Sparse Data Recovery Algorithm Based on BP Neural Network for Ultrasonic Guided Wave Imaging

Abstract

Ultrasonic guided wave (UGW) imaging quality is limited by the large number of sensors. In this paper, a sparse data recovery algorithm based on back forward (BP) neural network is proposed to solve the problem that the image quality deteriorates with the decrease of the number of sensors. The sparse data from sparse sensor array is up-sampled preprocessing by compressive sensing and then input to the BP neural network to further reduce the recovery error. Numerical results show that the recovery errors reduce from 10−3 and 10−2 to 10−6 for 32 and 16 sensors. After sparse data recovery, the recovered dense data is used for imaging. The average correlation coefficient related to the imaging quality of 32 sensors is improved to the level with 64 sensors. For 16 sensors imaging, the average correlation coefficient is also improved, but the image quality is still slightly reduced compared with 64 sensors.