Activation Civil 3D 2019 ##VERIFIED## Crack

Recently, with the development of machine learning classified as deep learning inspired by structure of the brain called artificial neural networks (ANN) [45], many algorithms have been proposed to perform object detection and image classification tasks. ANN is employed to solve many civil engineering problems [46,47,48,49,50]. Gao and Mosalam in [51] applied the transfer learning to detect damage images with structural method, and this method can reduce the computational cost by using the pre-trained neural network model. Meanwhile, the author needs to fine the neural network to perform the crack detection. Local patch information was employed to inspect crack information by convolutional neural networks (CNN) in [52]. In CrackNet [53], the algorithm improved pixel-perfect accuracy based on CNN by discarding pooling layers. In CrackNet-R [54], a recurrent neural network (RNN) is deployed to perform automatic crack detection on asphalt road. Cha et al. [55] adopted a sliding windows based on CNN to scan and detect road crack. Fan et al. in [56] proposed a structured prediction method to detect crack pixels with CNN. The small structured pixel images (27 27 pixels) was input into the neural network, which may generate overload for the computer memory. Ensemble network is proposed to perform crack detection and measure pavement cracks generated in road pavement [57]. Maeda et al. on [58] adopted object detection network architecture to detect crack images, and the network architecture can be transferred to a smartphone to perform road crack detection. Cha et al. used the Faster-RCNN to inspect road cracks [59]. Yang et al. in [60] adopted a fully convolutional network (FCN) to inspect road pavement cracks at pixel level, which can perform crack detection by end-to-end training. Li et al. in [61] employed the you-only-look-once v3 (YOLOv3)-Lite method to inspect the aircraft structures, and the depth wise separable convolution and feature pyramid were adopted to design the network architecture and joined the low- and high-resolution for crack detection. Jenkins et al. presented an encoder-decoder architecture to perform road crack detection, and the function of the encoder and decoder layers are used to reduce the size of input image to generate lower level feature maps, and obtain the resolution of the input data with up-sampling, respectively [62]. Tisuchiya et al. proposed a data augmentation method based on YOLOv3 to perform crack detection, which can increase the accuracy effectively [63].

Activation Civil 3D 2019 Crack