CN111754463B - Method for detecting CA mortar layer defects of ballastless track based on convolutional neural network - Google Patents

Method for detecting CA mortar layer defects of ballastless track based on convolutional neural network Download PDF

Info

Publication number
CN111754463B
CN111754463B CN202010489396.6A CN202010489396A CN111754463B CN 111754463 B CN111754463 B CN 111754463B CN 202010489396 A CN202010489396 A CN 202010489396A CN 111754463 B CN111754463 B CN 111754463B
Authority
CN
China
Prior art keywords
defect
neural network
convolutional neural
layer
data
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.)
Active
Application number
CN202010489396.6A
Other languages
Chinese (zh)
Other versions
CN111754463A (en
Inventor
赵维刚
陈甜甜
李荣喆
杨勇
田秀淑
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Shijiazhuang Tiedao University
Original Assignee
Shijiazhuang Tiedao University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Shijiazhuang Tiedao University filed Critical Shijiazhuang Tiedao University
Priority to CN202010489396.6A priority Critical patent/CN111754463B/en
Publication of CN111754463A publication Critical patent/CN111754463A/en
Application granted granted Critical
Publication of CN111754463B publication Critical patent/CN111754463B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

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
    • G06T7/0004—Industrial image inspection
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
    • G01N21/84—Systems specially adapted for particular applications
    • G01N21/88—Investigating the presence of flaws or contamination
    • G01N21/8851—Scan or image signal processing specially adapted therefor, e.g. for scan signal adjustment, for detecting different kinds of defects, for compensating for structures, markings, edges
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
    • 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
    • G06N3/045—Combinations of networks
    • 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
    • G06N3/047—Probabilistic or stochastic networks
    • 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
    • G06N3/048—Activation functions
    • 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
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
    • G01N21/84—Systems specially adapted for particular applications
    • G01N21/88—Investigating the presence of flaws or contamination
    • G01N21/8851—Scan or image signal processing specially adapted therefor, e.g. for scan signal adjustment, for detecting different kinds of defects, for compensating for structures, markings, edges
    • G01N2021/8854—Grading and classifying of flaws
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
    • G01N21/84—Systems specially adapted for particular applications
    • G01N21/88—Investigating the presence of flaws or contamination
    • G01N21/8851—Scan or image signal processing specially adapted therefor, e.g. for scan signal adjustment, for detecting different kinds of defects, for compensating for structures, markings, edges
    • G01N2021/8887—Scan or image signal processing specially adapted therefor, e.g. for scan signal adjustment, for detecting different kinds of defects, for compensating for structures, markings, edges based on image processing techniques
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/10—Image acquisition modality
    • G06T2207/10004—Still image; Photographic image
    • G06T2207/10012—Stereo images
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20081—Training; Learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20084—Artificial neural networks [ANN]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/30—Subject of image; Context of image processing
    • G06T2207/30108—Industrial image inspection

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • General Physics & Mathematics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • General Engineering & Computer Science (AREA)
  • Artificial Intelligence (AREA)
  • Evolutionary Computation (AREA)
  • General Health & Medical Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Mathematical Physics (AREA)
  • Computing Systems (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Software Systems (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Molecular Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Evolutionary Biology (AREA)
  • Probability & Statistics with Applications (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Quality & Reliability (AREA)
  • Signal Processing (AREA)
  • Chemical & Material Sciences (AREA)
  • Analytical Chemistry (AREA)
  • Biochemistry (AREA)
  • Immunology (AREA)
  • Pathology (AREA)
  • Image Analysis (AREA)

Abstract

The invention discloses a method for detecting a CA mortar layer defect of a ballastless track based on a convolutional neural network, which comprises the following steps: s1: converting the defect information into signal information; s2: converting the signal information into three-dimensional image information as a characteristic image; s3: establishing a data set; s4: training a convolutional neural network; s5: obtaining defect data to be detected; s6: preprocessing defect data to be detected; s7: and calculating a detection result. The invention solves the problems of inaccurate feature extraction, inaccurate defect size classification, low detection accuracy and the like in the prior art. The method is suitable for the technical field of defect detection of the CA mortar layer of the ballastless track.

