The　redundancy　and　complexity　of　combustion　flame　image　features　increase　the　difficulty　of　recognizing　the　combustion　condition　of　municipal　solid　waste　incineration　(MSWI).　The　complexity　of　solid　waste　components　and　the　inherent　nonlinearity,　time　variation　and　uncertainty　of　MSWI　process　lead　to　the　instability　of　flame　image　feature　distribution.　The　traditional　method　based　on　fixed　sliding　window　only　extract　fixed　size　features　and　cannot　reflect　global　and　local　features,　which　lead　to　low　recognition　accuracy　of　combustion　conditions.　To　address　the　issue,　this　paper　proposes　a　model　of　combustion　condition　recognition　in　MSWI　process　based　on　multi-scale　color　moment　features　and　random　forest　(RF).　Firstly,　the　image　is　pre-processed　by　defogging　and　denoising.　Then,　the　color　moment　features　of　different　scales　of　the　flame　image　is　extracted　via　using　sliding　windows　based　on　prior　setting　scales.　Finally,　taking　the　classification　accuracy　as　the　evaluation　criterion,　and　the　RF　algorithm　based　on　feature　selection　is　adopted　to　realize　accurate　identification　of　combustion　conditions.　Based　on　the　flame　image　data　of　an　MSWI　plant　in　Beijing,　this　method　has　been　validated　by　the　experimental　results.