This paper proposes a subpixel land cover mapping method based on multi-dictionary sparse representation.In this method

some known high spatial resolution land cover maps are used to formulate different dictionaries that represent distribution modes of different land cover classes

the unclassified subpixels are represented by each dictionary

and they are also classified according to the principle of reconstruction-error minimization and spectral distortion constraint.Experimental results both on artificial and real images show that the method deals with the diversity between different distribution modes of different land cover classes effectively

and achieves higher subpixel mapping accuracy and robustness than the other related methods.