A tunnel lining cavity detection method and system based on sound pulse signal
Through the detection method based on sound pulse signal, the tunnel lining hole is identified using the support vector machine model, which solves the problem of poor identification effect in the prior art, and achieves high-accuracy hole detection and structural damage-free effect.
Patent Information
- Application Number
- CN202211051407.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-08-31
- Publication Date
- 2024-11-19
- Estimated Expiration
- 2042-08-31
AI Technical Summary
The existing technology cannot effectively identify tunnel lining holes, and may cause damage to the structure, and cannot fully cover the policy supporting mechanism and coverage of the carbon emission trading market, and there is a problem of poor identification results.
Using a detection method based on sound pulse signal, the sound pulse signal is picked up by hitting the tunnel lining holes and solid parts, pre-processing and MFCC feature extraction, and a support vector machine model is built for training and recognition, and the low-dimensional feature vector conversion is transformed using kernel functions. is a high-dimensional space to achieve linear separability.
High accuracy identification of tunnel lining holes is achieved, avoiding damage to the structure, increasing detection performance, improving model reliability and information acquisition, and enhancing the practicality of detection.
Smart Images

Figure CN115420798B_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the technical field of tunnel lining cavity detection, and in particular to a tunnel lining cavity detection method and system based on sound pulse signals. Background Art
[0002] Gradually promoting the carbon emissions trading market is an important way to achieve the "dual carbon" goal. Since 2013, the pilot work of carbon emissions trading in various provinces has been launched. Although a lot of experience has been accumulated, many shortcomings have also been exposed, which directly restricts the establishment and implementation of a unified carbon market.
[0003] At present, the existing methods cannot make up for the factors of carbon emission trading market operation efficiency that are simply avoided from the macro level, cannot improve the policy supporting mechanism of carbon emission trading market, cannot expand the coverage of carbon emission trading market and optimize carbon emission trading market. Through the implementation of pilot carbon emission trading in various provinces, many problems have been exposed, such as total control and allocation of carbon quotas, incentive rules for trading entities, compliance system, carbon emission accounting and accounting system, and carbon financial innovation. Summary of the invention
[0004] In view of this, the present invention provides a method and system for detecting voids in tunnel lining based on sound pulse signals, at least to solve the problem of poor recognition effect in the prior art.
[0005] In a first aspect, the present invention provides a method for detecting a tunnel lining cavity based on an acoustic pulse signal, comprising:
[0006] S1: Hitting the hollow and solid parts of the tunnel lining to pick up the sound pulse signals generated by the hollow and solid parts of the tunnel lining respectively;
[0007] S2: converting the sound pulse signal into a digital signal and then preprocessing it, extracting the pulse signal with a set energy threshold to form a data set;
[0008] S3: After dividing the pulse samples of unequal lengths in the data set into frames of equal number, extracting the MFCC features of the two types of samples, namely, hollow and solid, in the data set;
[0009] S4: Building a support vector machine model, inputting the MFCC features of the two types of samples, hollow and solid, extracted from the data set into the support vector machine model for training to obtain target samples;
[0010] S5: Use the target sample to test the data set to obtain the recognition result of the target sample.
[0011] Furthermore, the preprocessing in S2 includes: setting 256 sampling points as one frame, and extracting a pulse signal with a set energy threshold based on a short-time average amplitude, wherein the pulse signal with a set energy threshold includes a single pulse signal.
[0012] Furthermore, the expression of MFCC features of two types of samples, hollow and solid, in the data set extracted in S3 is as follows:
[0013]
[0014] Among them, Cmfcc(i) is the MFCC feature parameter, N is the signal frame length, m(l) is the output of the Mel filter bank, l is the lth filter, i represents the i-th value, and L is the number of filter banks.
[0015] Furthermore, building the support vector machine model in S4 includes: finding an optimal hyperplane in the feature space.
[0016] Furthermore, the optimal hyperplane in S4 satisfies the following formula:
[0017]
[0018] Among them, ω and b represent the normal vector and constant term of the hyperplane respectively, st represents the satisfaction of the restriction conditions, (x i ) represents a mapping, ξ i is the relaxation factor, C represents the penalty factor, (x i ,y i ) represents a data sample.
[0019] Furthermore, the support vector machine model in S4 uses a kernel function to convert the low-dimensional feature vector into a linearly separable feature vector in a high-dimensional space.
