Thispaperisconcernedwithahighcharacteristicimageprocessingandrecognitionsystemthatisusedforinspectingreal-timeblemishes,streaksandcracksontheinnerwallsofhighaccuracypipes.Asaregulardetector,theBPneuralnetworkisusedforextractingfeaturesoftheimageinspectedandclassifyingtheseimages,ittakesfullyadvantageofthefunctionofartificialneuralnetwork,suchastheinformationdistributedmemory,largescaleself-adaptingparallelprocessing,highfault-tolerantabilityandsoforth.Besides,animprovedBPalgorithmisusedinthesystemfortrainingthenetwork,andmakingthelearningprocedureofthenetconvergestotheminimumofoverallsituationathighrate.
