An Improved EfficientDet-D3 Model for Internal Crack Detection in Bridge GPR Images

Internal crack detection in bridge roadway structures is critical for bridge health monitoring and maintenance. Ground Penetrating Radar (GPR), as a non-destructive testing (NDT) technique, has a wide range of applications in the identification of internal cracks in bridges. However, due to the complexity of the radar echo signals and the uncertainty of crack characteristics, existing methods still have challenges in detection accuracy and robustness. To this end, this paper proposes a crack detection method for GPR images based on improved EfficientDet-D3. We introduce the MoHAttention (Mixture of Heads Attention) mechanism to enhance the feature extraction capability and use an optimized feature fusion module in BiFPN (Bidirectional Feature Pyramid Network) to capture the crack information more effectively. Experimental results show that the method outperforms the existing mainstream target detection methods in AP<inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">50</inf>, AP<inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">75</inf>, and AR metrics, especially in recognizing small-scale cracks with higher accuracy and robustness. This study provides an efficient and accurate solution for automated crack detection with GPR images, which provides a new technical support for structural health monitoring of bridges.