A Method for Identifying Electromagnetic Radiation Signals of Active Cracks in Concrete Based on GoogLeNet Neural Network

Abstract

This invention discloses a method for identifying electromagnetic radiation signals from active concrete cracks based on a GoogLeNet neural network, comprising the following steps: acquiring a dataset of electromagnetic radiation signals from active concrete cracks and environmental noise; selecting initial samples from the electromagnetic radiation signals from active concrete cracks and environmental noise; increasing the sample size by linearly superimposing the initial samples; constructing a time-frequency domain image by applying continuous wavelet transform to the expanded samples; constructing a GoogLeNet convolutional neural network and setting initial parameters; training the GoogLeNet convolutional neural network and comprehensively judging the model solution results by determining whether the loss rate and accuracy meet the expected requirements and whether the number of iterations reaches the maximum requirement; saving the relevant network parameters to obtain a convolutional neural network model for identifying electromagnetic radiation signals from active concrete cracks based on GoogLeNet; identifying the data to be identified through the GoogLeNet convolutional neural network and outputting the identification results.

Description

Technical Field

[0001] This invention belongs to the field of signal recognition technology, specifically relating to a method for recognizing electromagnetic radiation signals from active cracks in concrete based on the GoogLeNet neural network. Background Technology

[0002] Concrete is a widely used material in engineering construction such as bridges, buildings, and dams. However, with changes in stress, deformation, and temperature of structural components, active cracks that cannot maintain stability can easily develop within concrete structures, posing potential risks and challenges to their structural stability. Once active cracks appear in concrete, their type and cause need to be analyzed, and their impact on structural safety needs to be assessed. Currently, methods such as ultrasonic monitoring, UAV image monitoring, and infrared monitoring are widely used in the identification of active cracks in concrete and the evaluation of structural performance. However, these monitoring methods are difficult to apply to the monitoring of active cracks within complex structures.

[0003] The generation of active cracks in concrete releases electromagnetic radiation signals, and identifying these signals can enable the detection of active cracks. However, the amplitude of the electromagnetic radiation signals generated by concrete cracking is small and easily interfered with by environmental noise. Therefore, it is necessary to study weak signal processing and identification methods to improve the success rate of detection. Commonly used signal processing methods, such as threshold denoising and frequency domain denoising, can perform preliminary decomposition of the signal, but they cannot reflect the frequency variation characteristics of the signal over time. Summary of the Invention

[0004] The purpose of this invention is to provide a method for identifying electromagnetic radiation signals of active cracks in concrete based on the GoogLeNet neural network, in order to solve the problems in the background art.

[0005] To achieve the above objectives, the present invention adopts the following technical solution:

[0006] A method for identifying electromagnetic radiation signals from active cracks in concrete based on the GoogLeNet neural network includes the following steps:

[0007] S1: Obtain the raw dataset of electromagnetic radiation signals and environmental noise from active concrete cracks;

[0008] S2: Select initial samples from the original dataset of electromagnetic radiation signals from active concrete cracks and environmental noise;

[0009] S3: Increase the sample size by linearly superimposing the initial samples;

[0010] S4: Apply continuous wavelet transform to the expanded samples to construct a time-frequency domain image;

[0011] S5: Construct the GoogLeNet convolutional neural network and set the initial parameters;

[0012] S6: Using the time-frequency domain images constructed by continuous wavelet transform on the expanded samples as the training sample set, train the GoogLeNet convolutional neural network. Judge the model solution results by judging whether the loss rate and accuracy meet the expected requirements and whether the number of iterations reaches the maximum requirement.

[0013] S7: Save the network-related parameters to obtain a convolutional neural network model based on GoogLeNet for recognizing electromagnetic radiation signals of multiple cracks in concrete.

[0014] S8: The data to be identified is identified through the GoogLeNet neural network, and the identification result is output.

[0015] In some embodiments, in S1, acquiring electromagnetic radiation signals and environmental noise from active concrete cracks includes: acquiring and recording electromagnetic radiation signals and environmental noise in real time using electromagnetic signal measuring instruments deployed on the concrete structure.

