When　robots　are　grasping,　they　often　face　objects　with　uncertain　poses　and　irregular　shapes,　so　it　is　difficult　to　estimate　the　grasping　pose　of　the　robot　accurately.　To　address　this　problem,　this　paper　proposes　a　robot　target　grasping　pose　detection　method　based　on　deep　learning.　The　method　is　able　to　face　different　target　detection　with　more　flexibility　and　accuracy.　This　method　uses　a　network　model　based　on　ShuffleNet　v2　combined　with　a　Region　Proposal　Networks(RPN)　to　predict　the　graspable　region　of　an　object.　In　this　networks,　classification　labels　are　used　to　output　the　angle　of　the　grasping　pose,　and　the　location　parameters　of　the　grasping　are　predicted　using　a　regression　method.　The　model　was　trained　on　the　Cornell　dataset　and　achieved　the　accuracy　of　91.12%　and　91.02%　on　the　instance　detection　and　object　detection　test　sets,　respectively.　The　processing　time　is　within　0.07s　per　image.　The　experiments　show　that　the　model　has　high　accuracy,　robustness,　and　stability　in　detecting　and　locating　single　or　multiple　grasped　objects　in　images.　©　2022　IEEE.