CN114742795A - Tunnel lining disease intelligent detection method based on deep learning - Google Patents
Tunnel lining disease intelligent detection method based on deep learning Download PDFInfo
- Publication number
- CN114742795A CN114742795A CN202210374768.XA CN202210374768A CN114742795A CN 114742795 A CN114742795 A CN 114742795A CN 202210374768 A CN202210374768 A CN 202210374768A CN 114742795 A CN114742795 A CN 114742795A
- Authority
- CN
- China
- Prior art keywords
- deep learning
- disease
- learning model
- layer
- region
- 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.)
- Withdrawn
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Computational Linguistics (AREA)
- Software Systems (AREA)
- Mathematical Physics (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computing Systems (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Quality & Reliability (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The invention discloses a tunnel lining disease intelligent detection method based on deep learning, which comprises the following steps of S1: making a tunnel lining disease data set formed by marked disease image samples; s2: dividing disease image samples in a tunnel lining disease data set into a training set and a testing set; s3: building a deep learning model; s4: training and parameter iteration are carried out on the built deep learning model through a training set, the detection effect of the trained deep learning model is evaluated through a test set, and the finally applied deep learning model is selected; s5: and detecting diseases in the tunnel image by using the selected deep learning model, and outputting the disease category and position information. According to the method, the tunnel image is used as a data source, the tunnel apparent diseases are intelligently identified and classified through the deep learning model, and the disease positions are positioned, so that the working efficiency is greatly improved compared with the conventional method which depends on manual disease judgment and evaluation.
Description
Technical Field
The invention relates to the technical field of tunnel defect detection, in particular to a tunnel lining defect detection method.
Background
The tunnel is affected by construction quality, operation age and external environment, and can generate cracks, deformation, damage, block falling, water leakage and other diseases, and threatens the safety, stability and durability of the tunnel structure.
The type and the development degree of the diseases are important indexes for evaluating the tunnel safety, the detection of the diseases at present mainly depends on manual judgment and evaluation, the subjectivity is strong, the efficiency is low, and the tunnel diseases are intelligentized, continuously and quickly identified.
Disclosure of Invention
In view of the above, the invention aims to provide an intelligent detection method for tunnel lining diseases based on deep learning, so as to solve the technical problems of strong subjectivity and low efficiency caused by manual disease judgment in the prior art.
The invention relates to a tunnel lining disease intelligent detection method based on deep learning, which comprises the following steps:
s1: acquiring images of a tunnel vault, a tunnel arch and side walls, manually detecting to select images with diseases, cutting the images with the diseases, marking the types and the position information of the diseases, and then manufacturing a tunnel lining disease data set formed by marked disease image samples;
s2: dividing disease image samples in a tunnel lining disease data set into a training set and a testing set;
s3: building a deep learning model, wherein the deep learning model comprises an input layer, a coding layer connected with the input layer, a position sensitive analysis layer connected with the coding layer, a non-maximum linear inhibition layer connected with the position sensitive analysis layer and an output layer connected with the non-maximum linear inhibition layer;
the input layer is used for receiving picture input;
the coding layer is used for processing the input picture to extract an interested area;
the position sensitivity analysis layer is used for carrying out position sensitivity score mapping and bounding box regression on the generated region of interest;
the non-maximum linear inhibition layer is used for filtering the output of the position sensitive analysis layer;
the input layer is used for outputting a disease category predicted value and a disease positioning predicted value of the region of interest;
s4: training and parameter iteration are carried out on the built deep learning model through a training set, the detection effect of the trained deep learning model is evaluated through a test set, and the deep learning model which is finally applied is selected;
s5: and detecting diseases in the tunnel image by using the selected deep learning model, and outputting the disease category and position information.
Further, the location sensitivity score mapping in step S3 specifically includes:
the first step is as follows: the region of interest is divided into k × k grids, and the size of each grid is
Wherein w is the width of the region of interest, h is the length of the region of interest, and pooling operation on a certain grid is defined as:
wherein r isc(i, j) is the response value of the c-th class under the (i, j) -th lattice, 0 ≦ i, j ≦ k-1; z is a radical ofi,j,cFinger k2One of (C +1) score plots, (x)0,y0) The coordinate value of the upper left corner of the interesting area is shown, n is the number of pixel points in the grid, and theta is a parameter of the network;
the second step is that: calculating the score of each category on the interested region, and averaging to obtain
Further, in step S3, the position-sensitive analysis layer divides each region of interest into 3 × 3 grids, and generates 9 position-sensitive score maps for each object type using one convolution layer, where the 9 score maps respectively describe relative position information of top left, top middle, top right, middle left, middle right, bottom left, bottom middle, and bottom right of one type, so as to obtain an output of 9 × channels (3+1), where 3 represents the number of disease types, and 1 represents the background;
and the position sensitive analysis layer is also used for boundary frame regression through a convolution layer with 4 x 9 channels, and finally, a 4-dimensional positioning prediction value is obtained after each region of interest passes through the position sensitive analysis layer.
