Layout　analysis　is　the　first　step　in　automatic　grading　and　other　OCR　tasks.　Although　various　layout　analysis　technologies　have　been　developed　for　different　application　scenarios,　existing　approaches　still　have　difficulty　in　achieving　high　accuracy　on　answer　sheet　analysis.　By　combining　the　deep　neural　network　object　detection　model　YOLOv5s　and　traditional　text　extraction　methods,　a　layout　analysis　method　for　answer　sheets　is　proposed　in　this　paper.　The　foreground　and　background　of　aligned　answer　sheets　are　separated　by　a　method　that　we　propose　based　on　the　variation　in　the　pixel　gradient.　Then,　a　YOLOv5s-DC　model　is　trained　to　detect　the　handwriting　texts　in　the　answer　sheets,　and　MSER　(Maximally　Stable　Extremal　Regions)　is　applied　to　the　answer　sheets　to　extract　the　missing　parts　and　background　texts.　The　decoupled　YOLO　head　replaces　the　original　YOLO　head　for　higher　performance.　Quality　focal　loss　and　efficient　IoU　loss　are　employed　during　loss　calculation　to　supervise　the　classification　and　regression,　respectively.　After　the　detection,　the　seam　carving　algorithm　and　other　rules　are　applied　to　the　bounding　boxes　to　improve　the　extraction　results.　The　mean　average　precision　of　YOLOv5s-DC　is　91.6%,　which　is　3.2%　higher　than　that　of　YOLOv5s.　The　pixel　accuracy　of　our　text　extraction　method　is　99.92%.　The　experimental　results　verify　that　our　method　can　effectively　and　accurately　analyze　the　layout　of　answer　sheets.