The mixed pixels estimation for the remotely sensed images using a nonlinear model and a filter of neighborhood pixels

Several papers have reported that errors in land cover classification have been generated especially around the edges of categories. Mixed pixels that include multiple categories are one of the error factors. All remotely sensed images include mixed pixels, and in particular the influence of mixed pixels in middle- or low-resolution images is more serious. In this paper a method using a nonlinear mixed pixel model and filtering of neighborhood pixels are proposed. In order to verify the effectiveness of the proposed method, six simulations were performed. Two patterns without the proposed filter were used for the linear and nonlinear models, and four patterns with the proposed filter were used for the proposed nonlinear method. The results of the simulations were quantitatively evaluated by calculating the coefficients of correlation between the simulation results obtained from Landsat-7/ETM+ images and the model answers made by QuickBird-2 images. It was confirmed that the coefficients of correlation of the proposed nonlinear model and the proposed filter were superior to the other results. (C) 2012 Wiley Periodicals, Inc. Electron Comm Jpn, 95(4): 110, 2012; Published online in Wiley Online Library. DOI 10.1002/ecj.10422