In　this　paper,　we　proposed　an　infrared　(IR)　ship　target　saliency　detection　method　based　on　improved　non-local　depth　features.　There　are　mainly　two　contributions.　First,　considering　the　low　contrast　characteristics　of　IR　images,　we　proposed　an　improved　lightweight　non-local　depth　feature　method　(Light-NLDF)　for　IR　ship　target　saliency　detection.　This　method　mainly　consists　of　three　parts,　CNN　based　feature　extraction,　top-down　feature　refinement　with　deconvolution,　and　improved　loss　function　by　adding　structural　similarity　index　(SSIM).　Secondly,　we　construct　an　IR　ship　target　image　dataset　for　saliency　detection.　This　dataset　includes　3,069　IR　images　and　ground-true　images　with　different　backgrounds　and　different　objects.　Experimental　results　show　that　our　proposed　method　is　robust　and　suitable　for　IR　ship　target　saliency　detection.　By　abandoning　the　module　of　contrast　calculation　and　the　fusion　of　global　and　local　features,　the　proposed　Light-NLDF　can　greatly　improve　the　efficiency　of　training　and　detecting.　Comparison　results　with　two　well　known　methods　demonstrated　that　the　proposed　Light-NLDF　achieves　a　satisfying　performance　for　IR　ship　target　saliency　detection　with　a　F-measure　of　75.21%　and　a　lightweight　model　with　a　size　of　82　MB.　©　2019　IEEE.