CN114266716A - Method for detecting leakage water risk of shield tunnel wall surface based on deep learning - Google Patents
Method for detecting leakage water risk of shield tunnel wall surface based on deep learning Download PDFInfo
- Publication number
- CN114266716A CN114266716A CN202010967797.8A CN202010967797A CN114266716A CN 114266716 A CN114266716 A CN 114266716A CN 202010967797 A CN202010967797 A CN 202010967797A CN 114266716 A CN114266716 A CN 114266716A
- Authority
- CN
- China
- Prior art keywords
- classification
- risk
- model
- wall surface
- trained
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 title claims abstract description 51
- 238000000034 method Methods 0.000 title claims abstract description 45
- 238000013135 deep learning Methods 0.000 title claims abstract description 38
- 238000013145 classification model Methods 0.000 claims abstract description 59
- 230000011218 segmentation Effects 0.000 claims abstract description 46
- 238000001514 detection method Methods 0.000 claims abstract description 43
- 238000012549 training Methods 0.000 claims abstract description 35
- 238000004040 coloring Methods 0.000 claims abstract description 18
- 238000013528 artificial neural network Methods 0.000 claims description 22
- 238000013527 convolutional neural network Methods 0.000 claims description 20
- 230000006870 function Effects 0.000 claims description 13
- 238000002790 cross-validation Methods 0.000 claims description 10
- 230000008569 process Effects 0.000 claims description 7
- 238000012545 processing Methods 0.000 claims description 6
- 238000005070 sampling Methods 0.000 claims description 6
- 238000013526 transfer learning Methods 0.000 claims description 6
- 230000003287 optical effect Effects 0.000 claims description 5
- 230000004913 activation Effects 0.000 claims description 3
- 238000004519 manufacturing process Methods 0.000 claims description 3
- 238000011176 pooling Methods 0.000 claims description 3
- 230000000694 effects Effects 0.000 abstract description 9
- 238000012544 monitoring process Methods 0.000 abstract description 9
- 201000010099 disease Diseases 0.000 abstract description 6
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 abstract description 6
- 238000012423 maintenance Methods 0.000 abstract description 4
- 238000010586 diagram Methods 0.000 description 12
- 238000012360 testing method Methods 0.000 description 8
- 230000003321 amplification Effects 0.000 description 4
- 238000003199 nucleic acid amplification method Methods 0.000 description 4
- 238000004364 calculation method Methods 0.000 description 3
- 238000009826 distribution Methods 0.000 description 3
- 238000002372 labelling Methods 0.000 description 2
- 238000005192 partition Methods 0.000 description 2
- 230000009471 action Effects 0.000 description 1
- 230000003213 activating effect Effects 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 238000003745 diagnosis Methods 0.000 description 1
- 230000036541 health Effects 0.000 description 1
- 238000003058 natural language processing Methods 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
Images
Landscapes
- Image Analysis (AREA)
Abstract
The invention provides a method for detecting the leakage water risk of the wall surface of a shield tunnel based on deep learning, which is used for detecting the leakage water risk of the wall surface of the shield tunnel according to a detection image obtained by shooting the wall surface of the shield tunnel and comprises the following steps: the method comprises the steps of carrying out classification type judgment on detection images through a classification model to obtain classification types which are risk-free, risk-processed and risk-unprocessed, positioning and coloring the detection images of which the classification types are risk-unprocessed to obtain coloring label images, outputting the coloring label images to provide warning effect and realize automatic positioning of the warned monitoring images, and training according to a tunnel wall classification data set and a tunnel wall segmentation data set. Therefore, the tunnel maintenance requirement aiming at the rapid increase can be reduced, the requirements of practical engineering on automatic and batch treatment of water leakage disease image recognition are met by sending related warning and automatic positioning, and the water leakage detection of the tunnel wall surface is more accurate and efficient.
Description
Technical Field
The invention relates to a method for detecting the risk of leakage water on the wall surface of a shield tunnel based on deep learning, and belongs to the technical field of tunnel risk detection.
