With　the　wide　application　of　hyperspectral　images　in　various　fields,　the　theory　of　sparse　representation　of　signals　has　been　getting　more　attention　from　researchers.　This　paper　presents　a　classified　redundant　dictionary-learning　algorithm　of　hyperspectral　image　based　on　geologic　feature.　Experimental　results　showed　that　hyperspectral　image　signals　could　be　expressed　with　the　greatest　fidelity.　The　algorithm　constructs　redundant　dictionary　library　through　a　sparse　decomposed　image　according　to　the　geologic　feature　based　on　clustering,　which　more　accurately　represents　the　spectral　signal　in　sparse　reconstruction.　To　apply　the　classification　dictionary　library　to　super-resolution　restoration　of　hyperspectral　image,　sparse　decomposed　hyperspectral　image　is　needed　to　obtain　high　and　low　resolutions　of　a　redundant　dictionary.　Then,　low　resolution　and　dictionary　of　hyperspectral　image　restoration　could　be　obtained.　The　algorithm　could　effectively　improve　image　resolution　to　ensure　the　quality　of　image　restoration.　©　2017　IEEE.