Further, in step S4, when the deep learning model is trained, an end-to-end method is adopted, and an algorithm of error back propagation and stochastic gradient descent is used, where a loss function of each batch of images during training is:
L(s,lx,y,w,h)＝Lcls(Sc*)+λ[c*＞0]Lreg(t,t*)
Lcls(Sc*)＝-log(Sc*)
wherein the content of the first and second substances,
as a function of classification loss, Lreg(t, t) is a regression loss function, using Smooth-L1, c is the real box class of the region of interest, t is the real box position information of the region of interest, λ ═ 1; c ═ 0 denotes the background, [ c ≧ 0 ] when c is the object class]Otherwise, the value is 0, namely when the category is judged to be background, the regression operation is not carried out,
is a category response.
Further, in step S4, the accuracy and the recall ratio index are used as the accuracy index of the test data, and the deep learning model with the best recall ratio index is selected as the model of the final application.
The invention has the beneficial effects that:
according to the tunnel lining disease intelligent detection method based on deep learning, tunnel images are used as data sources, apparent tunnel diseases can be intelligently identified and classified, the positions of the diseases can be positioned, and compared with the existing method which depends on manual judgment and disease evaluation, the working efficiency is greatly improved.
Drawings
Fig. 1 is a structural diagram of a deep learning model.
Detailed Description
The invention is further described below with reference to the figures and examples.
As shown in the figure, the intelligent tunnel lining disease detection method based on deep learning in the embodiment includes the following steps:
s1: acquiring images of the vault, the arch waist and the side wall of the tunnel, manually detecting to select images with diseases, cutting the images with the diseases, marking the types and the position information of the diseases, and then manufacturing a tunnel lining disease data set formed by marked disease image samples.
In the step, a tunnel detection vehicle is adopted to carry out non-interference rapid detection on a plurality of tunnels, the detection vehicle adopts 3 cameras to scan tunnel linings, the three cameras respectively collect images of tunnel vault, arch waist and side wall, the cameras adopt 4K line scanning cameras and adopt a half mode to collect, and the rapid collection of mass high-definition images of the tunnel linings can be realized.
In the step, according to the collected image, after manual detection and confirmation, common diseases and the proportion thereof are analyzed, and the type of the detected diseases, such as cracks, water seepage or peeling, is determined.
In the step, disease category and position information are marked by LabelIMG software.
The tunnel lining disease data set in this step contains 3 ten thousand picture samples.
S2: and dividing disease image samples in the tunnel lining disease data set into a training set and a testing set. In this step, the training set accounts for 90% of the total number of samples, and the testing set accounts for 10% of the total number of samples.
S3: and building a deep learning model, wherein the deep learning model comprises an input layer, a coding layer connected with the input layer, a position sensitive analysis layer connected with the coding layer, a non-maximum linear inhibition layer connected with the position sensitive analysis layer and an output layer connected with the non-maximum linear inhibition layer.
The input layer is used for accepting picture input, and the coding layer is used for processing the input picture to extract an interested area.
The position sensitive analysis layer is used for carrying out position sensitive score mapping and bounding box regression on the generated region of interest. The position sensitivity score mapping specific process is as follows:
the first step is as follows: the region of interest is divided into k × k grids, each grid having a size of
Wherein w is the width of the region of interest, h is the length of the region of interest, and pooling operation on a certain grid is defined as:
wherein r isc(i, j) is the response value of the c-th class under the (i, j) -th lattice, 0 ≦ i, j ≦ k-1; z is a radical ofi,j,cFinger k2One of (C +1) score plots, (x)0,y0) The coordinate value of the upper left corner of the interested area is obtained, n is the number of pixel points in the grid, and theta is a network parameter. In this step, each region of interest is specifically divided into 3 × 3 grids, and a convolution layer is used to generate 9 position-sensitive score maps for each object type, where the 9 score maps respectively describe the relative position information of the top left, top middle, top right, middle left, middle right, bottom left, bottom middle, and bottom right of a type, and obtain 9 × 1 channel output, where 3 represents the number of disease types, and 1 represents the background.
The second step is that: calculating the score of each category on the interested region, and averagingIs worthy of
And the position sensitive analysis layer is also used for boundary frame regression through a convolution layer with 4 x 9 channels, and finally, a 4-dimensional positioning prediction value is obtained after each region of interest passes through the position sensitive analysis layer.
