Traditional　manual　detection　method　of　crop　pests　is　a　quite　tedious　work　with　low　efficiency,　which　brings　great　inconvenience　to　the　control　and　removal　of　crop　pests　at　early　stage.　In　recently　years,　computer　vision　becomes　a　critical　and　promising　technique　for　pest　detection.　However,　limited　to　the　shape　and　size　of　the　pest　and　other　issues,　the　perforance　of　these　methods　are　not　so　effective　and　accurate.　In　order　to　improve　the　detection　accuracy,　we　propose　a　discriminative　method　for　pest　detection　on　leaves　based　on　low-rank　representation　and　sparsity.　By　utilizing　the　low　rank　characteristics　of　natural　images,　the　sparsity　of　the　noise　image　and　the　prior　knowledge　of　color　information　of　the　crop　pest　images,　our　method　decomposes　the　original　image　into　low-rank　image　and　sparse　noise　image,　which　contains　all　pests　on　the　leaf.　After　that,　the　crop　pests　with　leaf　can　be　separate　from　the　background　and　counted　effectively.　The　experimental　results　show　that　our　method　can　detect　pests　on　leaf　conveniently.　This　is　of　great　significance　for　future　pest　judgment　and　management.