Description

Method for detecting CA mortar layer defects of ballastless track based on convolutional neural network
Technical Field
The invention belongs to the technical field of ballastless track CA mortar layer defect detection, and particularly relates to a ballastless track CA mortar layer defect detection method based on a convolutional neural network.
Background
The Cement emulsified asphalt mortar (Cement-emulsified Asphalt Mortar, CA mortar)) layer is an elastic adjusting layer of the CRTSI and CRTSII plate type ballastless track, and the state of the elastic adjusting layer directly influences the durability of the ballastless track and the comfort and safety of train running. The damage of the CA mortar layer has hidden characteristics and cannot be directly observed. Therefore, research on nondestructive testing technology aiming at hidden damage of the CA mortar layer is important for early discovery and early repair of damage.
The impact echo method (IE) is the most commonly used method for detecting cracks and layering of a concrete structure, is convenient to operate, can detect on one side, has large detection depth, and is free from the influence of concrete materials and structural differences, so that the method is widely applied to detecting defects of a CA mortar layer. The main problems faced by the CA mortar layer void identification method are as follows: (1) The signal characteristic extraction is based on time-frequency domain characteristics, so that the signal expression is not comprehensive enough, and the characteristic parameter selection has difficulty. (2) Pattern recognition is based on traditional probability statistical methods and supervised machine learning methods, and human subjective participation is required, so that detection accuracy and speed are affected.
Disclosure of Invention
The invention provides a method for detecting defects of a CA mortar layer of a ballastless track based on a convolutional neural network, which aims to solve the problems that the feature extraction is not accurate enough, the defect size cannot be accurately classified, the detection accuracy is low and the like in the prior art.
In order to achieve the above purpose, the technical scheme adopted by the invention is as follows:
a ballastless track CA mortar layer defect detection method based on a convolutional neural network comprises the following steps:
S1: converting the defect information into signal information, and testing the defect model by using a scanning type impact echo test system to obtain signals of different defect types;
S2: converting the signal information into three-dimensional image information as a characteristic image, preprocessing data by using Hilbert yellow conversion to obtain a Hilbert spectrogram of the signal, and preprocessing the Hilbert spectrogram to be used as the characteristic image of the signal;
s3: establishing a data set, labeling characteristic images of echo signals, and establishing a training and testing data set;
s4: training a convolutional neural network according to the established data set and defect judgment requirements and a training algorithm;
s5: obtaining defect data to be detected, and detecting a test piece to be detected by using a scanning type impact echo test system to obtain the defect data to be detected;
S6: preprocessing the defect data to be detected, and preprocessing the data to be detected according to a preprocessing algorithm to obtain preprocessed data to be detected;
s7: and calculating a detection result, and inputting the preprocessed defect data to be detected into a trained convolutional neural network for calculation to obtain the detection result.
Further, when the thickness of the standard track plate is 0.2m, the frequency range of the impactor in the scanning type impact echo test system is 3 kHz-24 kHz, the sampling point is 2048, the sampling frequency of each point is 10 mu s, and the gain value is manually adjusted according to the actual test waveform condition.
Further, in steps S2 and S6, the preprocessing algorithm specifically includes: the method comprises the steps of introducing Hilbert yellow transformation to perform feature extraction on a measured impact echo signal, converting the signal into three-dimensional image data, imaging the obtained three-dimensional data, and performing normalization processing to obtain a feature image of the echo signal.
Further, the processing steps of the characteristic image are as follows:
I, hilbert spectrum analysis is carried out on the echo signals to obtain three-dimensional distribution conditions of time-frequency-amplitude values of the signals, and defect conditions are reflected;
intercepting a frequency spectrum image with the frequency of 0-20 KHz and the time of 0-10 mu s, reducing background interference, improving recognition accuracy and enabling signal characteristic expression to be more accurate;
