The　combustion　stability　of　solid　waste　in　the　incinerator　determines　the　operating　efficiency　and　pollutant　emission　concentration　of　the　municipal　solid　waste　incineration　(MSWI)　process.　At　present,　domain　experts　identify　the　combustion　state　and　manually　control　the　MSWI　process　has　the　problem　of　unstable　identification　results　and　low　intelligence.　A　combustion　state　recognition　method　for　the　MSWI　process　based　on　VGG19　depth　feature　migration　is　proposed　to　address　the　abovementioned　problems.　First,　the　original　flame　image　is　enhanced　by　rotation,　adding　noise,　and　other　data　enhancement　methods　to　expand　the　size　of　labeling　samples　to　overcome　the　high　cost　of　manual　labeling.　Second,　the　VGG19　model　based　on　ImageNet　pre-training　is　used　as　the　base　model,　and　the　output　of　the　last　layer　of　the　middle　layer　is　used　as　model　the　output　to　realize　feature　transfer　learning　by　enhancing　the　flame　image　dataset　and　fine-tuning　the　model　parameters.　Finally,　the　flame　feature　extracted　by　VGG19　is　used　as　the　input　of　the　improved　cascade　forest　to　build　the　combustion　slate　recognition　model.　The　experimental　results　show　that　the　recognition　rate　is　99.72%,　which　proves　the　effectiveness　of　the　method.