[0016] In some embodiments, in step S2, the initial sample is selected from three types of electromagnetic signals or noise from the field test of concrete cracking, specifically: electromagnetic radiation signal of active crack in concrete, electromagnetic radiation sudden noise, and electromagnetic radiation steady-state noise. Equal amounts of data are selected from each of these three types as three sets of data for the initial sample to form the initial sample.

[0017] In some embodiments, the electromagnetic radiation signal of active concrete cracks refers to the electromagnetic radiation signal of concrete cracking collected under conditions where there is no external noise interference; electromagnetic radiation sudden noise refers to electromagnetic noise whose amplitude changes suddenly over time in space, and electromagnetic radiation steady-state noise refers to electromagnetic noise whose amplitude does not change suddenly over time in space.

[0018] In some embodiments, the linear superposition processing step for the initial samples in S3 includes: randomly selecting 60% of each of the three sets of data in the initial samples; pairing the selected data in the same set into pairs, with the two data in a pair being denoted as data A and data B respectively; multiplying the value of each point of data A by k times to obtain data A′, and multiplying the value of each point of data B by (1-k) times to obtain data B′, where k is a random number between 0 and 1; adding data A′ and data B′ together to obtain data C, which serves as a new sample, thereby increasing the sample size by 30%.

[0019] In some embodiments, in step S4, the extended samples are subjected to continuous wavelet transform to construct a time-frequency domain image, and the specific method is as follows:

[0020] For a sample x(t), the following formula is used to perform a continuous wavelet transform, converting the time-domain signal into a two-dimensional time-frequency domain graph:

[0021]

[0022] Where CWT(a,b) is the result of continuous wavelet transform, and x(t) is the original one-dimensional time-domain signal. ψ(·) is the wavelet function, a is the scaling factor, b is the translation factor, and ψ * (·) is the complex conjugate of the wavelet function ψ(·).

[0023] In some embodiments, in step S5, the GoogleNet convolutional neural network is constructed, and the specific method is as follows:

[0024] Based on GoogLeNet, two auxiliary classifiers were removed; the output size of the fully connected layer that performs the classification task in the original network was changed to 3, and replaced with a new classification layer; at the same time, in order to further improve the learning ability of the network and reduce overfitting, an activation function was added after each convolutional layer.

[0025] In some embodiments, the initial parameters in S5 include: (1) Batch size, which refers to the number of samples selected in one training session; (2) Initial learning rate, which refers to the initial learning capacity of the neural network; (3) Iteration period, which refers to the total number of times all training datasets will be traversed; and (4) Maximum number of iterations, which refers to the maximum number of iterations of the internal parameters of the convolutional neural network structure.

[0026] In step S6, the comprehensive judgment of the model solution result by determining whether the loss rate and accuracy meet the expected requirements and whether the number of iterations meets the maximum requirement is as follows: if the loss rate is less than the loss rate threshold and the accuracy is greater than the accuracy threshold, then proceed to step S7; if the number of iterations is greater than or equal to the maximum number of iterations, then proceed to step S7; if neither of the above two conditions is met, then return to step S6 to train the GoogLeNet convolutional neural network.

[0027] In some embodiments, S7: Save network-related parameters including all data of convolutional layers, max pooling layers, average pooling layers, activation layers, and Inception kernels.

[0028] The beneficial effects of this invention compared to the prior art are:

[0029] This invention employs continuous wavelet transform to process the signal, converting a one-dimensional time-domain signal into a two-dimensional time-frequency domain graph. Signal recognition methods such as support vector machines, long short-term memory, deep learning-based clustering and segmentation, and machine vision can identify the processed signal. Among these methods, convolutional neural networks demonstrate excellent performance in image processing and classification.

[0030] This method uses continuous wavelet transform to decompose the electromagnetic radiation signal, converting it into a two-dimensional time-frequency domain graph. Combined with an improved GoogLeNet model, a trained recognition model is established to achieve rapid and accurate identification of electromagnetic radiation signals from active cracks in concrete. Attached Figure Description

[0031] Figure 1 This is a flowchart of a method for identifying electromagnetic radiation signals of active cracks in concrete based on a GoogLeNet neural network, according to an embodiment of the present invention.