The non-very large linear rejection layer is used to filter the output of the position sensitive analysis layer.
The input layer is used for outputting a disease category predicted value and a disease positioning predicted value of the region of interest.
S4: and training and parameter iteration are carried out on the built deep learning model through a training set, the detection effect of the trained deep learning model is evaluated through a test set, and the deep learning model which is finally applied is selected.
In this step, when the deep learning model is trained, an end-to-end method is adopted, an algorithm of error back propagation and random gradient descent is used, and a loss function of each batch of images during training is as follows:
L(s,lx,y,w,h)＝Lcls(Sc*)+λ[c*＞0]Lreg(t,t*)
Lcls(Sc*)＝-log(Sc*)
wherein the content of the first and second substances,
as a function of classification loss, Lreg(t, t) is a regression loss function, using Smooth-L1, c is the real box class of the region of interest, t is the real box position information of the region of interest, λ ═ 1; c ═ 0 denotes the background, [ c ≧ 0 ] when c is the object class]Otherwise, the value is 0, namely when the category is judged to be background, the regression operation is not carried out,
is a category response.
In the step, accuracy and recall rate indexes are adopted as precision indexes of test data, and a deep learning model with the best recall rate index is selected as a model for final application.
S5: and detecting diseases in the tunnel image by using the selected deep learning model, and outputting the disease category and position information.
According to the tunnel lining disease intelligent detection method based on deep learning, tunnel images are used as data sources, tunnel apparent diseases are intelligently identified and classified through a deep learning model, the positions of the diseases are located, and compared with the existing method which depends on manual judgment and disease evaluation, the method greatly improves the working efficiency.
Finally, the above embodiments are only for illustrating the technical solutions of the present invention and not for limiting, although the present invention has been described in detail with reference to the preferred embodiments, it should be understood by those skilled in the art that modifications or equivalent substitutions may be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all of them should be covered in the claims of the present invention.
Claims (5)
1. A tunnel lining disease intelligent detection method based on deep learning is characterized in that: the method comprises the following steps:
s1: acquiring images of a tunnel vault, a tunnel arch waist and a side wall, manually detecting to select an image with a disease, cutting the disease image, marking the disease type and position information, and then manufacturing a tunnel lining disease data set formed by marked disease image samples;
s2: dividing disease image samples in a tunnel lining disease data set into a training set and a testing set;
s3: building a deep learning model, wherein the deep learning model comprises an input layer, a coding layer connected with the input layer, a position sensitive analysis layer connected with the coding layer, a non-maximum linear inhibition layer connected with the position sensitive analysis layer and an output layer connected with the non-maximum linear inhibition layer;
the input layer is used for receiving picture input;
the coding layer is used for processing the input picture to extract an interested area;
the position sensitivity analysis layer is used for carrying out position sensitivity score mapping and bounding box regression on the generated region of interest;
the non-maximum linear inhibition layer is used for filtering the output of the position sensitive analysis layer;
the input layer is used for outputting a disease category predicted value and a disease positioning predicted value of the region of interest;
s4: training and parameter iteration are carried out on the built deep learning model through a training set, the detection effect of the trained deep learning model is evaluated through a test set, and the deep learning model which is finally applied is selected;
s5: and detecting diseases in the tunnel image by using the selected deep learning model, and outputting the disease category and position information.
2. The intelligent tunnel lining disease detection method based on deep learning as claimed in claim 1, characterized in that: the position sensitivity score mapping in step S3 specifically includes:
the first step is as follows: the region of interest is divided into k × k grids, and the size of each grid is
Wherein w is the width of the region of interest, h is the length of the region of interest, and pooling operation on a certain grid is defined as:
wherein r isc(i, j) is the response value of the c-th class under the (i, j) -th lattice, 0 ≦ i, j ≦ k-1; z is a radical ofi,j,cFinger k2One of (C +1) score plots, (x)0,y0) For the upper left corner of the region of interestCoordinate values, n is the number of pixel points in the grid, and theta is a parameter of the network;
the second step is that: calculating the score of each category on the region of interest, and averaging to obtain
3. The intelligent tunnel lining disease detection method based on deep learning as claimed in claim 2, characterized in that: in step S3, the position sensitive analysis layer divides each region of interest into 3 × 3 grids, and generates 9 position sensitive score maps for each object type using one convolution layer, where the 9 score maps respectively describe the relative position information of the upper left, upper middle, upper right, middle left, middle right, lower left, lower middle, and lower right of one type, so as to obtain 9 × 1 channel output, where 3 represents the number of disease types, and 1 represents the background;
and the position sensitive analysis layer is also used for boundary frame regression through a convolution layer with 4 x 9 channels, and finally, a 4-dimensional positioning prediction value is obtained after each region of interest passes through the position sensitive analysis layer.