and III, converting the cut RGB image into a gray image of 32×32 pixels, and taking the gray image as an input of a convolutional neural network.
Further, in step S4, the training algorithm selects a ReLU function as an activation function, a loss function adopts a cross entropy function, an output layer selects a softmax function, and model classification is completed, and in step S4, the convolutional neural network sequentially includes a first convolutional layer, a first pooling layer, a second convolutional layer, a second pooling layer, and a first full connection layer.
Further, the first convolution layer uses 6 convolution kernels of 5×5, the step size is 1, the image of 32×32 pixels is convolved, the size of the feature map after convolution is 28×28 pixels, the data after convolution operation is subjected to ReLU transformation, and then pooling is performed, and the size of the feature map is halved to be 14×14 pixels.
Further, the second convolution layer adopts 12 convolution kernels of 5×5 to convolve the pooled 14×14pixel feature maps to obtain 12 feature maps of 10×10 pixels, performs ReLU transformation on the convolved data, and then pools the data, and the feature map size is halved to be 5×5 pixels.
Further, the first full connection layer has a neuron number of 84.
Further, the output layer performs multi-category classification by using a softmax classifier, and outputs the category number according to the required classification.
Compared with the prior art, the invention adopts the structure, and the technical progress is that: the invention provides a method for extracting three-dimensional image features of echo signals based on HHT, which utilizes HHT to extract time-frequency-energy three-dimensional image features of original echo signals, expresses one-dimensional information in a three-dimensional space, increases absolute distance between the features and provides a basis for defect identification; and a CA mortar layer defect recognition method based on the convolutional neural network is provided, the convolutional neural network is constructed, classification recognition of the defect existence and defect size of the CA mortar layer is realized, and defect recognition accuracy is improved.
Drawings
The accompanying drawings are included to provide a further understanding of the invention and are incorporated in and constitute a part of this specification, illustrate the invention and together with the embodiments of the invention, serve to explain the invention.
In the drawings:
FIG. 1 is a flow chart of the present invention;
FIG. 2 is a schematic diagram of the principle of the shock echo of the present invention;
FIG. 3 is a schematic diagram of a feature image processing flow of the present invention;
FIG. 4 is a schematic diagram of a convolutional neural network according to the present invention;
FIG. 5 is a flow chart of an example of the present invention;
FIG. 6 is a layout of defects of a rail plate under four working conditions according to an example of the present invention;
fig. 7 is a diagram of recognition results of an example in the present invention.
Detailed Description
Preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are presented for purposes of illustration and explanation only and are not intended to limit the present invention.
The invention discloses a method for detecting a CA mortar layer defect of a ballastless track based on a convolutional neural network, which is shown in figures 1-7 and comprises the following steps:
Step one, a scanning type impact echo test system (IES) system is used for testing the track slab entity model to obtain 1200 defect-free signals, 400 defect signals of 0.12 multiplied by 0.2m, 400 defect signals of 0.2 multiplied by 0.3m and 400 defect signals of 0.12 multiplied by 0.4m, and training test is carried out as original data.
And step two, performing Hilbert yellow transformation on the obtained signals to obtain a Hilbert spectrogram. And cutting and graying the Hilbert spectrogram to obtain a characteristic picture of 32×32 pixels, wherein the characteristic image processing flow is shown in FIG. 3.
Step three, a normal signal data set X1 (1200 defect signals) and a defect signal data set X2 (1200 defect signals) are established, and defect characteristic signal data sets X3 (400 defect signals of 0.12x0.2mdefect), X4 (400 defect signals of 0.2x0.3mdefect) and X5 (400 defect signals of 0.12x0.4mdefect) are established.
And step four, training a convolutional neural network.
