Studies On Several Classifications Of Land Use Based On High-precision Remote Sensing Images

The development of remote sensing technology allows us to obtain a wealth ofinformation. Remote sensing image classification and application of these data, so that it canbe converted to the application information. The choice of classification method is a verycritical aspect of the selection of a reasonable classification will not only improve theclassification accuracy, but also to avoid spatial data redundancy and waste of resources.In this study, only the agricultural demonstration zone-Yangling, for example, toNovember25,2008ALOS high-resolution remote sensing image data source in the remotesensing image processing software of ERDAS IMAGIE9.2image correction, integration,tailoring and other pre-processing, the display of the entire process of traditional remotesensing classification-supervised and unsupervised classification, and analysis to show thewhole process of object-oriented classification methods in remote sensing image processingsoftware the ENVI4.8, and the classification principle. algorithm and the results werecompared. The study made the following main conclusions:1.Yangling District ALOS high-resolution remote sensing image classification,sub-categories according to the status of agriculture demonstration area, arable land, water,woodland, beaches, idle industrial land, residential areas, roads a total of eight main surfacefeatures.2.Unsupervised classification of remote sensing image of Yangling District, to get theresults in Figure spot broken road and beach points out, the overall classification accuracy of42.8%, kappa coefficient was0.33, indicating that non-supervised classification is not suitablefor high-resolution image classification.3.Yangling District of remote sensing image supervised classification to get the results inFigure spot broken various land types can be a clear separation of the overall classificationaccuracy was65%, kappa coefficient of0.6, this method is better than the unsupervisedclassification results a lot.4object-oriented classification of remote sensing images of Yangling District, set thesplit scale for the20、40、60、70、80to explore the best segmentation scale. The results showedthat: the object-oriented, regardless of which split the scale to get classification accuracy than the traditional pixel-based classification accuracy is high, and the final result in Figure avoidbroken polygons, which split the scale60is the most ideal The overall classification accuracyof79.8%, the kappa coefficient was0.71. Therefore, the object-oriented classification methodhas higher accuracy than the supervised classification and unsupervised classification, theoverall accuracy and kappa coefficient of accuracy in the classification results than thesupervised classification were increased by14.8%and11%.