CN-122023397-A - Bridge crack detection method based on deep learning

CN122023397ACN 122023397 ACN122023397 ACN 122023397ACN-122023397-A

Abstract

The invention relates to the technical field of bridge crack identification and discloses a bridge crack detection method based on deep learning, which comprises the steps of presetting an equidistant grid path along a bridge pavement, collecting an original image of the bridge pavement of each grid, carrying out pixel-level crack marking after filtering and enhancing treatment to generate a binary label image and a bridge pavement image after filtering and enhancing treatment, constructing an improved U-Net model, taking the bridge pavement image as the input of the improved U-Net model, training the improved U-Net model by taking the binary label image as the output, optimizing the trained improved U-Net model by adopting a mixed loss function to generate a trained improved U-Net model, acquiring an original image of the bridge pavement to be detected, and inputting the trained improved U-Net model to carry out crack detection after filtering and enhancing treatment to obtain a crack identification result.

Inventors

ZHAO YONGFEI
PU YUAN
CHEN ZHIPING
YU HUALI
DAI QIWEI
ZHOU XUEFU
MIN JIAN
FANG BIN
WANG RUIJI
HUANG YAN
YANG CHENGLONG

Assignees

四川交大工程检测咨询有限公司

Dates

Publication Date: 20260512
Application Date: 20260410

Claims (8)

1. The bridge crack detection method based on deep learning is characterized by comprising the following steps of: s1, presetting equidistant grid paths along a bridge pavement, collecting an original image of the bridge pavement of each grid, performing pixel-level crack marking after filtering and reinforcing treatment, and generating a binary label graph and a bridge pavement image after filtering and reinforcing treatment; s2, an improved U-Net model is built, and the improved U-Net model comprises a multi-scale feature extraction module, a multi-scale feature fusion module and a decoder; S3, taking the bridge pavement image as the input of an improved U-Net model, taking the binary label image as the output, training the improved U-Net model, optimizing the trained improved U-Net model by adopting a mixed loss function, and generating a trained improved U-Net model; S4, acquiring an original image of the bridge pavement to be detected, inputting a trained improved U-Net model to detect cracks after filtering and enhancing treatment, judging whether cracks exist, outputting a crack binary segmentation map if yes, and otherwise, judging that the bridge pavement to be detected does not exist.
2. The deep learning based bridge fracture detection method of claim 1, wherein the multi-scale feature extraction module comprises a first convolution block, a second convolution block, a third convolution block, a fourth convolution block, a first downsampling module, a second downsampling module, a third downsampling module, a fourth downsampling module, a first fracture saliency sub-network, a second fracture saliency sub-network, a third fracture saliency sub-network, and a fourth fracture saliency sub-network.
3. The method for detecting bridge cracking based on deep learning according to claim 2, wherein the decoder comprises a first deconvolution block, a second deconvolution block, a third deconvolution block, a fifth deconvolution block, a sixth convolution block, a seventh convolution block, and an output layer.
4. The method for detecting bridge cracks based on deep learning according to claim 3, wherein the step S3 specifically comprises: S31, taking a bridge pavement image as input data, taking a binary label graph as a supervision label, inputting the input data into a multi-scale feature extraction module of an improved U-Net model for multi-scale feature extraction, and generating a multi-scale feature graph and a crack response value corresponding to the multi-scale feature graph; S32, inputting the multi-scale feature map and the corresponding crack response value into a multi-scale feature fusion module to perform feature fusion, and generating a fusion feature map; S33, inputting the fusion feature map into a decoder for deconvolution, performing jump connection and convolution operation on the fusion feature map and the feature map of the corresponding scale of the multi-scale feature extraction module, gradually up-sampling, and finally outputting a crack binary segmentation map; s34, calculating a mixed loss function of the Dice loss and the cross entropy loss, and taking the mixed loss function as a mixed loss value of the crack binary segmentation graph and the binary label graph; And S35, updating network parameters of the improved U-Net model by using the mixed loss value to obtain a trained improved U-Net model.
5. The method for detecting the bridge crack based on the deep learning according to claim 4, wherein the process of inputting the input data into the multi-scale feature extraction module of the improved U-Net model to extract the multi-scale features and generating the multi-scale feature map and the corresponding crack response value thereof is as follows: After input data is input into a first convolution block to carry out convolution, batch normalization and activation operations, the input data is input into a first downsampling module to carry out downsampling, a first scale feature map is generated, meanwhile, the first scale feature map is input into a first crack significance sub-network to carry out convolution, global average pooling and activation operations, and crack response values of the first scale feature map are generated by extracting crack semantic features; Inputting the first scale feature map into a second convolution block for convolution, batch normalization and activation operation, inputting the first scale feature map into a second downsampling module for downsampling to generate a second scale feature map, inputting the second scale feature map into a second crack significance sub-network for convolution, global average pooling and activation operation, and generating a crack response value of the second scale feature map by extracting crack semantic features; Inputting the second scale feature map into a third convolution block for convolution, batch normalization and activation operation, inputting the second scale feature map into a third downsampling module for downsampling to generate a third scale feature map, inputting the third scale feature map into a third crack significance sub-network for convolution, global average pooling and activation operation, and generating a crack response value of the third scale feature map by extracting crack semantic features; and inputting the third scale feature map into a fourth convolution block for convolution, batch normalization and activation operation, inputting the third scale feature map into a fourth downsampling module for downsampling to generate a fourth scale feature map, inputting the fourth scale feature map into a fourth crack significance sub-network for convolution, global average pooling and activation operation, and extracting crack semantic features to generate a crack response value of the fourth scale feature map.
6. The method for detecting a bridge crack based on deep learning according to claim 5, wherein the step S32 specifically comprises: Normalizing the crack response values of the multi-scale feature map by adopting a softmax function to obtain the self-adaptive weight of each scale feature map, namely: Wherein, the Represent the first The attention weight of the scale feature map, 、 Respectively represent the first First, the The fracture response values of the scale feature map, Representing an exponential function; And carrying out weighted summation on the multi-scale feature map and the self-adaptive weight corresponding to the multi-scale feature map to generate a fusion feature map, namely: Wherein, the The fusion profile is represented by a graph of features, Represent the first And (5) a scale characteristic diagram.
7. The method for detecting bridge cracks based on deep learning according to claim 6, wherein step S33 specifically comprises: s331, inputting the fusion feature map into a first deconvolution block for deconvolution operation, then splicing the fusion feature map with a third-scale feature map to obtain a first splicing feature, inputting the first splicing feature into a fifth deconvolution block for convolution, batch normalization and activation operation to obtain a first feature map; s332, inputting the first characteristic diagram into a second deconvolution block for deconvolution operation, then splicing the first characteristic diagram with the second scale characteristic diagram to obtain a second spliced characteristic, and inputting the second spliced characteristic into a sixth deconvolution block for convolution, batch normalization and activation operation to obtain a second characteristic diagram; S333, inputting the second characteristic diagram into a third convolution block for deconvolution operation, then splicing the second characteristic diagram with the first scale characteristic diagram to obtain a third spliced characteristic, inputting the third spliced characteristic into a seventh convolution block for convolution, batch normalization and activation operation to obtain a third characteristic diagram; S334, performing convolution operation on the third feature map input/output layer to obtain a crack binary segmentation map.
8. The method for detecting bridge cracks based on deep learning according to claim 7, wherein the formula for calculating the mixed loss function of the Dice loss and the cross entropy loss is: Wherein, the The mixing loss function is represented by a function of the mixing loss, Representing the Dice loss function, Representing the cross-entropy loss function, Representing pixel points Is used to determine the true value of (a), A logarithmic function is represented and is used to represent, Representing pixel points Is used to determine the predicted value of (c), Representing a smooth term.

