In order to solve the problem of low accuracy in ultrasonic characterization of defect detection,a new method based on BP neural network and D-S evidence theory is proposed to identify defects in ultrasonic inspection.Firstly,a fusion model based on BP neural network and evidence theory is proposed.The BP neural network is used to fuse the feature layer,and its output is taken as the basic probability distribution function of the evidence source.Secondly,in the fusion of decision-making level,the validity of the evidence source is evaluated by introducing an effective factor λ to evaluate the reliability of the evidence source.Therefore,all the evidence sources are fused after the reliability evaluation.Finally,the defect data of a certain type of aeronautical material are obtained by means of ultrasonic testing,and the method proposed in this paper is verified.The results show that:compared with the traditional D-S evidence theory,this method can identify the flaws more accurately and improve the accuracy of flaw recognition.