Background
With the vigorous development of the economy of China, the convenience and diversification requirements of cities on traffic are higher and higher, and the construction of subway tunnels and highway river-crossing tunnels is more and more. Along with the rapidly increased tunnel maintenance requirement, the rapid and accurate identification and diagnosis of the subway shield tunnel structure diseases, particularly the water leakage diseases, are urgently needed. The health detection of the shield tunnel by using the computer vision is a new trend at home and abroad in recent years, theoretically, the effect of higher robustness can be realized with very low cost, but the identification effect of the leakage water disease image at present cannot meet the requirement of actual engineering.
Deep learning is one of the major breakthroughs obtained in the field of artificial intelligence in the last decade, and has great success in the fields of speech recognition, natural language processing, computer vision, image and video analysis, multimedia and the like.
The current tunnel wall leakage water risk detection method is not manual detection, but positioning by adopting a segmentation method, and the automatic realization that warning can be provided and positioning can be carried out on the warning is not available.
Disclosure of Invention
In order to solve the problems, the invention provides a shield tunnel wall leakage water risk detection method based on deep learning, which adopts the following technical scheme:
the invention provides a method for detecting the risk of leakage water on the wall surface of a shield tunnel based on deep learning, which is used for detecting the risk of leakage water on the wall surface of the shield tunnel according to a detection image obtained by shooting the wall surface of the shield tunnel and is characterized by comprising the following steps: step 1-1, classifying the detection images through a pre-trained classification model to obtain classification categories corresponding to the detection images, wherein the classification categories comprise no risk, processed risk and unprocessed risk; step 1-2, judging classification types; step 1-3, when the classification category is judged to be risk-free and risk-treated, outputting a risk-free result; step 1-4, when the classification type is judged to be at risk unprocessed, carrying out water leakage positioning on the detection image of which the classification type is at risk unprocessed through a pre-trained segmentation model, and processing the detection image to obtain a coloring label image; step 1-5, outputting coloring label graph to perform early warning, wherein the training process of classifying models and segmenting models comprises the following steps: step 2-1, performing down-sampling on a tunnel wall optical image collected in advance to a uniform size to manufacture a tunnel wall classification data set; step 2-2, the tunnel wall surface classification data sets are respectively split into tunnel wall surface classification data sets which contain risk unprocessed, risk processed and risk-free, the portions which are risk unprocessed are marked with risk or not through a segmentation marking tool, and the marked portions are further used as tunnel wall surface segmentation data sets; 2-3, training based on the tunnel wall surface classification data set to obtain a classification model; and 2-4, training to obtain a classification model based on the tunnel wall segmentation data set.
The method for detecting the water leakage risk of the shield tunnel wall surface based on the deep learning, provided by the invention, can also have the technical characteristics that the step 2-3 comprises the following substeps: step 2-3-1, a convolutional neural network and model parameters of the convolutional neural network are called, and the convolutional neural network is obtained by obtaining a large-scale open source data set based on transfer learning; step 2-3-2, modifying the last full connection layer of the convolutional neural network into a global pooling module, a1 x 1 convolutional kernel structure and a classifier as a classification model to be trained; step 2-3-3, calculating the classification probability of the classification model to be trained according to the classifier in the classification model to be trained on the tunnel wall surface classification data set; 2-3-4, taking the class corresponding to the maximum classification probability as a classification class; 2-3-5, updating model parameters of the classification model to be trained by using a cross entropy loss function in a random gradient descending manner; 2-3-6, calculating the accuracy of classification categories obtained by the classification model and the model training speed; 2-3-7, training the classification model to be trained by adjusting the hyper-parameters of the classification model to be trained and by a cross validation method; and 2-3-8, repeating the steps 2-3-3 to 2-3-7, and outputting the model parameter with the highest accuracy and the classification model to be trained corresponding to the hyper-parameter as a classification model.
