Abstract: A recognitionsystemwaspresentedtoaccomplishspacetargetclassification.Firstly,Imagefeaturevectorswereextractedaccordingtoinvariantregionmoments,shapeandimagedescriptorsofspaceobjects.Andthen,clusteringalgorithmisappliedtotheclassifyspacetarget.AnimprovedkernelclusteringalgorithmbasedonVoronoidistancewasproposedwhichhadacrispermembershipfunctionandwasrobustfornoiseandoutliers.Experimentsshowthattheimprovedkernelfuzzyclusteringalgorithmismoreaccurateandvalidthanthatoftheconventionalmethods.

Key words: Image recognition, Kernel cluster method, Feature extraction, Space target