Deep　learning　has　excellent　achievements　in　computer　vision　such　as　pedestrian　detection.　Based　on　YOLOv3　algorithm,　this　paper　introduces　CBAM　to　improve　the　feature　representation　ability　and　improve　the　overall　accuracy　of　target　detection.　Then　the　easy　to　implement　CSPNet　is　used　to　reduce　the　computational　workload　while　maintaining　the　original　accuracy.　The　experimental　results　based　on　INRIA　dataset　show　that　after　the　introduction　of　CBAM　attention　module　and　CSPNet,　the　efficiency　and　accuracy　of　object　detection　are　improved,　so　it　has　better　performance.　　©　2023　IEEE.