The method for detecting the water leakage risk of the shield tunnel wall surface based on the deep learning, provided by the invention, can also have the technical characteristics that the step 2-4 comprises the following substeps: step 2-4-1, calling a full convolution neural network and model parameters of the full convolution neural network, wherein the model parameters of the full convolution neural network and the full convolution neural network are obtained by obtaining a large-scale open source data set based on transfer learning; step 2-4-2, removing the last full connection layer of the full convolution neural network, applying the model parameters of the full convolution neural network to the full convolution neural network, and judging the category of each pixel through a sigmoid activating function of an output layer so as to obtain a segmentation model to be trained; 2-4-3, training a segmentation model to be trained through a tunnel wall segmentation data set, and updating model parameters by adopting a random gradient descent strategy through a cross entropy loss function; 2-4-4, adjusting hyper-parameters of the segmentation model to be trained to perform cross validation and calculating the accuracy of the segmentation model to be trained corresponding to the current hyper-parameters; and 2-4-5, repeating the steps 2-4-3 and 2-4-4, and outputting the segmentation model to be trained corresponding to the hyper-parameters and the model parameters with the highest accuracy in the cross validation as the segmentation model.
Action and Effect of the invention
According to the method for detecting the water leakage risk of the shield tunnel wall based on the deep learning, the monitoring image is classified and classified through a classification model, then classification classes containing no risk, processed risk and unprocessed risk are obtained, then the detection image classified and classified as unprocessed risk is further subjected to water leakage positioning and coloring through a segmentation model so as to obtain a coloring label image, finally the coloring label image is output so as to provide a warning effect and realize automatic positioning of the warned monitoring image, wherein the classification model and the segmentation model make a tunnel wall optical image into a tunnel wall classification dataset by reducing and sampling the uniform size, the made tunnel wall classification dataset is divided into three parts of unprocessed risk, processed risk and unprocessed risk, and the unprocessed risk part is marked as the tunnel wall segmentation dataset, thereby performing training. Therefore, the method for detecting the risk of the leakage water on the wall surface of the shield tunnel based on the deep learning can reduce the requirement for tunnel maintenance which is increased sharply, meet the requirement of the actual engineering on the identification of the leakage water disease image by sending out a warning about the condition of risk untreated in the monitoring image and automatically positioning the risk untreated part in the monitoring image, and realize the automatic learning function more accurately and more efficiently, thereby realizing the automatic and batch processing of the leakage water detection image on the wall surface of the tunnel.
Drawings
Fig. 1 is a flowchart of a method for detecting the risk of water leakage on the wall surface of a shield tunnel based on deep learning in the embodiment of the present invention;
fig. 2 is a schematic diagram of three classification categories in the method for detecting the risk of water leakage on the wall surface of the shield tunnel based on deep learning in the embodiment of the present invention;
FIG. 3 is a flowchart of a training classification model of a method for detecting the risk of water leakage on the wall surface of a shield tunnel based on deep learning in an embodiment of the present invention;
fig. 4 is a flowchart of a training segmentation model of a shield tunnel wall leakage risk detection method based on deep learning in the embodiment of the present invention.
Fig. 5 is a data distribution of a tunnel wall segmentation data set of the method for detecting the risk of water leakage on the wall surface of the shield tunnel based on deep learning in the embodiment of the present invention;
FIG. 6 is a diagram of relevant model parameters of a classification model in a method for detecting the risk of water leakage on the wall surface of a shield tunnel based on deep learning according to an embodiment of the present invention;
FIG. 7 is a diagram illustrating a relevant model hyper-parameter condition of a classification model in a method for detecting the risk of water leakage on the wall surface of a shield tunnel based on deep learning according to an embodiment of the present invention;
fig. 8 is a schematic diagram illustrating a result of test 1 in the method for detecting the risk of water leakage on the wall surface of the shield tunnel based on deep learning according to the embodiment of the present invention; and
fig. 9 is a schematic diagram of a result of test 2 in the method for detecting the risk of water leakage on the wall surface of the shield tunnel based on deep learning in the embodiment of the present invention.
Detailed Description
In order to make the technical means, creation features, achievement purposes and effects of the method easy to understand, the method for detecting the water leakage risk of the shield tunnel wall surface based on deep learning is specifically described below with reference to the embodiments and the accompanying drawings.
< example >
Fig. 1 is a flowchart of a method for detecting the risk of water leakage on the wall surface of a shield tunnel based on deep learning in the embodiment of the present invention.
