This　study　aims　to　achieve　online　detection　of　the　surface　chemistry　properties　of　mineral　particles　during　the　flotation　process　using　surface-enhanced　Raman　spectroscopy（SERS）.　A　high-quality　SERS　substrate　is　prepared　on　the　surface　of　indium-tin　oxide　（ITO）conductive　glass　through　electrodeposition　of　a　gold　nanoparticles　film.　Subsequently，　an　inert　silicon　dioxide　mono-molecule　isolation　layer　is　formed　on　the　gold　surface　through　selfassembly，　enabling　the　in-situ　detection　of　adsorbed　flotation　collector　molecules　on　the　surface　of　mineral　particles.　Addressing　the　variable　ore　grade　characteristics　during　the　production　process，　the　competitive　adaptive　reweighted　sampling（CARS）　and　automatic　multiscale-based　peak　detection（AMPD）algorithms　are　employed　to　extract　feature　signals　from　the　SERS　data　of　samples　with　varying　chalcocite　content.　This　study　employs　the　backpropagation　neural　network　（BP）　and　partial　least　squares　（PLS）　algorithms　to　construct　models　to　predict　the　chalcocite　content　based　on　the　saturation　adsorption　of　the　reagent.　After　comparison，　the　AMPD　algorithm　is　found　to　more　accurately　reflect　the　characteristic　peaks　of　adsorbed　molecules，　and　the　accuracy　of　the　BP　neural　network　algorithm　surpasses　that　of　the　PLS　algorithm.　The　AMPD-BP　neural　network　model　has　a　root　mean　square　error　of　prediction　（RMSEP）　of　0.02664，　with　a　correlation　coefficient　（R）　of　0.9755，　indicating　its　excellent　predictive　performance.　SERS　combined　with　machine　learning　can　be　used　for　online　detection　of　surface　properties　of　mineral　particles　during　flotation，　providing　methodological　support　for　the　detection　of　reagent　adsorption　amount　and　flotation　intelligent　optimization　system.　©　2025,　Youke　Publishing　Co.,Ltd.　All　rights　reserved.