The invention discloses a pipeline new defect identification method based on an evolutionary fuzzy maximum and minimum neural network, which can be used for continuous training on the basis of an existing model, and comprises the following steps of: 1) establishing a source model by utilizing an original mark sample of a pipeline; 2) screening out a new depth mark sample subset Dnew from the newlymeasured mark data set; 3) on the basis of the source model, continuously performing training by utilizing the new mark subset Dnew, and establishing a new mark sample identification model; and 4) performing pipeline defect identification by utilizing the trained new model. According to the method, an existing model can be utilized to adapt to a new depth defect size, meanwhile, an evolutionary pipeline defect recognition framework is provided, continuous training of the network can be achieved, and the problem that once a new depth size defect exists, all samples need to be retrained is avoided.