As shown in fig. 1, a method for detecting the risk of water leakage on the wall surface of a shield tunnel based on deep learning specifically includes steps 1-1 to 1-5.
Step 1-1, classifying the detection images through a pre-trained classification model to obtain classification categories corresponding to the detection images, wherein the classification categories comprise risk-free, risk-processed and risk-free, and then entering step 1-2.
Fig. 2 is a schematic diagram of three classification categories in the shield tunnel wall leakage risk detection method based on deep learning in the embodiment of the present invention.
In this embodiment, the detection image is an image of leakage water on the wall surface of the shield tunnel obtained by shooting the wall surface of the shield tunnel through an instrument, the shooting operation may be initiated by an operator instruction to be sent, or the system may periodically perform a check to autonomously initiate, and forward reasoning is performed on the detection image through a classification model to obtain a classification category corresponding to the detection image, as shown in fig. 2, fig. 2 (a) shows that the classification category is a detection image at risk and unprocessed, fig. 2 (b) shows that the classification category is a detection image at risk and fig. 2 (c) shows that the classification category is a detection image at risk and is a detection image at no risk.
And step 1-2, judging classification types. When the classification category is judged to be risk-free and risk-treated, entering step 1-3; when the classification category is determined to be at risk unprocessed, then step 1-5 is entered.
And 1-3, outputting a risk-free result and entering an ending state.
In this embodiment, the output risk-free result indicates that the wall surface subjected to the risk detection of water leakage on the wall surface of the shield tunnel does not have water leakage, and the wall surface does not need to be processed.
Step 1-4, performing water leakage positioning on the detection images with risk and no processing classification types through a pre-trained segmentation model, processing the detection images to obtain a coloring label map, and then entering step 1-5.
The coloring label image is an image which can visually locate the leakage position by coloring the detection image according to the inherent label in the image.
The steps of the training process of the classification model and the segmentation model include steps 2-1 to 2-4.
And 2-1, performing down-sampling on the pre-collected tunnel wall optical image to a uniform size, performing sample amplification to manufacture a tunnel wall classification data set, and then entering the step 2-2.
In this example, the size obtained by down-sampling is 432px by 648 px.
Fig. 4 is a data distribution of a tunnel wall classification dataset of the method for detecting the risk of water leakage on the wall surface of the shield tunnel based on deep learning in the embodiment of the present invention.
The sample amplification of the tunnel wall classification dataset is realized through operations of symmetry, turning, noise addition and the like, and the data distribution of the tunnel wall classification dataset obtained through the sample amplification is shown in fig. 4.
Step 2-2, the tunnel wall surface classification data sets are respectively split into tunnel wall surface classification data sets which comprise risk unprocessed, risk processed and risk-free data and are subjected to sample amplification, the risk unprocessed parts are marked with or without risks through a segmentation marking tool, the marked parts are further used as tunnel wall surface segmentation data sets, and then the step 2-3 is carried out.
In this embodiment, the used segmentation labeling tool is Matlab, and the generated single-channel picture is formed by labeling and coloring the risk area of each picture with unprocessed risk.
And 2-3, training to obtain a classification model based on the tunnel wall surface classification data set, and then entering the step 2-4.
Fig. 2 is a flowchart of a training classification model of a shield tunnel wall leakage risk detection method based on deep learning in the embodiment of the present invention.
As shown in FIG. 2, the training process of the classification model in step 2-3 includes steps 2-3-1 through 2-3-8.
And 2-3-1, calling a convolutional neural network and model parameters of the convolutional neural network, wherein the model parameters of the convolutional neural network and the convolutional neural network are obtained by obtaining a large-scale open source data set based on transfer learning, and then entering the step 2-3-2.
Fig. 5 is a diagram of relevant model parameters of a classification model in a shield tunnel wall leakage risk detection method based on deep learning in an embodiment of the present invention.
In this embodiment, the called convolutional neural network is a keras deep learning framework and a densenert classification network, the model parameters of the convolutional neural network are parameters of a pre-training model obtained by training an ImageNet public data set, the model parameters of the called convolutional neural network are set as shown in fig. 5, and the accuracy rate obtained by verifying the tunnel wall classification data set through the classification model under the model parameter setting is 98%.