[0020] Furthermore, the kernel function includes a radial basis function.
[0021] In a second aspect, the present invention provides a tunnel lining cavity based on a sound pulse signal.
[0022] Detection system,
[0023] include:
[0024] Picking module, preprocessing module, feature extraction module, modeling module and recognition module,
[0025] The pickup module is used to strike the hollow and solid parts of the tunnel lining, and respectively pick up the sound pulse signals generated after the hollow and solid parts of the tunnel lining are struck;
[0026] The preprocessing module is used to convert the sound pulse signal into a digital signal and then perform preprocessing, and extract the pulse signal with a set energy threshold to form a data set;
[0027] The feature extraction module is used to extract MFCC features of two types of samples, namely, hollow and solid, in the data set after dividing the pulse samples of unequal lengths in the data set into frames of equal number;
[0028] The modeling module is used to build a support vector machine model, and input the MFCC features of the two types of samples, hollow and solid, extracted from the data set into the support vector machine model for training to obtain target samples;
[0029] The recognition module is used to test the data set using the target sample to obtain a recognition result of the target sample.
[0030] Furthermore, the preprocessing module is used to set 256 sampling points as one frame, and extract a pulse signal with a set energy threshold based on the short-time average amplitude, wherein the pulse signal with a set energy threshold includes a single pulse signal.
[0031] Furthermore, the modeling module uses a kernel function to build a support vector machine model and converts a low-dimensional feature vector into a linearly separable feature vector in a high-dimensional space.
[0032] Compared with the prior art, the present invention has the following beneficial effects:
[0033] 1. The present invention uses acoustic means to identify the impact echo of the cavity, which is simpler than other methods and will not cause damage to the tunnel lining structure, and is highly practical;
[0034] 2. The feature parameters extracted by the present invention have the same dimension, and a variable frame length method is used to realize complete pulse signal analysis, which will not cause information loss, can obtain more pulse information, and improve the detection performance of the model;
[0035] 3. The present invention uses pattern recognition technology to build an SVM model to detect holes, which demonstrates good performance in classification and regression problems and obtains higher model accuracy and reliability. BRIEF DESCRIPTION OF THE DRAWINGS
[0036] In order to more clearly illustrate the technical solutions in the present invention, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.
[0037] Figure 1It is a flow chart of a tunnel lining cavity detection method based on sound pulse signals provided by the present invention;
[0038] FIG. 2( a ) is a time domain waveform of an acoustic signal of knocking on a tunnel lining provided by the present invention; FIG. 2( b ) is a time domain waveform of a single acoustic pulse signal provided by the present invention;
[0039] Figure 3 It is a schematic diagram of the improved MFCC extraction steps provided by the present invention;
[0040] Figure 4 It is a schematic diagram of a tunnel lining cavity detection system based on sound pulse signals provided by the present invention. DETAILED DESCRIPTION
[0041] In the following description, specific details such as specific system structures, technologies, etc. are provided for the purpose of illustration rather than limitation, so as to provide a thorough understanding of the embodiments of the present invention. However, it should be clear to those skilled in the art that the present invention may be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to prevent unnecessary details from obstructing the description of the present invention.
[0042] A method for detecting voids in a tunnel lining based on sound pulse signals provided by the present invention will be described in detail below with reference to the accompanying drawings.
[0043] Figure 1 It is a flow chart of a tunnel lining cavity detection method based on sound pulse signals provided by the present invention;
[0044] like Figure 1 As shown, the tunnel lining void detection method includes:
[0045] FIG. 2( a ) is a time domain waveform of a sound signal of knocking on a tunnel lining provided by the present invention.
[0046] S1: Hitting the hollow and solid parts of the tunnel lining to pick up the sound pulse signals generated by the hollow and solid parts of the tunnel lining respectively;
[0047] S2: converting the sound pulse signal into a digital signal and then preprocessing it, extracting the pulse signal with a set energy threshold to form a data set;
[0048] FIG. 2( b ) is a time domain waveform of a single acoustic pulse signal provided by the present invention.
[0049] The preprocessing in S2 includes: setting 256 sampling points as one frame, and extracting a pulse signal with a set energy threshold based on a short-time average amplitude, wherein the pulse signal with a set energy threshold includes a single pulse signal.