Training a DE-CNN (Defect Exist-Convolutional Neural Network) model by using a data set X1 and a data set X2 to judge the void position, wherein the output types are two types, namely defective and non-defective; and selecting a ReLU function of the DE-CNN model as an activation function. The loss function adopts a cross entropy function, and the output layer selects a softmax function to finish model classification. In the fourth step, the convolutional neural network sequentially comprises a first convolutional layer, a first pooling layer, a second convolutional layer, a second pooling layer and a first full-connection layer. Wherein:
The first convolution layer uses 65 x 5 convolution kernels with a step size of 1. The image of 32×32 pixels is convolved, and the size of the feature map after convolution is 28×28 pixels. And carrying out ReLU transformation on the data subjected to convolution operation, and then carrying out pooling, wherein the size of the characteristic diagram is halved to be 14 multiplied by 14 pixels.
The second convolution layer uses 12 convolution kernels of 5 x 5. Convolving the pooled 14×14pixel feature maps resulted in 12 feature maps of 10×10 pixels. And carrying out ReLU transformation on the data subjected to convolution operation, and then carrying out pooling, wherein the size of the characteristic diagram is halved to be 5 multiplied by 5 pixels.
The first full connection layer has a neuron number of 84.
Output layer: with the softmax classifier, the output is classified into two classes, defective and non-defective.
The DS-CNN (Defect Size-Convolutional Neural Network) model is trained by using the data sets X3, X4 and X5 to judge the Size of the void, and the output types are three, namely 0.12m multiplied by 0.2m void, 0.2m multiplied by 0.3m void and 0.3m multiplied by 0.4m void. And selecting a ReLU function of the DE-CNN model as an activation function. The loss function adopts a cross entropy function, and the output layer selects a softmax function to finish model classification. In the fourth step, the convolutional neural network sequentially comprises a first convolutional layer, a first pooling layer, a second convolutional layer, a second pooling layer and a first full-connection layer.
The first convolution layer uses 65 x 5 convolution kernels with a step size of 1. The image of 32×32 pixels is convolved, and the size of the feature map after convolution is 28×28 pixels. And carrying out ReLU transformation on the data subjected to convolution operation, and then carrying out pooling, wherein the size of the characteristic diagram is halved to be 14 multiplied by 14 pixels.
The second convolution layer uses 12 convolution kernels of 5 x 5. Convolving the pooled 14×14pixel feature maps resulted in 12 feature maps of 10×10 pixels. And performing ReLU transformation on the data subjected to convolution operation, and then performing pooling, wherein the size of the characteristic diagram is halved to be 5×5 pixels.
The first full connection layer has a neuron number of 84.
Output layer: three types of output are respectively 0.12m×0.2m empty, 0.2m×0.3m empty and 0.3m×0.4m empty by using a softmax classifier.
And fifthly, obtaining data to be detected. And testing the first, second, third and fourth track plates to obtain four-plate echo signal data.
And step six, preprocessing echo signal data of the four plates to obtain a characteristic image serving as data to be detected.
And step seven, inputting the data of the four plates into the trained DE-CNN model and DS-CNN model for multiple times to detect defects, wherein the detection result is shown in figure 7, and the verification result shows that the average recognition rate of the CA mortar layer defect recognition model based on HHT-CNN can reach 98.75%.
In summary, the invention provides the echo signal three-dimensional image feature extraction method based on HHT, which utilizes HHT to extract the time-frequency-energy three-dimensional image feature of the original echo signal, expresses one-dimensional information in a three-dimensional space, increases the absolute distance between the features and provides a basis for defect identification; and a CA mortar layer defect recognition method based on the convolutional neural network is provided, the convolutional neural network is constructed, classification recognition of the defect existence and defect size of the CA mortar layer is realized, and defect recognition accuracy is improved.
Finally, it should be noted that: the foregoing description is only a preferred embodiment of the present invention, and the present invention is not limited thereto, but it is to be understood that modifications and equivalents of some of the technical features described in the foregoing embodiments may be made by those skilled in the art, although the present invention has been described in detail with reference to the foregoing embodiments. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention should be included in the scope of the claims of the present invention.