And 2-3-2, modifying the last full connection layer of the convolutional neural network into a global pooling module, taking a1 x 1 convolutional kernel structure and a classifier as a classification model to be trained, and then entering the step 2-3-3.
And 2-3-3, calculating the tunnel wall surface classification data set according to the classifier in the classification model to be trained to obtain the classification probability of the classification model to be trained, and then entering the step 2-3-4.
In this embodiment, the classifier in the classification model to be trained is softmax, and a total of n classification classes S are setk k∈(0,n]Then, the calculation formula of the softmax result is:
where i represents a certain classification category of the k classification categories, giA value representing the classification category.
And 2-3-4, taking the class corresponding to the maximum classification probability as a classification class, and then entering the step 2-3-5.
And 2-3-5, updating model parameters of the classification model to be trained in a random gradient descending mode by using a cross entropy loss function, and then entering the step 2-3-6.
And 2-3-6, calculating the accuracy of the classification type obtained by the classification model to be trained, and then entering the step 2-3-7.
And 2-3-7, training the classification model to be trained by adjusting the hyper-parameters of the classification model to be trained and by a cross validation method, and then entering the step 2-3-8.
And 2-3-8, repeating the steps 2-3-3 to 2-3-7, outputting the model parameter with the highest accuracy and the classification model to be trained corresponding to the hyper-parameter as a classification model, and then entering an ending state.
Fig. 6 shows a relevant model hyper-parameter condition of a classification model in the method for detecting the risk of water leakage on the wall surface of the shield tunnel based on deep learning in the embodiment of the present invention.
The hyper-parameter training is performed under the model parameter setting, the relevant model hyper-parameter setting is shown in fig. 6, and the accuracy of the classification model under the hyper-parameter setting is 83%.
And 2-4, training to obtain a classification model based on the tunnel wall segmentation data set, and then entering an ending state.
Fig. 3 is a flowchart of a training segmentation model of a shield tunnel wall leakage risk detection method based on deep learning in an embodiment of the present invention.
As shown in FIG. 3, the training process of the classification model in step 2-4 includes step 2-4-1 to step 2-4-5.
And 2-4-1, calling a full convolution neural network and model parameters of the full convolution neural network, wherein the full convolution neural network and the model parameters of the full convolution neural network are obtained by obtaining a large-scale open source data set based on transfer learning, and then entering the step 2-4-2.
In this embodiment, the called full convolution neural network is a network partitioned by a keras deep learning framework and a deplab 3, and the adopted open source data set is an ImageNet data set.
And 2-4-2, removing the last full connection layer of the full convolution neural network, applying the model parameters of the full convolution neural network to the full convolution neural network, judging the category of each pixel through a sigmoid activation function of an output layer so as to obtain a segmentation model to be trained, and then entering the step 2-4-3.
And 2-4-3, training the segmentation model to be trained through the tunnel wall segmentation data set, updating model parameters by adopting a random gradient descent strategy through a cross entropy loss function, and then entering the step 2-4-4.
In this embodiment, the tunnel wall partition data set is marked with a color on the risk area with risk of unprocessed of each picture through a Matlab partition marking tool, so as to generate a single-channel picture, such single-channel picture is converted into a color-marked label graph, and the name of the generated single-channel picture and the name of the label graph are as follows: a10.png — > A10_ label.
And 2-4-4, adjusting the hyper-parameters of the segmented model to be trained, performing cross validation, calculating the accuracy of the segmented model corresponding to the current hyper-parameters, and then entering the step 2-4-5.
And 2-4-5, repeating the steps 2-4-3 and 2-4-4, outputting the segmentation model to be trained corresponding to the hyper-parameters and the model parameters with the highest accuracy in the cross validation as the segmentation model, and then entering an ending state.
And 1-5, outputting a coloring label graph so as to perform early warning, and ending the process.
Fig. 7 is a schematic diagram illustrating a result of test 1 in the method for detecting the risk of water leakage on the wall surface of the shield tunnel based on deep learning in the embodiment of the present invention.
