A Dual-Branch Artifact-Resistant Network for Air-Coupled Ultrasonic Localization Imaging of Composite Damage

Composite materials are prone to low-velocity impact during service. This impact causes invisible internal damage and threatens structural safety. Air-coupled ultrasonic Lamb wave testing offers non-contact advantages. However, line scanning sparsity and anisotropic noise challenge traditional imaging methods. These challenges affect reconstruction efficiency and anti-artifact capability. To address this, this paper proposes a Dual-branch Artifact-Resistant Fast Imaging Network (DARF-Net). The network achieves end-to-end localization and reconstruction of composite damage. The method performs line scans along 0° and 90° directions. It feeds Lamb wave signals from two orthogonal directions into independent encoding branches. This extracts dual-view complementary features. The bottleneck layer uses element-wise addition fusion. It integrates dual-view information, enhances true damage responses, and suppresses noise. Residual blocks are embedded in the decoder to improve gradient propagation and reconstruction details. The network output undergoes morphological post-processing to suppress residual artifacts. Experiments use composite plate samples. Results show a mean center positioning deviation of 0.133 mm. The mean absolute diameter deviation is 0.188 mm. The method maintains high localization accuracy for damage offset 10 mm from the training grid. Single-image inference takes 14.79 ms. Validation is performed on calibrated surrogate damage produced by coupled mass blocks. The proposed method balances imaging accuracy and inference efficiency. It provides a feasible solution for rapid field evaluation of composite structures.