[0050]
[0051] Among them, x n (m) is the mth data value of the nth frame, and M is the analysis frame length for extracting pulses, which is 256. n It is a representation of the change in signal amplitude. It is compared with a preset threshold, and the continuous frames that exceed the threshold are extracted as a pulse sample. Figure 3 It is a schematic diagram of the improved MFCC extraction steps provided by the present invention.
[0052] S3: After dividing the pulse samples of unequal lengths in the data set into frames of equal number, extracting the MFCC features of the two types of samples, namely, hollow and solid, in the data set;
[0053] Dividing the pulse samples of unequal lengths in the data set into frames of equal number in S3 includes:
[0054] The Hamming window is used to uniformly divide the time length of the pulse samples in the data set into 5 frames.
[0055] For the classifier model, the input parameter dimension is required to be unified. Since the time length of the pulse signal generated by the knocking may be inconsistent, if only the traditional MFCC extraction method is used, the characteristic parameter dimension of the pulse signal with a longer time is also longer, and it needs to be intercepted to ensure the normal training of the model, which may cause some key information loss. Therefore, the present invention improves the MFCC extraction process: according to the time length of the initial pulse signal, it is uniformly divided into 5 frames using the Hamming window:
[0056] x w (n) = x(n)*w(n)
[0057] Among them, x w (n) is the windowed data, x(n) is a frame of sample data, w(n) represents the window function, and w(n) is:
[0058]
[0059] Wherein, N is the data frame length.
[0060] The step S3 further includes: if the number of sampling points in the individual frame is less than 2 I , then the frame is padded with zeros, where I is a natural number.
[0061] Regardless of the time length of the initial signal, all samples are divided into 5 frames. The difference is that the time length of each frame of samples of unequal length is different. Then MFCC is extracted from these 5 frames. If the number of sampling points in an individual frame is less than 2 I, the frame is padded with zeros. Through such processing, the dimension of the feature matrix extracted from pulse samples of unequal lengths is consistent, which meets the requirements of model input, and does not lose any pulse information, thereby improving the recognition accuracy.
[0062] After each frame is filled with zeros, the high-frequency part of the signal is emphasized, and then the frequency domain energy distribution is extracted using Fourier transform:
[0063]
[0064] Among them, the data frame length is equal to the Hamming window length.
[0065] X(k) is the Fourier transform result, where k is the frequency index.
[0066] The power spectrum is obtained by squaring the spectrum, and then filtered using the Mel filter bank. The energy after filtering is:
[0067]
[0068] Where L is the number of filter banks, W l (k) is the frequency response of the Mel filter bank, expressed as:
[0069]
[0070] Where o(l) represents the lth filter in the Mel filter bank. After logarithmic operation and discrete cosine transform, the improved MFCC parameters can be obtained.
[0071] The expression of MFCC features of two types of samples, hollow and solid, extracted from the data set in S3 is as follows:
[0072]
[0073] Among them, Cmfcc(i) is the MFCC feature parameter, N is the signal frame length, m(l) is the output of the Mel filter bank, l is the lth filter, i represents the i-th value, and L is the number of filter banks.
[0074] Finally, the first 15 MFCC coefficients are extracted from each frame, and each sample forms a 5*15 MFCC feature matrix.
[0075] S4: Building a support vector machine model, inputting the MFCC features of the two types of samples, hollow and solid, extracted from the data set into the support vector machine model for training to obtain target samples;
[0076] The construction of the support vector machine model in S4 includes: finding an optimal hyperplane in the feature space.
[0077] SVM is constructed as a classification model for detecting tunnel lining voids. SVM searches for the optimal hyperplane in the feature space.
[0078] Among them, SVM, full spelling Support Vector Machine, is a generalized linear classifier that performs binary classification of data in a supervised learning manner.
[0079] The optimal hyperplane in S4 satisfies the following formula:
[0080]
[0081] Among them, ω and b represent the normal vector and constant term of the hyperplane respectively, st represents the satisfaction of the restriction conditions, (x i ) represents a mapping, ξ i is the relaxation factor, C represents the penalty factor, (x i ,y i ) represents a data sample.
[0082] The support vector machine model in S4 uses a kernel function to convert a low-dimensional feature vector into a linearly separable feature vector in a high-dimensional space. The kernel function includes: a radial basis function.
[0083] SVM uses a kernel function to convert a low-dimensional feature vector into a linearly separable feature vector in a high-dimensional space, and uses the radial basis function (RBF) as the kernel function.