Fig. 8 is a schematic diagram illustrating a result of test 2 in the method for detecting the risk of water leakage on the wall surface of the shield tunnel based on deep learning in the embodiment of the present invention.
As shown in fig. 7 and 8, fig. 7 and 8 are schematic diagrams illustrating test results of a test image by a method for detecting the risk of water leakage on the wall surface of the shield tunnel based on deep learning, in which fig. 7 (a) and 8 (a) are diagrams of test images at risk, and fig. 7 (b) and 8 (b) are diagrams of test images at risk.
Examples effects and effects
According to the method for detecting the water leakage risk of the shield tunnel wall surface based on the deep learning, classification judgment is carried out on a monitoring image through a classification model, then classification classes containing no risk, processed risk and unprocessed risk are obtained, then the detection image of which the classification class is unprocessed at risk is further subjected to water leakage positioning and coloring through a segmentation model so as to obtain a coloring label image, finally the coloring label image is output so as to provide a warning effect and realize automatic positioning of the warned monitoring image, wherein the classification model and the segmentation model make a tunnel wall surface classification data set by carrying out down sampling on an optical image of the tunnel wall surface to a uniform size, and the made tunnel wall surface classification data set is divided into three parts of unprocessed at risk, processed risk and no risk, and marking the part with risks unprocessed as a tunnel wall segmentation data set so as to train. Therefore, the method for detecting the leakage water risk of the wall surface of the shield tunnel based on the deep learning can reduce the requirement for tunnel maintenance which is increased sharply, and can enable a detector to process related warning areas in time more quickly by sending a warning about the condition that the risk is not processed in a monitoring image and automatically positioning the risk-not-processed part in the monitoring image, thereby meeting the requirement of actual engineering on the identification of leakage water disease images, realizing the automatic learning function more accurately and more efficiently, and further realizing the automatic and batch processing of the leakage water detection images of the wall surface of the tunnel.
In the embodiment, when a classification model is trained, a convolutional neural network is called and a last full connection layer is removed, so that training time and training cost can be reduced in a mode of training an initialization parameter through a model, meanwhile, through a mode of descending a random gradient, calculation overhead caused by loss calculation through a cross entropy loss function can be reduced, a model with high precision and good generalization performance can be obtained by adjusting the adjusted classification model through a mode of adjusting a hyper-parameter, and therefore, the method is more reliable and can be better applied to practical engineering, and finally, overfitting can be reduced to a certain extent through a cross verification mode and effective information as much as possible can be obtained from limited data.
In the embodiment, when a segmentation model is trained, a full convolutional neural network is called to realize model training initialization parameters by removing a final full connection layer and modifying a sigmoid activation function of an output layer, so that training time and training cost are reduced, meanwhile, a method for adjusting training hyper-parameter setting and adjusting a training submodule structure is adjusted, so that a convolutional neural network classification model with higher efficiency, good generalization performance and high precision is obtained, and the convolutional neural network classification model is more reliable and can be better applied in practical engineering, and finally overfitting can be reduced to a certain extent by a cross validation method and effective information as much as possible can be obtained from limited data.
The above-described embodiments are merely illustrative of specific embodiments of the present invention, and the present invention is not limited to the description of the above-described embodiments.
Claims (3)
1. The utility model provides a shield tunnel wall leakage water risk detection method based on degree of depth study for carry out shield tunnel wall leakage water risk detection according to the detection image that obtains of shooting shield tunnel wall, its characterized in that includes:
step 1-1, classifying the detection image through a pre-trained classification model to obtain classification categories corresponding to the detection image, wherein the classification categories comprise no risk, processed risk and unprocessed risk;
step 1-2, judging the classification type;
step 1-3, when the classification category is judged to be risk-free and risk-treated, outputting a risk-free result;
step 1-4, when the classification type is judged to be at risk unprocessed, carrying out water leakage positioning on the detection image with the classification type at risk unprocessed through the pre-trained segmentation model, and processing the detection image to obtain a coloring label map;
step 1-5, outputting the coloring label graph to perform early warning,
the training process of the classification model and the segmentation model comprises the following steps:
step 2-1, performing down-sampling on a tunnel wall optical image collected in advance to a uniform size to manufacture a tunnel wall classification data set;
step 2-2, the tunnel wall surface classification data sets are respectively split into tunnel wall surface classification data sets which contain risk unprocessed, risk processed and risk-free, the risk unprocessed parts are marked with or without risks through a segmentation marking tool, and the marked parts are further used as tunnel wall surface segmentation data sets;
step 2-3, training based on the tunnel wall surface classification data set to obtain the classification model;
and 2-4, training to obtain the classification model based on the tunnel wall segmentation data set.