Claims (9)

1. A ballastless track CA mortar layer defect detection method based on a convolutional neural network is characterized by comprising the following steps:
s1: converting the defect information into signal information, and testing the track slab entity model by using a scanning type impact echo test system to obtain signals with different defect types;
S2: converting the signal information into three-dimensional image information serving as a characteristic image, wherein the specific steps are as follows: preprocessing the signal information by using Hilbert yellow transformation to obtain a Hilbert spectrogram of the signal, and preprocessing the Hilbert spectrogram to serve as a characteristic image of the signal;
S3: establishing a data set, labeling the characteristic images of the signal information, and establishing a training and testing data set;
s4: training a convolutional neural network according to the established data set and defect judgment requirements and a training algorithm;
s5: obtaining defect data to be detected, and detecting a test piece to be detected by using a scanning type impact echo test system to obtain the defect data to be detected;
S6: preprocessing the defect data to be detected, and preprocessing the data to be detected according to a preprocessing algorithm to obtain preprocessed data to be detected;
s7: and calculating a detection result, and inputting the preprocessed defect data to be detected into a trained convolutional neural network for calculation to obtain the detection result.
2. The method for detecting the defect of the CA mortar layer of the ballastless track based on the convolutional neural network, which is characterized by comprising the following steps of: when the thickness of the standard track plate is 0.2m, the frequency range of an impactor in the scanning type impact echo test system is 3 kHz-24 kHz, sampling points are 2048, sampling frequency of each point is 10 mu s, and the gain value is manually adjusted according to the actual test waveform condition.
3. The method for detecting the defect of the CA mortar layer of the ballastless track based on the convolutional neural network according to claim 1, wherein in the steps S2 and S6, the preprocessing algorithm specifically comprises: the Hilbert yellow transformation is introduced to perform feature extraction on the measured impact echo signals, the signals are converted into three-dimensional image data, the obtained three-dimensional data are imaged, and normalization processing is performed to obtain feature images of the signal information.
4. The method for detecting the defect of the CA mortar layer of the ballastless track based on the convolutional neural network, which is characterized by comprising the following steps of:
carrying out Hilbert spectrum analysis on the signal information to obtain the three-dimensional distribution condition of time-frequency-amplitude of the signal, and reflecting the defect condition;
intercepting a frequency spectrum image with the frequency of 0-20 KHz and the time of 0-10 mu s, reducing background interference, improving identification accuracy and enabling signal characteristic expression to be more accurate;
and III, converting the cut RGB image into a gray image of 32×32 pixels, and taking the gray image as an input of a convolutional neural network.
5. The method for detecting the defect of the CA mortar layer of the ballastless track based on the convolutional neural network, which is characterized by comprising the following steps of: in step S4, the training algorithm selects a ReLU function as an activation function, a loss function adopts a cross entropy function, an output layer selects a softmax function, and model classification is completed, and in step S4, the convolutional neural network sequentially includes a first convolutional layer, a first pooling layer, a second convolutional layer, a second pooling layer, and a first full-connection layer.
6. The method for detecting the defect of the CA mortar layer of the ballastless track based on the convolutional neural network, which is characterized by comprising the following steps of: the first convolution layer adopts 6 convolution kernels with the length of 1 being 5×5, convolves an image with the length of 32×32 pixels, the size of a characteristic diagram after convolution is 28×28 pixels, performs ReLU transformation on data after convolution operation, and then performs pooling, wherein the size of the characteristic diagram is halved to be 14×14 pixels.
7. The method for detecting the defect of the CA mortar layer of the ballastless track based on the convolutional neural network, which is characterized by comprising the following steps of: and a second convolution layer adopts 12 convolution kernels of 5 multiplied by 5 to convolve the pooled 14 multiplied by 14pixel characteristic graphs to obtain 12 characteristic graphs of 10 multiplied by 10 pixels, and performs ReLU transformation on the convolved data, and then the pooled characteristic graphs are halved to be 5 multiplied by 5 pixels.
8. The method for detecting the defect of the CA mortar layer of the ballastless track based on the convolutional neural network, which is characterized by comprising the following steps of: the first full connection layer has a neuron number of 84.
9. The method for detecting the defect of the CA mortar layer of the ballastless track based on the convolutional neural network, which is characterized by comprising the following steps of: and the output layer is used for classifying multiple categories by using a softmax classifier and outputting the category number according to the required classification.
CN202010489396.6A 2020-06-02 2020-06-02 Method for detecting CA mortar layer defects of ballastless track based on convolutional neural network Active CN111754463B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010489396.6A CN111754463B (en) 2020-06-02 2020-06-02 Method for detecting CA mortar layer defects of ballastless track based on convolutional neural network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010489396.6A CN111754463B (en) 2020-06-02 2020-06-02 Method for detecting CA mortar layer defects of ballastless track based on convolutional neural network

Publications (2)

Publication Number Publication Date
CN111754463A CN111754463A (en) 2020-10-09
CN111754463B true CN111754463B (en) 2024-05-14

Family

ID=72674513

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010489396.6A Active CN111754463B (en) 2020-06-02 2020-06-02 Method for detecting CA mortar layer defects of ballastless track based on convolutional neural network

Country Status (1)

Country Link
CN (1) CN111754463B (en)