[0084] S5: Use the target sample to test the data set to obtain the recognition result of the target sample.
[0085] Figure 4 It is a schematic diagram of a tunnel lining cavity detection system based on sound pulse signals provided by the present invention.
[0086] Based on the same inventive concept, the present invention also provides a tunnel lining cavity detection system based on sound pulse signals, comprising:
[0087] Picking module, preprocessing module, feature extraction module, modeling module and recognition module,
[0088] The pickup module is used to strike the hollow and solid parts of the tunnel lining, and respectively pick up the sound pulse signals generated after the hollow and solid parts of the tunnel lining are struck;
[0089] The preprocessing module is used to convert the sound pulse signal into a digital signal and then perform preprocessing, and extract the pulse signal with a set energy threshold to form a data set;
[0090] The feature extraction module is used to extract MFCC features of two types of samples, namely, hollow and solid, in the data set after dividing the pulse samples of unequal lengths in the data set into frames of equal number;
[0091] The modeling module is used to build a support vector machine model, and input the MFCC features of the two types of samples, hollow and solid, extracted from the data set into the support vector machine model for training to obtain target samples;
[0092] The recognition module is used to test the data set using the target sample to obtain a recognition result of the target sample.
[0093] Example 1
[0094] The method of the present invention obtains C and ξ by training the pulse data set. i The optimal values of are C=1 and ξ i = 12. After testing the data set, the recognition accuracy of the improved MFCC reached 89.9%, while the recognition rate of the MFCC before the improvement was only 86.2%, an increase of 3.7 percentage points. The results show that the proposed method can effectively detect tunnel lining voids and effectively improve the recognition results of voids.
[0095] All the above optional technical solutions can be arbitrarily combined to form optional embodiments of the present application, which will not be described one by one here.
[0096] It should be understood that the order of execution of the steps in the above embodiment does not necessarily mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiment of the present invention.
[0097] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention, and should be included in the protection scope of the present invention.
Claims
1. A method for detecting voids in tunnel lining based on sound pulse signals, characterized in that: include: S1: Hitting the hollow and solid parts of the tunnel lining to pick up the sound pulse signals generated by the hollow and solid parts of the tunnel lining respectively; S2: converting the sound pulse signal into a digital signal and then preprocessing it, extracting the pulse signal with a set energy threshold to form a data set; S3: After dividing the pulse samples of unequal lengths in the data set into frames of equal number, extracting the MFCC features of the two types of samples, namely, hollow and solid, in the data set; S4: Building a support vector machine model, inputting the MFCC features of the two types of samples, hollow and solid, extracted from the data set into the support vector machine model for training to obtain target samples; S5: Using the target sample to test the data set to obtain a recognition result of the target sample; Dividing the pulse samples of unequal lengths in the data set into frames of equal number in S3 includes: Using a Hamming window, the time length of the pulse samples in the data set is uniformly divided into 5 frames; x w (n)=x(n)*w(n) Among them, x w (n) is the windowed data, x(n) is a frame of sample data, w(n) represents the window function, and w(n) is: Wherein, N is the data frame length; The step S3 further includes: if the number of sampling points in the individual frame is less than 2 I , then the frame is padded with zeros, where I is a natural number; Regardless of the initial signal time length, all samples are divided into 5 frames; then MFCC is extracted from these 5 frames. If the number of sampling points in an individual frame is less than 2 I , the frame is padded with zeros, so that the dimension of the feature matrix extracted from pulse samples of unequal lengths is consistent, which meets the requirements of model input, and does not lose any pulse information, thereby improving the recognition accuracy; After each frame is filled with zeros, the high-frequency part of the signal is emphasized, and then the frequency domain energy distribution is extracted using Fourier transform: Among them, the data frame length is equal to the Hamming window length; X(k) is the Fourier transform result, where k is the frequency index; The power spectrum is obtained by squaring the spectrum, and then filtered using the Mel filter bank. The energy after filtering is: Where L is the number of filter banks, W is l (k) is the frequency response of the Mel filter bank, expressed as: Among them, o(l) represents the lth filter in the Mel filter bank. After logarithmic operation and discrete cosine transform, the improved MFCC parameters can be obtained; The expression of MFCC features of two types of samples, hollow and solid, extracted from the data set in S3 is as follows: Among them, Cmfcc(i) is the MFCC feature parameter, N is the signal frame length, m(l) is the output of the Mel filter bank, l is the lth filter, i represents the i-th value, and L is the number of filter banks.