2. The method for detecting the water leakage risk of the shield tunnel wall surface based on the deep learning of claim 1 is characterized in that:
wherein, the step 2-3 comprises the following substeps:
step 2-3-1, a convolutional neural network and model parameters of the convolutional neural network are called, and the convolutional neural network is obtained by obtaining a large-scale open source data set based on transfer learning;
step 2-3-2, modifying the last full connection layer of the convolutional neural network into a global pooling module, a1 x 1 convolutional kernel structure and a classifier as a classification model to be trained;
2-3-3, calculating the tunnel wall surface classification data set according to the classifier in the classification model to be trained to obtain the classification probability of the classification model to be trained;
step 2-3-4, the category corresponding to the maximum classification probability is taken as the classification category;
2-3-5, updating the model parameters of the classification model to be trained by using a cross entropy loss function in a random gradient descending manner;
2-3-6, calculating the accuracy of the classification categories obtained by the classification model;
2-3-7, training the classification model to be trained by adjusting the hyper-parameters of the classification model to be trained and a cross validation method;
and 2-3-8, repeating the steps 2-3-3 to the steps 2-3-7, and outputting the model parameter with the highest accuracy and the classification model to be trained corresponding to the hyper-parameter as classification models.
3. The method for detecting the risk of water leakage on the wall surface of the shield tunnel based on the deep learning of claim 1, which is characterized in that:
wherein, the step 2-4 comprises the following substeps:
step 2-4-1, calling a full convolution neural network and the model parameters of the full convolution neural network, wherein the model parameters of the full convolution neural network and the full convolution neural network are obtained by obtaining a large-scale open source data set based on transfer learning;
step 2-4-2, removing the last full connection layer of the full convolution neural network, applying the model parameters of the full convolution neural network to the full convolution neural network, and judging the category of each pixel through a sigmoid activation function of an output layer so as to obtain a segmentation model to be trained;
2-4-3, training the segmentation model to be trained through the tunnel wall segmentation data set, and updating model parameters by adopting a random gradient descent strategy through a cross entropy loss function;
2-4-4, adjusting the hyper-parameters of the segmentation model to be trained to perform cross validation and calculating the accuracy of the segmentation model to be trained corresponding to the current hyper-parameters;
and 2-4-5, repeating the steps 2-4-3 and the steps 2-4-4, and outputting the segmentation model to be trained corresponding to the hyper-parameters and the model parameters with the highest accuracy in cross validation as the segmentation model.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010967797.8A CN114266716B (en) | 2020-09-15 | 2020-09-15 | Method for detecting leakage water risk of shield tunnel wall based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010967797.8A CN114266716B (en) | 2020-09-15 | 2020-09-15 | Method for detecting leakage water risk of shield tunnel wall based on deep learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114266716A true CN114266716A (en) | 2022-04-01 |
| CN114266716B CN114266716B (en) | 2024-04-02 |
Family
ID=80824141
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010967797.8A Active CN114266716B (en) | 2020-09-15 | 2020-09-15 | Method for detecting leakage water risk of shield tunnel wall based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114266716B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116935016A (en) * | 2023-07-28 | 2023-10-24 | 上海济目科技有限公司 | Integrated tunnel construction site risk identification and laser positioning method |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2019136946A1 (en) * | 2018-01-15 | 2019-07-18 | 中山大学 | Deep learning-based weakly supervised salient object detection method and system |
