Sparse denoising method for flip chip ultrasonic signals based on memory-enhanced deep unfolding network

A sparse denoising method for ultrasonic signals based on memory-enhanced deep unfolding network is proposed in this paper, which is intended to accurately reconstruct high-frequency ultrasonic signals in flip chip detection. To address information loss during the iterative process in deep unfolding networks, two memory enhancement mechanisms are designed. In order to mitigate information loss between adjacent stages, a multi-channel high-throughput memory mechanism is proposed, which introduces high-throughput information from the previous stage into the current stage, thus promoting the transmission of information between adjacent stages and ensuring maximum information flow. A cross-stage memory enhancement mechanism is applied to promote information fusion between different stages, which is based on Richardson extrapolation and calculates the output error between every two adjacent stages in order to update the sparse coefficient of the current stage. The proposed method achieves an average increase of 13.75 dB in signal-to-noise ratio at different levels of noise, with an average reduction of 20.56 in root mean square error. The results demonstrate that the proposed method can effectively captures weak echo features while removing the noise, achieving satisfactory denoising performance.