In　this　paper,　we　propose　a　novel　hand-object　interaction　reconstruction　method　based　on　diffusion　model　and　signed　distance　fields　(SDFs).　This　method　uses　a　multi-scale　feature　extractor,　an　attention　bridge,　a　segmentation　branch　module　based　on　vector　merging,　and　reconstructs　the　3D　structure　of　hand　and　object　from　monocular　color　RGB　images.　In　this　study,　a　multi-scale　feature　extractor　is　used　to　extract　image　features,　and　the　feature　representation　ability　is　enhanced　by　the　attention　mechanism.　In　addition,　a　diffusion　model　is　introduced　into　the　hand　and　object　shape　estimation　in　HOI　to　optimize　the　feature　extraction,　and　the　Marching　Cubes　algorithm　is　used　to　extract　zero　isosurfaces　from　the　signed　distance　field　and　convert　them　into　grid　form　to　obtain　the　reconstruction　result　of　the　three　dimensional　surface.　Through　ObMan　data　set,　we　verified　the　HOI　method,　and　the　reconstruction　results　showed　obvious　improvement　in　detail　and　quality.　©　The　Institution　of　Engineering　&　Technology　2024.