Abstract: This paper proposes a baseline-free probability imaging method integrating density clustering and semantic weighting to overcome the reliance on defect-free reference signals and boundary artifacts in traditional air-coupled Lamb wave imaging. Full-path response signals are collected via orthogonal scanning. A joint feature vector is constructed using wavelet low-frequency approximation coefficients and a normalized symmetric difference factor, followed by density-based spatial clustering of applications with noise (DBSCAN) to unsupervisedly classify scan paths into “healthy” or “damaged” states. The resulting semantic labels are embedded as adaptive weights into an improvedreconstruction algorithm for probabilistic inspection of damage (RAPID) model, suppressing artifact paths and dynamically constructing a soft baseline. Experiments on CFRP delamination defects show that the method, requiring no prior baseline, reduces the average size measurement error by 50.5% compared to traditional RAPID. For a 40 mm×20 mm×0.05 mm defect, measurement accuracy in the X/Y directions has improved by 81.7% and 65.5%, respectively, while effective identification of a 20 mm×20 mm×0.05 mm defect is achieved. Stable imaging performance is maintained even at a 5 dB SNR. This method effectively eliminates the baseline dependency and enhances detection accuracy and robustness, demonstrating significant engineering potential.

Key words: baseline-free imaging, density clustering, semantic weighting, air-coupled Lamb waves, CFRP, boundary artifact suppression