As　the　heterogeneous　system　based　on　CPU-GPU　architecture　has　a　strong　performance　advantages,　so　that　it　has　been　widely　used　in　many　fields　especially　image　recognition,　in　order　to　enhance　the　transmission　efficiency　on　HSA　(Heterogeneous　System　Architecture)　Platform.　In　this　paper,　we　design　a　naive　Bayesian　image　classification　algorithm　on　the　APU,　which　supports　the　HSA　standard,　and　design　a　set　of　resource　dynamic　scheduling　strategy　for　the　eigenvalue　extraction　of　the　algorithm.　The　experimental　results　show　that　this　strategy　can　save　29.56%　execution　time　on　average.