Families Citing this family (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112732092B (en) * 2021-01-22 2023-04-07 河北工业大学 Surface electromyogram signal identification method based on double-view multi-scale convolution neural network
CN112862764A (en) * 2021-01-26 2021-05-28 中国铁道科学研究院集团有限公司 Method and device for identifying ballastless track bed gap damage and storage medium
CN113064166B (en) * 2021-03-22 2023-01-06 石家庄铁道大学 Method and device for detecting thickness of thin layer defect of multilayer concrete structure and terminal
CN113655002B (en) * 2021-08-30 2023-11-24 华侨大学 Recycled aggregate quality detection system of surface mortar based on hyperspectral technology
CN115856078B (en) * 2022-11-28 2023-11-07 西南交通大学 Self-feedback-regulation ballastless track mortar hidden damage intelligent detection system

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2018214195A1 (en) * 2017-05-25 2018-11-29 中国矿业大学 Remote sensing imaging bridge detection method based on convolutional neural network
CN110243934A (en) * 2019-05-30 2019-09-17 中国计量大学上虞高等研究院有限公司 A kind of ultrasonic weld seam detection method based on wavelet convolution neural network
CN110269625A (en) * 2019-05-31 2019-09-24 杭州电子科技大学 A kind of electrocardio authentication method and system of novel multiple features fusion

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2018214195A1 (en) * 2017-05-25 2018-11-29 中国矿业大学 Remote sensing imaging bridge detection method based on convolutional neural network
CN110243934A (en) * 2019-05-30 2019-09-17 中国计量大学上虞高等研究院有限公司 A kind of ultrasonic weld seam detection method based on wavelet convolution neural network
CN110269625A (en) * 2019-05-31 2019-09-24 杭州电子科技大学 A kind of electrocardio authentication method and system of novel multiple features fusion

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
基于卷积神经网络的轨道表面缺陷检测技术实现;刘孟轲;吴洋;王逊;;现代计算机(专业版)(第29期);全文 *

Also Published As

Publication number Publication date
CN111754463A (en) 2020-10-09

Similar Documents

Publication Publication Date Title
CN111754463B (en) Method for detecting CA mortar layer defects of ballastless track based on convolutional neural network
CN112462346B (en) Ground penetrating radar subgrade disease target detection method based on convolutional neural network
CN113362326B (en) Method and device for detecting defects of welding spots of battery
CN104964886A (en) Welded member fatigue stress and strain real-time non-contact type monitoring method
CN113009447B (en) Road underground cavity detection and early warning method based on deep learning and ground penetrating radar
CN102928435A (en) Aircraft skin damage identification method and device based on image and ultrasound information fusion
CN104599280A (en) Road crack intersection extraction method and system
CN115879343B (en) Ballastless track self-compacting concrete void two-stage identification method
CN114581764B (en) Underground structure crack disease discriminating method based on deep learning algorithm
CN108844856B (en) Sleeve grouting defect nondestructive testing method based on impact elastic wave and machine learning
CN111783616B (en) Nondestructive testing method based on data-driven self-learning
CN115420806A (en) Nondestructive ultrasonic detection method based on neural network and image fusion
CN114445397A (en) Strip steel defect detection method based on shallow neural network
CN115656335A (en) Method for quickly detecting and identifying internal defects of bearing ring
CN113536924A (en) Method for detecting internal damage type of steel fiber concrete structure
CN111325724B (en) Tunnel crack region detection method and device
CN115830302B (en) Multi-scale feature extraction fusion power distribution network equipment positioning identification method
CN116958645A (en) Scratch identification method for full-automatic integrated automobile paint repair
CN112801110B (en) Target detection method and device for image distortion correction of linear array camera of rail train
CN115791803A (en) Deep-buried tunnel surrounding rock blasting damage test system and test method
CN112508862B (en) Method for enhancing magneto-optical image of crack by improving GAN
CN110988140B (en) Intelligent identification method for fatigue cracks of orthotropic steel bridge deck slab
CN114881938A (en) Grain size detection method and system based on wavelet analysis and neural network
CN110910373B (en) Identification method of orthotropic steel bridge deck fatigue crack detection image
CN114708190A (en) Road crack detection and evaluation algorithm based on deep learning

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