Composite materials are widely employed in modern industries such as aerospace, biomedical, and automotive sectors due to their superior mechanical properties. However, these materials are susceptible to defects introduced during manufacturing or service. To address the need for efficient, automated, and non-destructive damage detection, this study proposes a K-means-aided wavenumber-adaptive wavefield filtering technique for highlighting local anomalies in the wavefield images of composite plates. The method eliminates the need for prior knowledge of material properties or manual tuning of processing parameters. In the experiments, guided waves were excited using a piezoelectric transducer and measured by a scanning laser Doppler vibrometer over a dense spatial measurement grid. The proposed approach involves wavefield images preprocessing, automatic mask generation via K-means clustering, and post-filtering in both the wavenumber and spatial domains. A stability enhancer function and Gaussian smoothing were incorporated to improve clustering robustness and reduce artifacts. Validation through both numerical simulations and experimental tests on glass fiber-reinforced polymer specimens with various PZT excitation positions demonstrates the method’s high accuracy in detecting and localizing artificial defects. The technique maintains consistent performance regardless of the PZT placement. This work presents a fully automated, parameter-free technique for defect visualization in composite structures, significantly enhancing the practicality and robustness of guided-wave-based non-destructive testing systems.