Efficient Denoising Method for Ultrasonic Signals Using IA-ICEEMDAN-VME

Ultrasonic testing is a widely used non-destructive technique for detecting internal defects in materials. In practical applications, ultrasonic signals often suffer from low signal-to-noise ratios (SNR) due to reflection, transmission, and refraction within the tested object, causing defect-related echoes to be obscured and reducing detection accuracy. Exiting decomposition-based denoising methods typically rely on fixed or adaptive thresholds to classify Intrinsic Mode Functions (IMFs), and subsequently reconstruct the signal based on the selected effective IMFs. However, IMFs near the threshold are difficult to distinguish as either informative signal or noise, leading to misclassification and suboptimal reconstruction. To address this challenge, we propose IA-ICEEMDAN-VME, a denoising framework tailored for low-SNR ultrasonic signals. It employs a Pearson correlation coefficient-guided residual adaptive iterative strategy to directly extract high-value IMFs without thresholding. The residual from each iteration is reconstructed as the input for the next iteration, gradually building an optimal set of IMFs. To further suppress potential residual noise within selected IMFs, Variational Mode Extraction (VME) is applied, enhancing signal clarity and fidelity. Experiments on both simulated and real-world datasets demonstrate that IA-ICEEMDAN-VME outperforms existing methods such as ICEEMDAN and ICEEMDAN-wavelet, achieving higher SNR and lower RMSE under noisy conditions. The proposed method offers a robust solution for ultrasonic non-destructive testing in low-SNR environments, enabling more accurate recovery of defect signals.