2. The tunnel lining void detection method according to claim 1, characterized in that: The preprocessing in S2 includes: setting 256 sampling points as one frame, and extracting a pulse signal with a set energy threshold based on a short-time average amplitude, wherein the pulse signal with a set energy threshold includes a single pulse signal.
3. The tunnel lining cavity detection method according to claim 1, characterized in that: The construction of the support vector machine model in S4 includes: finding an optimal hyperplane in the feature space.
4. The tunnel lining cavity detection method according to claim 3 is characterized in that: The optimal hyperplane in S4 satisfies the following formula: Among them, ω and b represent the normal vector and constant term of the hyperplane respectively, st represents the satisfaction of the restriction conditions, (x i ) represents a mapping, ξ i is the relaxation factor, C represents the penalty factor, (x i ,y i ) represents a data sample.
5. The tunnel lining cavity detection method according to claim 3, characterized in that: The support vector machine model in S4 uses a kernel function to convert a low-dimensional feature vector into a linearly separable feature vector in a high-dimensional space.
6. The method for detecting voids in tunnel lining according to claim 5, characterized in that: The kernel function includes a radial basis function.
7. A tunnel lining cavity detection system based on sound pulse signals, characterized in that: include: Picking module, preprocessing module, feature extraction module, modeling module and recognition module, The pickup module is used to strike the hollow and solid parts of the tunnel lining, and respectively pick up the sound pulse signals generated after the hollow and solid parts of the tunnel lining are struck; The preprocessing module is used to convert the sound pulse signal into a digital signal and then perform preprocessing, and extract the pulse signal with a set energy threshold to form a data set; The feature extraction module is used to extract MFCC features of two types of samples, namely, hollow and solid, in the data set after dividing the pulse samples of unequal lengths in the data set into frames of equal number; The modeling module is used to build a support vector machine model, and input the MFCC features of the two types of samples, hollow and solid, extracted from the data set into the support vector machine model for training to obtain target samples; The recognition module is used to test the data set using the target sample to obtain a recognition result of the target sample; The modeling module divides the pulse samples of unequal lengths in the data set into frames of equal number, including: Using a Hamming window, the time length of the pulse samples in the data set is uniformly divided into 5 frames; x w (n)=x(n)*w(n) Among them, x w (n) is the windowed data, x(n) is a frame of sample data, w(n) represents the window function, and w(n) is: Wherein, N is the data frame length; The modeling module further includes: if the number of sampling points in the individual frame is less than 2 I , then the frame is padded with zeros, where I is a natural number; Regardless of the initial signal time length, all samples are divided into 5 frames; then MFCC is extracted from these 5 frames. If the number of sampling points in an individual frame is less than 2 I , the frame is padded with zeros, so that the dimension of the feature matrix extracted from pulse samples of unequal lengths is consistent, which meets the requirements of model input, and does not lose any pulse information, thereby improving the recognition accuracy; After each frame is filled with zeros, the high-frequency part of the signal is emphasized, and then the frequency domain energy distribution is extracted using Fourier transform: Among them, the data frame length is equal to the Hamming window length; X(k) is the Fourier transform result, where k is the frequency index; The power spectrum is obtained by squaring the spectrum, and then filtered using the Mel filter bank. The energy after filtering is: Where L is the number of filter banks, W is l (k) is the frequency response of the Mel filter bank, expressed as: Among them, o(l) represents the lth filter in the Mel filter bank. After logarithmic operation and discrete cosine transform, the improved MFCC parameters can be obtained; The expression of MFCC features of two types of samples, hollow and solid, extracted from the data set in the modeling module is as follows: Among them, Cmfcc(i) is the MFCC feature parameter, N is the signal frame length, m(l) is the output of the Mel filter bank, l is the lth filter, i represents the i-th value, and L is the number of filter banks.
8. The tunnel lining void detection system according to claim 7, characterized in that: The preprocessing module is used to set 256 sampling points as one frame, and extract a pulse signal with a set energy threshold based on a short-time average amplitude, wherein the pulse signal with a set energy threshold includes a single pulse signal.
9. The tunnel lining void detection system according to claim 8, characterized in that: The modeling module uses a kernel function to build a support vector machine model and converts a low-dimensional feature vector into a linearly separable feature vector in a high-dimensional space.
