US 12,450,718 B1
Quantification method for microscopic cracks inside 3D printed concrete and system thereof
Baidian Li, Nanchang (CN); Xiangyu Wang, Nanchang (CN); Junbo Sun, Changzhou (CN); Fei Wu, Nanchang (CN); Jianqun Wang, Xiangtan (CN); Yangqing Liu, Nanchang (CN); Bo Huang, Xiangtan (CN); Weixiang Shi, Nanchang (CN); Qiaoming Guo, Nanchang (CN); and Yufei Wang, Nanchang (CN)
Assigned to Jiangxi Transportation Investment Group Co., Ltd., Nanchang (CN); East China Jiaotong University, Nanchang (CN); and Liyang Institute of Smart City, Chongqing University, Changzhou (CN)
Filed by Jiangxi Transportation Investment Group Co., Ltd., Nanchang (CN); East China Jiaotong University, Nanchang (CN); and Liyang Institute of Smart City, Chongqing University, Changzhou (CN)
Filed on May 9, 2025, as Appl. No. 19/203,267.
Claims priority of application No. 202410906818.3 (CN), filed on Jul. 8, 2024.
Int. Cl. G06T 7/00 (2017.01); G06T 5/70 (2024.01); G06T 7/13 (2017.01); G06V 10/26 (2022.01); G06V 10/30 (2022.01); G06V 10/762 (2022.01); G06V 10/82 (2022.01)
CPC G06T 7/0004 (2013.01) [G06T 5/70 (2024.01); G06T 7/13 (2017.01); G06V 10/26 (2022.01); G06V 10/30 (2022.01); G06V 10/763 (2022.01); G06V 10/82 (2022.01); G06T 2207/20028 (2013.01); G06T 2207/20081 (2013.01); G06T 2207/20084 (2013.01); G06T 2207/30132 (2013.01)] 9 Claims
OG exemplary drawing
 
1. A quantification method for microscopic cracks inside 3D printed concrete, comprising the following steps:
1) Preparing a microscopic crack image dataset
using backscattered electron (BSE) to obtain defect image data of printing areas of 3D printed concrete, forming the microscopic crack image dataset, wherein the microscopic crack image dataset comprises 3D printed concrete microscopic crack images, non 3D printed concrete microscopic crack images, and concrete images without microscopic cracks;
2) Treating of improved attention-guided denoising neural network (LADNet)
constructing an improved attention-guided denoising neural network (IADNet), wherein the improved attention-guided denoising neural network (IADNet) comprises a sparse block (SB), a feature enhancement block (FEB), an attention block (AB), and a reconstruction block (RB);
the sparse block (SB) comprises five convolutional layers and four dilated convolutional layers, each layer comprising a Leaky ReLU function stacked with convolution or dilated convolution, connected in a order of one convolutional layer, one dilated convolutional layer, two convolutional layers, two dilated convolutional layers, two convolutional layers, and one dilated convolutional layer;
the feature enhancement block (FEB) comprises a convolutional layer, a convolution operation, a compression concatenation operation, a convolution operation, and a hyperbolic tangent activation function connected in sequence; an output result of the sparse block (SB) is used as an input of the feature enhancement block (FEB);
the attention block (AB) comprises 1×1 convolution operation and vector dot product operation; an input of the 1×1 convolution operation is connected to an output processed by the hyperbolic tangent activation function of the feature enhancement block (FEB), and features obtained by the 1×1 convolution operation are used as weight vectors to perform vector dot product operation with an output of a first convolution operation in the feature enhancement block (FEB); an output of the vector dot product operation is connected to the reconstruction block (RB) to obtain an output denoised image;
training the improved attention-guided denoising neural network (IADNet) using the microscopic crack image dataset to obtain a trained improved attention-guided denoising neural network (IADNet), then applying the trained improved attention-guided denoising neural network (IADNet) to denoise the images to construct a crack segmentation dataset;
3) Constructing MCR-Former network for quantitative measurement of microscopic crack images
training the MCR-Former network using the crack segmentation dataset to obtain a trained MCR-Former neural network;
performing denoising treatment using the trained improved attention-guided denoising neural network (IADNet), then inputting vector data into the trained MCR-Former neural network to obtain a microscopic crack segmentation result, and then further calculating a microscopic crack width based on the microscopic crack segmentation result.