SwinTrans CrackNet: A multi-scale frequency domain fusion framework for bridge crack detection

Crack detection is a key task in bridge structure health monitoring. Accurate segmentation of crack areas is of great significance for bridge safety assessment and maintenance. However, in complex backgrounds, existing methods still face challenges such as difficulty in identifying small cracks and false detection caused by texture interference. To this end, this paper uses Swin Trans-UNet as the baseline model to optimize the crack segmentation task. Firstly, multi-scale feature fusion is introduced to perform crack fusion at different resolutions to enhance the model's perception of cracks of different scales and improve the detection effect of small cracks. Secondly, combined with frequency domain features, discrete Fourier transform (DFT) is used to convert the image into frequency domain space to extract high-frequency information, thereby enhancing the network's ability to express crack features and improving segmentation accuracy. Finally, Dice Loss is introduced as a loss function to further optimize the model training process and improve the segmentation effect. Experimental results show that the improved model has better detection performance and robustness in the crack segmentation task, providing more accurate technical support for bridge structure health monitoring.