.9%,88.1%and93.03%and kappa coefficient of BPalgorithm and SOFM network was0.8431and0.9128. In general, theclassification results of the more obvious areas of the natural properties of theclassification of land cover used the SOFM network better than the genetic andBP algorithm.(2) The experiment of land cover classification of the more significantsocio-economic attributes of Hohhot used on direct BP classification methodand tolerant rough set classification method for BP network. The experimentalresults showed that the addition to the traffic space and agricultural land usedmore than90%on the classification accuracy of direct BP classification method;the residential land and waters that up to94%on the the classification accuracyof land, especially; the results of both classification methods were better inindustrial warehousing space the classification. The different of classificationaccuracy between the two classification methods that the direct BP classificationmethod and tolerant rough set classification method for BP network was87.96%and91.70%, kappa coefficient was0.8896and0.9298for the other types of land.In general, the classification results of the more survey region of theclassification of land cover used the tolerant rough set classification method forBP network better than the direct BP classification method.The classification results will be different that different region of thetypical choice of algorithms and models studied on a variety of paper ofdomestic and overseas. The selection of algorithms and models a directinfluenced on the results of remote sensing image. We should be the selectionthe best algorithms and models based on the classification of land cover featuresfor remote sensing image. The experiment results showed, the SOFM network better than the genetic and BP algorithm in the natural attributes; the tolerantrough set classification method for BP network better than the direct BPclassification method in socio-economic attributes.