To　address　common　issues　of　missed　and　false　detections　in　small　object　detection　tasks,　the　PFR-YOLOv8s　algorithm,　incorporating　pixel　and　feature　rearrangement,　is　proposed.　The　pixel　rearrangement　feature　extraction　module　(PRFE)　with　attention　mechanisms　is　constructed　to　preserve　fine-　grained　information　and　capture　crucial　features　of　small　objects　in　complex　backgrounds.　A　multi-　feature　fusion　mechanism　(MFF)　is　designed　to　fully　exploit　contextual　and　multi-scale　information　of　small　objects.　Based　on　MFF,　the　feature　rearrangement　neck　framework　(FR-Neck)　enriches　target　features　through　feature　rearrangement,　enhancing　the　model’s　perception　of　small　objects.　The　Inner-　GIoU　loss　function　is　introduced　to　improve　the　learning　capability　for　small　object　samples　in　dense　scenes.　Experimental　results　show　that　this　algorithm　improves　average　detection　accuracy　by　5.8　percentage　points　over　the　original　YOLOv8s　model　on　the　VisDrone2019　dataset,　significantly　enhancing　small　object　detection　capabilities.　Additionally,　average　accuracy　on　the　railway　worker　safety　small　object　dataset　improves　by　2.1　percentage　points,　validating　the　algorithm’s　generalization　ability.　©　2025　Journal　of　Computer　Engineering　and　Applications　Beijing　Co.,　Ltd.;　Science　Press.　All　rights　reserved.