Research On High - Resolution Remote Sensing Image Change Detection Method Based On Object - Oriented

High resolution remote sensing images have such features as clear imaging, richtexture, precise positioning and large internal heterogeneity, which have madetraditional change detection methods based on pixels cannot make full use of imageinformation, and produce low classification accuracy. Meanwhile, these methods cancause a large number of spatial data redundancies and waste of resources, which cannot be well adapted to the characteristics of high-resolution images. Therefore, thestudy of high-resolution images change detection method has a strong practicalsignificance.This article takes IKONOS images of Tongzhou District of Beijing on October31,2008and November1,2010as an example to study high-resolution remote sensingimages change detection methods based on the object-oriented approach. First of all,Ratio of Mean Difference to Neighbors (ABS) to Standard Deviation (RMAS)method was used to test segmentation results based on multi-scale segmentationalgorithm of eCognition software. Then Object Correlation Images (OCIS) werecreated using the segmentation results. Automated binary change detection withmultiple variables method and decision tree classification method were used toexplore its efforts in change detection.The results show that the RMAS method can accurately determine segmentationscales of various types of land surface. Compared with automated binary changedetection with single variables, automated binary change detection with multiplevariables method produces better result. And automated binary change detection withOCIs shows higher change detection accuracy than difference images. For differenceimages, KAPPA coefficient of the first band is0.7,0.69for the second band,0.72forthe third band, and0.76for the fourth band. However, KAPPA coefficient of fourbands combined image is0.87. While for OCIs, KAPPA coefficient of correlationband is0.86,0.7for slope band,0.63for intercept band. But KAPPA coefficient is0.89when the three bands combine. What’s more, compared with other methods,decision tree classification method also makes better classification results. Theoverall accuracy of pixel-based decision tree classification is81%, and its KAPPAcoefficient is0.71; for object-oriented decision tree classification method, the overallaccuracy is88%, KAPPA coefficient is0.75; for nearest neighbor classification basedon the OCIs, the overall accuracy is87%, KAPPA coefficient is0.81; for decision tree classification based on OCIs, the overall accuracy is88%, Kappa coefficient is0.82.Found by the above studies, accuracies of object-oriented change detectionmethod is higher than the pixel-based methods, the object-oriented approach caneffectively avoid the phenomenon of salt and pepper.