Description

Bridge crack detection method based on deep learning Technical Field The invention relates to the technical field of bridge crack identification, in particular to a bridge crack detection method based on deep learning. Background The bridge pavement cracks are important expression forms of bridge structure damage, and timely and accurately detecting crack information is important for bridge maintenance and prolonging the service life of the bridge. At present, two methods of traditional manual detection and automatic detection based on image processing are mainly used for bridge pavement crack detection. The traditional manual detection relies on the field visual observation of detection personnel, has low efficiency and high labor intensity, is easily influenced by subjective experience and environmental illumination, and has high small crack omission rate. In the automatic detection method based on image processing, the traditional algorithms such as threshold segmentation, edge detection and the like are adopted in the early method, the interference resistance to illumination change, road surface stains and the like is weak, the crack segmentation precision is low, the traditional method based on deep learning mostly directly adopts a general segmentation model such as an original U-Net, is not optimized aiming at the characteristics of slender and discontinuous bridge road surface cracks and low contrast ratio, the multi-scale feature extraction is insufficient, the small crack feature is easily submerged by the background, and therefore the extraction precision is poor, and the engineering practical requirements are difficult to meet. Disclosure of Invention Aiming at the defects in the prior art, the invention provides a bridge crack detection method based on deep learning, which is used for solving the problem of low detection precision of the existing bridge pavement crack detection method. In order to achieve the aim of the invention, the invention adopts the following technical scheme: The bridge crack detection method based on deep learning comprises the following steps: s1, presetting equidistant grid paths along a bridge pavement, collecting an original image of the bridge pavement of each grid, performing pixel-level crack marking after filtering and reinforcing treatment, and generating a binary label graph and a bridge pavement image after filtering and reinforcing treatment; s2, an improved U-Net model is built, and the improved U-Net model comprises a multi-scale feature extraction module, a multi-scale feature fusion module and a decoder; S3, taking the bridge pavement image as the input of an improved U-Net model, taking the binary label image as the output, training the improved U-Net model, optimizing the trained improved U-Net model by adopting a mixed loss function, and generating a trained improved U-Net model; S4, acquiring an original image of the bridge pavement to be detected, inputting a trained improved U-Net model to detect cracks after filtering and enhancing treatment, judging whether cracks exist, outputting a crack binary segmentation map if yes, and otherwise, judging that the bridge pavement to be detected does not exist. The invention has the following beneficial effects: The bridge crack detection method based on deep learning provided by the invention realizes automatic crack identification by systematically acquiring images through a preset grid path and combining an improved U-Net model, remarkably reduces manual inspection cost and time, enhances the perceptibility of the model to micro cracks and complex backgrounds through multi-scale feature extraction and self-adaptive fusion, realizes accurate positioning of the cracks, ensures that the model outputs clear crack-free judgment or crack segmentation results through binary label graph supervision training, reduces the probability of missing detection and false detection, and finally realizes quick, accurate and automatic detection of bridge pavement cracks, and provides reliable technical support for infrastructure maintenance. Drawings Fig. 1 is a schematic flow chart of a bridge crack detection method based on deep learning according to the present invention; FIG. 2 is a schematic diagram of an original image of a bridge Liang Lumian according to an example; FIG. 3 is a schematic illustration of labeling of an original image of a bridge Liang Lumian in an embodiment; FIG. 4 is a schematic diagram of the structure of the improved U-Net model in the embodiment; FIG. 5 is a schematic structural diagram of a multi-scale feature extraction module according to an embodiment; Fig. 6 is a schematic diagram of a decoder in an embodiment. Detailed Description The following description of the embodiments of the present invention is provided to facilitate understanding of the present invention by those skilled in the art, but it should be understood that the present invention is not limited to the scope of the embodiments