Automatic Pixel-level pavement sealed crack detection using Multi-fusion U-Net network

The Multi-fusion U-Net network based on U-Net is proposed to attain pixel-level detection of sealed cracks. The multi-fusion module, dual attention mechanism, and Atrous Spatial Pyramid Pooling (ASPP) are designed to efficiently capture the details of sealed cracks. The 3163 image set is divided into training, validation, and testing datasets. The training data consist of 2463 image sets. The Multi-fusion U-Net outperforms U-Net and DANet during the training process. The test experimental results indicate that the F-measure and IOU of the Multi-fusion U-Net on the 200 test images are 84.36 % and 72.95 % respectively. Compared with seven state-of-the-art models (i.e., DANet, MACSNet, U-Net, SegNet, DeepLabV3+, PSPNet, SegFormer), the proposed network exhibits higher detection accuracy on the 200 testing images. The average time to process the images for all networks was 49.78 ms/frame, and the proposed network processed the images in 57 ms/frame. Real-time detection of sealed cracks is feasible.