The　flame　combustion　state　inside　the　furnace　is　important　feedback　information　for　intelligent　optimization　control　of　the　municipal　solid　waste　incineration　(MSWI)　process.　However,　none　of　the　existing　studies　on　the　identification　of　MSWI　combustion　state　have　proposed　a　combustion　state　classification　criteria,　which　with　actual　physical　significance　and　interpretability　for　the　incineration　characteristics　of　MSWI.　As　a　result,　it　is　difficult　to　build　a　recognition　model　for　the　MSWI　process　based　on　the　domain　expert　cognitive　mechanism　and　valid　reference　data.　To　solve　this　problem,　combined　with　the　experience　of　industry　experts　and　research　results　in　related　fields,　the　construction　of　MSWI　process　flame　combustion　state　classification　criteria　and　benchmark　database　was　studied　in　this　paper.　Firstly,　the　problem　of　combustion　slate　identification　is　described,　and　the　existing　methods　of　combustion　state　identification　based　on　combustion　lines　arc　analyzed.　Next,　the　classification　criteria　are　elaborated　based　on　normal　combustion,　partial　combustion,　channeling　combustion　and　smoldering　combustion.　Then,　the　image　database　of　the　flame-burning　state　which　can　be　used　for　machine　learning　is　constructed.　Finally,　the　flame-burning　image　database　is　modeled　and　tested　based　on　a　variety　of　classical　algorithms　in　the　field　of　machine　vision.　The　results　show　that　the　accuracy　of　most　methods　for　flame　slate　classification　is　more　than　80%.　The　validity　of　the　proposed　classification　criteria　and　flame　image　database　is　greatly　validated.