It　is　important　to　accurately　identify　the　combustion　state　of　the　municipal　solid　waste　incineration　(MSWI)　processes.　Stable　state　not　only　can　greatly　improve　the　combustion　efficiency,　but　also　can　ensure　safety　of　the　MSWI　processes.　What’s　more,　the　pollution　emission　concentration　would　be　greatly　reduced.　Aiming　at　the　situation　that　domain　experts　identify　the　combustion　state　in　terms　of　self-experience　in　the　actual　MSWI　processes,　this　study　proposes　an　efficient　method　based　on　improved　deep　forest　(IDF).　First,　the　image　preprocessing　methods　such　as　defogging　and　denoising,　were　used　to　preprocess　the　combustion　flame　image　to　obtain　a　clear　one.　Then,　the　multi-source　features　(brightness,　flame　and　color)　were　extracted.　Finally,　the　multi-source　features　were　used　as　the　input　of　cascade　forest　module　in　terms　of　substituting　multi-grained　scanning　module.　Therefore,　a　combustion　state　recognition　model　of　MSWI　processes　based　on　IDF　was　established.　Based　on　actual　flame　images　of　industrial　processes,　many　experiments　has　been　done.　The　results　showed　that　the　constructed　model　can　reach　a　recognition　accuracy　of　95.28%.　©　2022,　The　Author(s),　under　exclusive　license　to　Springer　Nature　Singapore　Pte　Ltd.