[0032] Figure 2 These are the time-domain characteristics of electromagnetic radiation signals from active concrete cracks, electromagnetic radiation abrupt noise, and electromagnetic radiation steady-state environmental noise in the initial samples of this invention embodiment;

[0033] Figure 3 These are the frequency domain characteristics of electromagnetic radiation signals from active concrete cracks, electromagnetic radiation abrupt noise, and electromagnetic radiation steady-state environmental noise in the initial samples of this invention.

[0034] Figure 4 This is a time-frequency domain image constructed from samples through continuous wavelet transform in the embodiments of the present invention;

[0035] Figure 5 This is a diagram showing the recognition effect of the electromagnetic radiation signal recognition method for active concrete cracks based on the GoogLeNet neural network in an embodiment of the present invention.

[0036] Figure 6 This is a diagram of the neural network model architecture based on GoogLeNet in an embodiment of the present invention. Detailed Implementation

[0037] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions in the embodiments of this application will be described in more detail below with reference to the accompanying drawings. In the drawings, the same or similar reference numerals denote the same or similar components or components having the same or similar functions throughout. The described embodiments are some, but not all, of the embodiments of this application. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this application, and should not be construed as limiting this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0038] The embodiments of this application will now be described in detail with reference to the accompanying drawings.

[0039] Example 1:

[0040] like Figure 1-6 As shown, a method for identifying electromagnetic radiation signals from active cracks in concrete based on the GoogLeNet neural network includes the following steps:

[0041] S1: Obtain the raw dataset of electromagnetic radiation signals and environmental noise from active concrete cracks;

[0042] S2: Select initial samples from the original dataset of electromagnetic radiation signals from active concrete cracks and environmental noise;

[0043] S3: Increase the sample size by linearly superimposing the initial samples;

[0044] S4: Apply continuous wavelet transform to the expanded samples to construct a time-frequency domain image;

[0045] S5: Construct the GoogLeNet convolutional neural network and set the initial parameters;

[0046] S6: Using the time-frequency domain images constructed by continuous wavelet transform on the expanded samples as the training sample set, train the GoogLeNet convolutional neural network. Judge the model solution results by judging whether the loss rate and accuracy meet the expected requirements and whether the number of iterations reaches the maximum requirement.

[0047] In step S6, the comprehensive judgment of the model solution result by determining whether the loss rate and accuracy meet the expected requirements and whether the number of iterations meets the maximum requirement is as follows: if the loss rate is less than the loss rate threshold and the accuracy is greater than the accuracy threshold, then proceed to step S7; if the number of iterations is greater than or equal to the maximum number of iterations, then proceed to step S7; if neither of the above two conditions is met, then return to step S6 to train the GoogLeNet convolutional neural network.

[0048] S7: Save the network-related parameters to obtain a convolutional neural network model for recognizing electromagnetic radiation signals of multiple cracks in concrete based on GoogLeNet; S7: saving the network-related parameters includes all data of convolutional layers, max pooling layers, average pooling layers, activation layers, and Inception kernels.

[0049] S8: The data to be identified is identified through the GoogLeNet neural network, and the identification result is output.

[0050] In some embodiments of the present invention, acquiring electromagnetic radiation signals and environmental noise from active concrete cracks includes: acquiring and recording electromagnetic radiation signals and environmental noise in real time using electromagnetic signal measuring instruments deployed on concrete structures.

[0051] In some embodiments of the present invention, the initial sample is selected as follows: the original dataset is divided into three categories: electromagnetic radiation signal of active crack in concrete, electromagnetic radiation sudden noise, and electromagnetic radiation steady-state noise. Equal amounts of data are selected from each category as three sets of data for the initial sample to form the initial sample.

[0052] In some embodiments of the present invention, the processing steps for the linear superposition of the initial samples include: randomly selecting 60% of each of the three sets of data in the initial samples; pairing the selected data in the same set into pairs, with the two data in a pair being denoted as data A and data B respectively; changing the value of each point of data A to k times the original value to obtain data A′, and changing the value of each point of data B to (1-k) times the original value to obtain data B′, where k is a random number between 0 and 1; adding data A′ and data B′ together to obtain data C, which serves as a new sample, thereby increasing the sample size by 30%.

