Hu, BaoxinJung, Won Mo2020-11-132020-11-132020-072020-11-13http://hdl.handle.net/10315/37902Information on the severity of pavement cracks is critical for pavement repair services. This study aimed to exploit the applications of deep learning networks to improve the detection and classification of pavement cracks. An improved Convolution Neural Network (CNN) with structured prediction was proposed. Also, Fully Convolutional Network (FCN), U-Net, and attention U-Net were implemented and explored with different optimizers and loss functions. The developed networks were tested on the data collected in Ontario, Canada with the purpose of localizing the cracks and identifying their severity levels based on three categories (low, medium, and high). The results showed that the improved CNN approach performed better than its original version with the F1-score increased from (5.21%, 30.85%, 83.51%) to (19.63%, 55.60%, 85.89%) for the detection of the cracks with the three severities. Furthermore, FCN, U-Net, and attention U-Net achieved slightly better results than the improved CNN approach with the F1-scores of (32.08%, 68.82%, 89.89%), (40.06%, 69.97%, 89.07%), and (40.53%, 71.27%, 89.95%), respectively.Author owns copyright, except where explicitly noted. Please contact the author directly with licensing requests.Environmental scienceElectronic Thesis or Dissertation2020-11-13Crack DetectionPavementConvolution Neural NetworkCNNFully Convolutional NetworkFCNPavement distressInstance segmentationU-NetAttention U-NetDeep LearningMachine LearningArtificial Intelligence (AI)Image processingComputer Vision