Compressive Sensing and Deep Learning Enhanced Imaging Algorithm for Sparse Guided Wave Array

Abstract Aiming at the problem of image quality reduction caused by sparse array in guided wave detection, an enhanced algorithm based on improved compressive sensing and deep learning is proposed in this paper so as to realize high-quality imaging with a small number of sensors. The enhancement algorithm consists of two parts: the sparse guided wavefield is up-sampled by the improved compressed sensing, and then the up-sampled guided wavefield is input into U-net for further recovery. After compressive sensing and deep learning enhancement, the recovered wavefield is close to the dense wavefield. Simulation is carried out and the results verify the feasibility of the method. In training and validation, the losses evaluated by mean square error (MSE) are 1.62 × 10-4 and 2.18 × 10-5 for 32 sensors and 1.65 × 10-4 and 3.44 × 10-5 for 16 sensors. Imaging performance is also verified by Pearson’s coefficient. The Pearson’s coefficient is improved from 0.9218 to 0.9517 with 32 sensors, and improved from 0.8896 to 0.9487 with 16 sensors.