[0053] In some embodiments of the present invention, a continuous wavelet transform is used on the expanded sample to construct a time-frequency domain image. The specific method is as follows: For sample x(t), a continuous wavelet transform is performed using the following formula to convert the time-domain signal into a two-dimensional time-frequency domain image:

[0054]

[0055] Where CWT(a,b) is the result of continuous wavelet transform, and x(t) is the original one-dimensional time-domain signal. ψ(·) is the wavelet function, a is the scaling factor, b is the translation factor, and ψ * (·) is the complex conjugate of the wavelet function ψ(·).

[0056] In some embodiments of the present invention, the initial parameters include: (1) batch size; (2) initial learning rate; (3) iteration period; and (4) maximum number of iterations.

[0057] In some embodiments of the present invention, the original dataset is divided into three categories: electromagnetic radiation signals from active concrete cracks, electromagnetic radiation abrupt noise, and electromagnetic radiation steady-state environmental noise; 1000 data points are selected from each category to form an initial sample, and the time-domain and frequency-domain features of the initial sample are as follows: Figure 2 , Figure 3 As shown; the sample size is increased by linear superposition of the initial samples, raising the sample size of each group to 1300; continuous wavelet transform is applied to the expanded samples to construct a time-frequency domain image, as shown in the figure. Figure 4As shown; construct a GoogLeNet convolutional neural network, setting the initial parameters as follows: batch size set to 15, initial learning rate set to 0.0001, iteration period set to 5, and maximum number of iterations set to 72; divide each group of time-frequency domain images into training and validation sets in a 10:3 ratio, and train the constructed GoogLeNet convolutional neural network; save the network-related parameters, and use the GoogLeNet neural network to recognize the data to be identified. The recognition effect is shown in the figure. Figure 5 As shown, it can be seen that the method provided by the present invention has a good recognition effect.

[0058] This invention discloses a method for identifying electromagnetic radiation signals from active concrete cracks based on a GoogLeNet neural network. The method involves: acquiring electromagnetic radiation signals from active concrete cracks and environmental noise; selecting initial samples from these samples; increasing the sample size through linear superposition of the initial samples; constructing a time-frequency domain image using continuous wavelet transform on the expanded samples; constructing a GoogLeNet convolutional neural network and setting initial parameters; training the GoogLeNet convolutional neural network and comprehensively judging the model's solution results by determining whether the loss rate and accuracy meet expected requirements and whether the number of iterations reaches the maximum requirement; saving the network's relevant parameters to obtain a convolutional neural network model for identifying electromagnetic radiation signals from active concrete cracks based on GoogLeNet; identifying the data to be identified using the GoogLeNet convolutional neural network and outputting the identification results. This invention achieves rapid and accurate identification of electromagnetic radiation signals from active concrete cracks by introducing the GoogLeNet convolutional neural network into the identification of such signals.

