A multiple-blockage identification scheme for buried pipeline via acoustic signature model and SqueezeNet

Abstract

Due to buried pipes’ complex and changeable operating conditions and insufficient effective data samples, the multiple-blockage recognition model lacks great adaptability and generalization. A multiple-blockage identification scheme for buried pipeline via acoustic signature model and SqueezeNet is proposed, including three stages of acoustic signature model, feature learning with SqueezeNet, and blockage recognition and classification. First, indirect sound intensity and dynamic threshold are proposed to ensure segmentation that the typical individual acoustic signature images. Then, the lightweight SqueezeNet is used to extract the acoustic signature feature. Among them, only the data of healthy and single blockage to train the SqueezeNet recognition model. Finally, the unknown multiple blockages are divided into multiple independent individual parts. According to the results, the model can distinguish between the testing samples that contain all the blocking types and the unknown type of multiple blockages and achieve an average accuracy of 99.88% and 97%, respectively.