Tianjie Zhang – Data Science
Title: TJ Zhang’s Comprehensive Exam
Pavement crack segmentation using deep learning methods can improve crack segmentation accuracy, but in many cases, the training dataset is lacking or uneven, making it insufficient to train an accurate segmentation model. In this work, an integrated APC-GAN and AttuNet framework is proposed as an automated pavement surface crack pixel-level segmentation strategy for small training datasets. First, an automated pavement cracks generative adversarial network (APC-GAN) is designed for the pavement cracks data as an image augmentation method, which is modified and improved from a traditional deep convolutional generative adversarial network (DCGAN). Then, a novel pixel-level semantic segmentation structure, Attunet, is proposed by introducing the attention module into the convolutional network structure. Another AttuNet-min is modified by replacing the max pooling layer and activation function in AttuNet. In order to assess the performance of APC-GAN and AttuNet framework, an open-source dataset DeepCrack is used, which only contains 300 training images. The results show that APC-GAN can produce clearer and more distinct pavement images than DCGAN and more diversity than the traditional augmentation method. The AttuNet model with APC-GAN can reach higher accuracy in evaluation metrics than other augmentation methods. As for the segmentation model comparison, APC-GAN and AttuNet framework gain the highest value in recall, F1 score, mean Intersection over Union (mIoU) and mean pixel accuracy (mPA) among all models, including U-Net, DeepLabv3, FCN and LRASPP, while the AttuNet-min gain the highest mean precision.