ABSTRACT Microwave nondestructive testing (NDT) techniques show promise for composite inspection due to microwave signals’ ability to penetrate and interact with internal structures. However, current microwave imaging approaches have poor spatial resolution, struggling to distinguish defects from defect-free regions. This limits reliable subsurface analysis and widespread adoption. This paper introduces a novel multi-frequency microwave imaging fusion method using an open-ended waveguide and deep learning to enhance defect detection accuracy in carbon fiber-reinforced polymer (CFRP) composites. The proposed technique employs an optimized feature extraction strategy to improve differentiation between defective and sound areas for superior subsurface visualization. The method leverages VGG-19 for efficient feature extraction and parallel processing to merge information from multi-frequency images. Our method significantly improved detection accuracy and F1 score, surpassing non-deep learning image fusion techniques by at least 50%. The optimized fusion strategy enables clear visualization of various defect types across multiple subsurface layers. Our approach also reduces computation time versus standard VGG implementation by 3–4×, showing scalability. The results demonstrate the proposed method’s potential to overcome existing constraints and provide rapid, accurate subsurface analysis of complex CFRP structures through an optimized deep learning framework, holding significance for expanding microwave NDE applications.
Read full abstract