The　flare　pilot　is　the　key　to　ensure　that　the　flare　system　is　ignited　at　any　time.　At　present,　domestic　and　foreign　companies　mainly　use　thermocouples,　flame　ionization　detectors　and　other　sensors　to　detect　the　working　condition　of　the　flare　pilot.　These　electronic　components　have　the　hysteretic　and　vulnerable　problems　due　to　extreme　heat,　thermal　shock　and　vibration,　which　may　well　further　lead　to　the　failure　to　ignite　the　flare　gas　with　flare　pilot　in　time　and　cause　production　accidents.　In　view　of　the　above　problems,　this　paper　proposes　an　anomaly　detector　of　flare　pilot　based　on　deep　learning　technology.　First　of　all,　we　made　an　anomaly　detection　database　for　flare　pilot,　which　was　shot　in　a　domestic　petrochemical　company.　Then,　we　designed　a　specific　siamese　network　based　on　the　triplet　loss　function　for　learning　the　similarity　between　the　samples,　in　order　to　apply　to　the　curing　scene　of　the　flare　pilot.　Finally,　it　is　determined　whether　the　flare　pilot　is　abnormal　according　to　the　similarity　between　the　input　picture　and　the　positive　and　negative　samples.　Experimental　results　demonstrate　that　the　proposed　method　can　be　effectively　used　for　the　anomaly　detection　of　flare　pilot.