ABSTRACT
This paper presents an effective method for intelligent detection of pavement distresses. Firstly, the unmanned aerial vehicle (UAV) is used to obtain original images of pavement distress. The images are grayscale binarized by MATLAB. Then, three algorithms based on improved convolutional neural networks (CNN), including Visual Geometry Group (VGG), Resident Net(ResNet), and GoogLENet, are used to investigate the detection accuracy of pavement distresses based on Pytorch. The results show that each neural network can identify and classify pavement distresses. The GoogLENet model has the best training performance with an accuracy of 98% and a loss degree of only 0.3%. It is recommended that the GoogLENet model has a stronger capacity for pavement distress identification and classification.