| CN110909657A (en) * | 2019-11-19 | 2020-03-24 | 上海勘察设计研究院(集团)有限公司 | Method for identifying apparent tunnel disease image |
| CN111507990A (en) * | 2020-04-20 | 2020-08-07 | 南京航空航天大学 | Tunnel surface defect segmentation method based on deep learning |
| CN111612758A (en) * | 2020-05-19 | 2020-09-01 | 南京航空航天大学 | Tunnel surface defect detection method based on convolutional neural network |
-
2020
- 2020-09-15 CN CN202010967797.8A patent/CN114266716B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2019136946A1 (en) * | 2018-01-15 | 2019-07-18 | 中山大学 | Deep learning-based weakly supervised salient object detection method and system |
| CN110909657A (en) * | 2019-11-19 | 2020-03-24 | 上海勘察设计研究院(集团)有限公司 | Method for identifying apparent tunnel disease image |
| CN111507990A (en) * | 2020-04-20 | 2020-08-07 | 南京航空航天大学 | Tunnel surface defect segmentation method based on deep learning |
| CN111612758A (en) * | 2020-05-19 | 2020-09-01 | 南京航空航天大学 | Tunnel surface defect detection method based on convolutional neural network |
Non-Patent Citations (1)
| Title |
|---|
| 薛亚东;高健;李宜城;黄宏伟;: "基于深度学习的地铁隧道衬砌病害检测模型优化", 湖南大学学报(自然科学版), no. 07, 21 July 2020 (2020-07-21) * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116935016A (en) * | 2023-07-28 | 2023-10-24 | 上海济目科技有限公司 | Integrated tunnel construction site risk identification and laser positioning method |
| CN116935016B (en) * | 2023-07-28 | 2024-03-08 | 上海济目科技有限公司 | Integrated tunnel construction site risk identification and laser positioning method |
Also Published As
| Publication number | Publication date |
|---|---|
| CN114266716B (en) | 2024-04-02 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111275688B (en) | Small target detection method based on context feature fusion screening of attention mechanism | |
| CN108830188B (en) | Vehicle detection method based on deep learning | |
| CN109949316B (en) | Power grid equipment image weak supervision example segmentation method based on RGB-T fusion | |
| CN110070074B (en) | Method for constructing pedestrian detection model | |
| CN109871875B (en) | Building change detection method based on deep learning | |
| CN110532961B (en) | Semantic traffic light detection method based on multi-scale attention mechanism network model | |
| Wen et al. | Deep learning-based ash content prediction of coal flotation concentrate using convolutional neural network | |
| CN112614119A (en) | Medical image region-of-interest visualization method, device, storage medium and equipment | |
| CN112613428B (en) | Resnet-3D convolution cattle video target detection method based on balance loss | |
| CN110659601B (en) | Depth full convolution network remote sensing image dense vehicle detection method based on central point | |
| Ajagbe et al. | Investigating the efficiency of deep learning models in bioinspired object detection | |
| CN112686242B (en) | Fine-grained image classification method based on multilayer focusing attention network | |
| Chen et al. | Road damage detection and classification using mask R-CNN with DenseNet backbone | |
| CN112949510A (en) | Human detection method based on fast R-CNN thermal infrared image | |
| CN114266716A (en) | Method for detecting leakage water risk of shield tunnel wall surface based on deep learning | |
| Shui et al. | End-to-end cell recognition by point annotation | |
| Chen et al. | An adaptive hybrid attention based convolutional neural net for intelligent transportation object recognition | |
| CN116883650A (en) | Image-level weak supervision semantic segmentation method based on attention and local stitching | |
| CN115458174A (en) | Method for constructing intelligent diagnosis model of diabetic retinopathy | |
| Shishkin et al. | Implementation of yolov5 for detection and classification of microplastics and microorganisms in marine environment | |
| Tumuluru et al. | SMS: SIGNS MAY SAVE–Traffic Sign Recognition and Detection using CNN | |
| CN115223001A (en) | Medical image identification method based on transfer learning | |
| Kovács et al. | Hungarian Traffic Sign Detection and Classification using Semi-Supervised Learning | |
| Suvetha et al. | Automatic Traffic Sign Detection System With Voice Assistant | |
| Srivastava | Retraining of object detectors to become suitable for trash detection in the context of autonomous driving |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |