To address the inaccurate detection of highway cracks by current models

an intelligent detection method based on remote sensing images from unmanned aerial vehicle(UAV)aerial sensors is proposed.Based on the U-Net architecture

the DAR-Unet pixel-wise crack detection model is developed by combining the deeply separable residual block(DR-Block)

the spatial pyramid fusion attention module(SPFAM)

and the receptive field block(RFB).A dedicated dataset is constructed using 1046 high-quality highway remote sensing images collected by UAVs.On this self-developed dataset

DAR-Unet achieves a mean intersection over union(mIoU)and an F1-score of 76.41% and 74.24%

respectively

which are higher than those of mainstream models.The model is further integrated with UAVs to construct an Internet of Things-based highway crack detection system.Practical tests show excellent performance

verifying the effectiveness of DAR-Unet in highway crack detection using remote sensing images.