Automatic Identification and Quantitative Research of Pavement Cracks based on UAV Images and U-Net

Abstract

Aiming at the problems of low efficiency, poor safety and insufficient quantitative accuracy of traditional manual pavement crack detection, an automatic identification and quantitative method for pavement cracks integrating UAV aerial photography technology and improved U-Net deep learning model is proposed. The DJI Phantom 4 RTK UAV is used to collect high-resolution images under different pavement types and lighting conditions. A dataset containing 2000 original images is constructed with pixel-level annotation, and expanded to 8000 images through data augmentation technology to improve the generalization ability of the model. The Convolutional Block Attention Module (CBAM) is embedded in the skip connections of the U-Net benchmark model, and a composite loss function combining Dice Loss and Focal Loss is designed to solve the problems of class imbalance and background interference. Experimental results show that the improved CBAM-U-Net model achieves a mean Intersection over Union (mIoU) of 85.3% and an F1-Score of 89.7% on the test set, which is significantly better than comparative methods such as Canny edge detection, FCN and basic U-Net. Based on the segmentation results, accurate quantification of crack length and width is realized through morphological processing, skeleton extraction and geometric calculation, with length error of ±0.82% and width error of ±3.65%, and an integrated prototype system is developed. This method provides efficient and accurate data support for pavement maintenance decisions and has important engineering application value.