4. The intelligent tunnel lining disease detection method based on deep learning as claimed in claim 1, characterized in that: in step S4, when the deep learning model is trained, an end-to-end method is adopted, and an algorithm of error back propagation and random gradient descent is used, where a loss function of each batch of images during training is:
wherein the content of the first and second substances,
as a function of classification loss, Lreg(t, t) is a regression loss function, using Smooth-L1, c is the real box class of the region of interest, t is the real box position information of the region of interest, λ ═ 1; c ═ 0 denotes the background, [ c ≧ 0 ] when c is the object class]If not, 0, namely when the category is judged as background, no regression operation is carried out, Sc*Is a category response.
5. The intelligent tunnel lining disease detection method based on deep learning as claimed in claim 1, characterized in that: in step S4, the accuracy and the recall ratio index are used as the accuracy index of the test data, and the deep learning model with the best recall ratio index is selected as the model for the final application.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210374768.XA CN114742795A (en) | 2022-04-11 | 2022-04-11 | Tunnel lining disease intelligent detection method based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210374768.XA CN114742795A (en) | 2022-04-11 | 2022-04-11 | Tunnel lining disease intelligent detection method based on deep learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114742795A true CN114742795A (en) | 2022-07-12 |
Family
ID=82281437
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210374768.XA Withdrawn CN114742795A (en) | 2022-04-11 | 2022-04-11 | Tunnel lining disease intelligent detection method based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114742795A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115638816A (en) * | 2022-09-13 | 2023-01-24 | 郑州大轩电子科技有限公司 | Wisdom agricultural information monitoring system |
-
2022
- 2022-04-11 CN CN202210374768.XA patent/CN114742795A/en not_active Withdrawn
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115638816A (en) * | 2022-09-13 | 2023-01-24 | 郑州大轩电子科技有限公司 | Wisdom agricultural information monitoring system |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108537215B (en) | Flame detection method based on image target detection | |
| CN108364280B (en) | Method and equipment for automatically describing structural crack and accurately measuring width | |
| CN108710919A (en) | A kind of crack automation delineation method based on multi-scale feature fusion deep learning | |
| CN113392775B (en) | Sugarcane seedling automatic identification and counting method based on deep neural network | |
| CN107665492B (en) | Colorectal panoramic digital pathological image tissue segmentation method based on depth network | |
| CN108711148B (en) | Tire defect intelligent detection method based on deep learning | |
| CN111289512B (en) | Rice grain alkali elimination value high-throughput determination method based on deep convolutional neural network | |
| CN111797920B (en) | Remote sensing extraction method and system for depth network impervious surface with gate control feature fusion | |
| den Bieman et al. | Deep learning video analysis as measurement technique in physical models | |
| CN114548278A (en) | In-service tunnel lining structure defect identification method and system based on deep learning | |
| CN108898096B (en) | High-resolution image-oriented information rapid and accurate extraction method | |
| CN111161224A (en) | Casting internal defect grading evaluation system and method based on deep learning | |
| CN113610778A (en) | Bridge surface crack detection method and system based on semantic segmentation | |
| CN113420619A (en) | Remote sensing image building extraction method | |
| CN111507972A (en) | Tunnel surface defect detection method combining convolutional neural network and support vector machine | |
| CN115223063A (en) | Unmanned aerial vehicle remote sensing wheat new variety lodging area extraction method and system based on deep learning | |
| CN114742795A (en) | Tunnel lining disease intelligent detection method based on deep learning | |
| CN115512247A (en) | Regional building damage grade assessment method based on image multi-parameter extraction | |
| CN115272826A (en) | Image identification method, device and system based on convolutional neural network | |
| CN114332534B (en) | Hyperspectral image small sample classification method | |
| CN111815625A (en) | Early lung cancer prediction method based on deep learning | |
| CN115830302B (en) | Multi-scale feature extraction fusion power distribution network equipment positioning identification method | |
| CN109978832B (en) | Twisted pair lay detection method based on edge reconstruction | |
| CN112084941A (en) | Target detection and identification method based on remote sensing image | |
| CN115690597A (en) | Remote sensing image urban ground feature change detection method based on depth background difference |
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 | ||
| WW01 | Invention patent application withdrawn after publication |
Application publication date: 20220712 |
|
| WW01 | Invention patent application withdrawn after publication |