[0059] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for identifying electromagnetic radiation signals from active cracks in concrete based on a GoogLeNet neural network, characterized in that, Including: the following step: S1: Obtain the raw dataset of electromagnetic radiation signals and environmental noise from active concrete cracks; S2: Select initial samples from the original dataset of electromagnetic radiation signals from active concrete cracks and environmental noise; S3: Increase the sample size by linearly superimposing the initial samples; S4: Apply continuous wavelet transform to the expanded samples to construct a time-frequency domain image; S5: Construct the GoogLeNet convolutional neural network and set the initial parameters; S6: Using the time-frequency domain images constructed by continuous wavelet transform on the expanded samples as the training sample set, train the GoogLeNet convolutional neural network. Judge the model solution results by judging whether the loss rate and accuracy meet the expected requirements and whether the number of iterations reaches the maximum requirement. S7: Save the network-related parameters to obtain a convolutional neural network model based on GoogLeNet for recognizing electromagnetic radiation signals of multiple cracks in concrete. S8: The data to be identified is identified through the GoogLeNet neural network, and the identification result is output. In step S3, the linear superposition processing step for the initial samples includes: randomly selecting 60% from each of the three sets of data in the initial samples; pairing the selected data within the same set into pairs, with the two data points in each pair denoted as data A and data B; and multiplying the value of each point in data A by k times to obtain data B. The value of each point in data B is multiplied by (1-k) times to obtain the data. Where k is a random number between 0 and 1; [The data is then processed / conducted]. and data The data are added together to obtain data C, which serves as a new sample, thereby increasing the sample size by 30%. In step S4, the extended samples are subjected to continuous wavelet transform to construct a time-frequency domain image. The specific method is as follows: For the sample The following formula is used to perform continuous wavelet transform, converting the time-domain signal into a two-dimensional time-frequency domain graph: ； Where CWT(a, b) is the result of continuous wavelet transform, and x(t) is the original one-dimensional time domain signal; These are wavelet functions, where a is the scaling factor and b is the translation factor. It is a wavelet function The complex conjugate; In step S5, the GoogleNet convolutional neural network is constructed using the following method: Based on GoogLeNet, two auxiliary classifiers were removed; the output size of the fully connected layer that performs the classification task in the original network was changed to 3, and replaced with a new classification layer; an activation function was added after each convolutional layer.

2. The method for identifying electromagnetic radiation signals of active cracks in concrete based on the GoogLeNet neural network according to claim 1, characterized in that, In step S1, acquiring electromagnetic radiation signals and environmental noise from active concrete cracks includes: real-time acquisition and recording of electromagnetic radiation signals and environmental noise using electromagnetic signal measuring instruments deployed on the concrete structure.

3. The method for identifying electromagnetic radiation signals of active cracks in concrete based on the GoogLeNet neural network according to claim 1, characterized in that, In step S2, the initial sample is selected from three types of electromagnetic signals or noise from the field test of concrete cracking, specifically: electromagnetic radiation signal of active crack in concrete, electromagnetic radiation sudden noise, and electromagnetic radiation steady-state noise. Equal amounts of data are selected from each of these three types as three sets of data for the initial sample.

4. The method for identifying electromagnetic radiation signals of active cracks in concrete based on the GoogLeNet neural network according to claim 3, characterized in that, The electromagnetic radiation signal of active cracks in concrete refers to the electromagnetic radiation signal collected when concrete cracks, under conditions where there is no external noise interference; electromagnetic radiation sudden noise refers to electromagnetic noise in space whose amplitude changes suddenly over time, while electromagnetic radiation steady-state noise refers to electromagnetic noise in space whose amplitude does not change suddenly over time.

5. The method for identifying electromagnetic radiation signals of active cracks in concrete based on the GoogLeNet neural network according to claim 1, characterized in that, In S5, the initial parameters include: (1) Batch size, which refers to the number of samples selected in one training session; (2) Initial learning rate, which refers to the initial learning capacity of the neural network; (3) Iteration period, which refers to the total number of times all training datasets will be traversed; (4) Maximum number of iterations, which refers to the maximum number of iterations of the internal parameters of the convolutional neural network structure; (5) Loss rate threshold; (6) Accuracy threshold.

6. The method for identifying electromagnetic radiation signals of active cracks in concrete based on the GoogLeNet neural network according to claim 1, characterized in that, In step S6, the comprehensive judgment of the model solution result by determining whether the loss rate and accuracy meet the expected requirements and whether the number of iterations meets the maximum requirement is as follows: if the loss rate is less than the loss rate threshold and the accuracy is greater than the accuracy threshold, then proceed to step S7; if the number of iterations is greater than or equal to the maximum number of iterations, then proceed to step S7; if neither of the above two conditions is met, then return to step S6 to train the GoogLeNet convolutional neural network.

7. The method for identifying electromagnetic radiation signals of active cracks in concrete based on the GoogLeNet neural network according to claim 1, characterized in that, S7: Save all data related to network parameters, including convolutional layers, max pooling layers, average pooling layers, activation layers, and Inception kernels.

Images