Due　to　strong　noises　from　system　components　and　measurement　devices,　arc　faults　would　become　weak　to　bring　about　detection　challenges.　Therefore,　new　measurements　should　be　taken　to　acquire　obvious　arc　fault　features　under　complex　operation　disturbances　in　photovoltaic　application　scenes.　In　this　article,　weak　arc　fault　signals　are　acquired　from　the　designed　experimental　and　simulation　platform　with　different　system　structures　and　signal　acquisition　devices　first.　Arc　fault　features　would　become　weak　in　initial　transient　arc　fault　stages　and　higher　frequency　bands　above　15.6　kHz,　which　could　not　be　directly　acquired　by　applying　designed　digital　filters　with　existing　wavelets.　Next,　the　ant　colony　algorithm-based　stochastic　resonance　(ACA-SR)　method　is　proposed　to　enhance　arc　fault　features,　which　is　verified　to　be　effective　under　various　inverter　and　resistor　conditions.　Then,　the　feature　enhancement　ability　of　stochastic　resonance　method　is　evaluated　to　select　the　optimal　arc　fault　feature,　which　is　proven　to　have　the　average　feature　enhancement　ability　of　3.23　times.　The　selected　wavelet　is　proven　to　have　the　superiority　of　symmetry　and　shorter　support　width.　Finally,　the　proposed　method　is　conducted　with　universal　conditions　containing　weak　arc　faults　via　numerical　and　hardware　tests,　which　is　proven　to　improve　the　detection　accuracy　of　arc　faults　by　24.57%　on　average　compared　with　existing　methods.　©　1963-2012　IEEE.