With　the　rapid　development　of　computer　vision　and　deep　learning　technologies,　object　detection　plays　a　crucial　role　in　various　fields.　Due　to　the　complexity　and　diversity　of　underwater　environments,　underwater　vision　tasks　face　great　challenges.　In　this　paper,　we　improved　the　YOLOv8　detection　model　for　the　fish　detection　task　in　complex　environments.　Specifically,　we　proposed　a　multi-scale　feature　fusion　method　(MFANet)　that　combines　multiscale　weighted　feature　fusion　with　multiscale　attention　to　efficiently　capture　multilevel　features　of　fishes　in　complex　environments.　We　designed　a　multiscale　weighted　feature　fusion　module　that　combines　feature　maps　at　different　scales　to　enhance　the　model＇s　representation　of　the　target.　By　introducing　a　multi-scale　attention　mechanism,　the　model　is　able　to　pay　more　attention　to　important　feature　regions.　We　conducted　experiments　on　real　underwater　complex　environment　datasets.　The　results　show　that　our　proposed　improved　method　significantly　enhances　the　performance　of　the　YOLOv8　model　in　fish　detection　tasks　and　lays　a　good　foundation　for　subsequent　underwater　detection　tasks.　©　2024　IEEE.