Object　detection　is　the　task　of　classifying　and　locating　objects.　Deep　learning-based　object　detection　is　the　current　mainstream,　however,　detecting　small　scale　targets　is　still　challenging.　To　achieve　accurate　detection　of　small　objects,　we　propose　a　bilateral　dense　feature　circulation　method　(BCM).　First,　to　make　the　network　strategically　extract　features　that　contain　useful　location　information,　we　design　a　bilateral　feature　circulation　module　in　a　dense　nested　way　to　detect　small　objects.　Second,　in　order　to　fully　utilize　and　fuse　each　scale　feature,　we　sequentially　perform　the　channel　and　spatial　attention　to　each　layer　in　the　feature　fusion　stage.　Third,　through　repetitive　and　progressive　feature　extraction　and　fusion　under　the　instruction　of　the　attention　module,　we　attain　sufficient　information　to　detect　small　objects.　We　conducted　experiments　on　PASCAL　VOC　and　Tsinghua-Tencent　100k.　The　experiment　results　show　that　the　proposed　model　improves　the　mAP　of　small　objects　on　PASCAL　VOC　by　1.8%　and　Tsinghua-Tencent　100k　by　2.7%　compared　with　SOTA　methods.　©　2024　Technical　Committee　on　Control　Theory,　Chinese　Association　of　Automation.