CN112557510A - Concrete pavement void intelligent detection device and detection method thereof - Google Patents
Concrete pavement void intelligent detection device and detection method thereof Download PDFInfo
- Publication number
- CN112557510A CN112557510A CN202011463253.4A CN202011463253A CN112557510A CN 112557510 A CN112557510 A CN 112557510A CN 202011463253 A CN202011463253 A CN 202011463253A CN 112557510 A CN112557510 A CN 112557510A
- Authority
- CN
- China
- Prior art keywords
- void
- module
- sample
- detection
- sound
- 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.)
- Pending
Links
- 239000011800 void material Substances 0.000 title claims abstract description 170
- 238000001514 detection method Methods 0.000 title claims abstract description 132
- 238000001228 spectrum Methods 0.000 claims abstract description 25
- 238000000034 method Methods 0.000 claims abstract description 21
- 230000009467 reduction Effects 0.000 claims abstract description 20
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 19
- 125000004122 cyclic group Chemical group 0.000 claims abstract description 15
- 238000010586 diagram Methods 0.000 claims abstract description 10
- 238000012545 processing Methods 0.000 claims abstract description 10
- 230000003044 adaptive effect Effects 0.000 claims abstract description 6
- 239000000523 sample Substances 0.000 claims description 110
- 239000013598 vector Substances 0.000 claims description 65
- 238000012549 training Methods 0.000 claims description 51
- 239000013074 reference sample Substances 0.000 claims description 42
- 238000013528 artificial neural network Methods 0.000 claims description 26
- 230000005284 excitation Effects 0.000 claims description 26
- 230000002457 bidirectional effect Effects 0.000 claims description 18
- 229910000831 Steel Inorganic materials 0.000 claims description 17
- 239000010959 steel Substances 0.000 claims description 17
- 238000000605 extraction Methods 0.000 claims description 12
- 239000011159 matrix material Substances 0.000 claims description 12
- 230000000306 recurrent effect Effects 0.000 claims description 8
- 230000004913 activation Effects 0.000 claims description 7
- 238000009434 installation Methods 0.000 claims description 4
- 238000012360 testing method Methods 0.000 claims description 3
- 238000001914 filtration Methods 0.000 abstract description 8
- 238000005070 sampling Methods 0.000 abstract description 4
- 230000008901 benefit Effects 0.000 abstract description 3
- 238000003062 neural network model Methods 0.000 abstract 1
- 230000000638 stimulation Effects 0.000 abstract 1
- 230000001629 suppression Effects 0.000 abstract 1
- 230000005236 sound signal Effects 0.000 description 10
- 230000006870 function Effects 0.000 description 7
- 230000006872 improvement Effects 0.000 description 7
- 238000007781 pre-processing Methods 0.000 description 7
- 238000004364 calculation method Methods 0.000 description 6
- 238000006243 chemical reaction Methods 0.000 description 4
- 230000009471 action Effects 0.000 description 3
- 230000008569 process Effects 0.000 description 3
- 238000004458 analytical method Methods 0.000 description 2
- 230000002401 inhibitory effect Effects 0.000 description 2
- 238000011176 pooling Methods 0.000 description 2
- 238000009825 accumulation Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 230000021615 conjugation Effects 0.000 description 1
- DMBHHRLKUKUOEG-UHFFFAOYSA-N diphenylamine Chemical compound C=1C=CC=CC=1NC1=CC=CC=C1 DMBHHRLKUKUOEG-UHFFFAOYSA-N 0.000 description 1
- 238000009432 framing Methods 0.000 description 1
- 238000012423 maintenance Methods 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000000149 penetrating effect Effects 0.000 description 1
- 230000008447 perception Effects 0.000 description 1
- 230000004044 response Effects 0.000 description 1
- 230000003595 spectral effect Effects 0.000 description 1
- 239000002352 surface water Substances 0.000 description 1
- 230000001052 transient effect Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N29/00—Investigating or analysing materials by the use of ultrasonic, sonic or infrasonic waves; Visualisation of the interior of objects by transmitting ultrasonic or sonic waves through the object
- G01N29/04—Analysing solids
- G01N29/045—Analysing solids by imparting shocks to the workpiece and detecting the vibrations or the acoustic waves caused by the shocks
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N29/00—Investigating or analysing materials by the use of ultrasonic, sonic or infrasonic waves; Visualisation of the interior of objects by transmitting ultrasonic or sonic waves through the object
- G01N29/44—Processing the detected response signal, e.g. electronic circuits specially adapted therefor
- G01N29/4472—Mathematical theories or simulation
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N29/00—Investigating or analysing materials by the use of ultrasonic, sonic or infrasonic waves; Visualisation of the interior of objects by transmitting ultrasonic or sonic waves through the object
- G01N29/44—Processing the detected response signal, e.g. electronic circuits specially adapted therefor
- G01N29/4481—Neural networks
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
- G01S19/00—Satellite radio beacon positioning systems; Determining position, velocity or attitude using signals transmitted by such systems
- G01S19/38—Determining a navigation solution using signals transmitted by a satellite radio beacon positioning system
- G01S19/39—Determining a navigation solution using signals transmitted by a satellite radio beacon positioning system the satellite radio beacon positioning system transmitting time-stamped messages, e.g. GPS [Global Positioning System], GLONASS [Global Orbiting Navigation Satellite System] or GALILEO
- G01S19/42—Determining position
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
- G01S5/00—Position-fixing by co-ordinating two or more direction or position line determinations; Position-fixing by co-ordinating two or more distance determinations
- G01S5/18—Position-fixing by co-ordinating two or more direction or position line determinations; Position-fixing by co-ordinating two or more distance determinations using ultrasonic, sonic, or infrasonic waves
- G01S5/20—Position of source determined by a plurality of spaced direction-finders
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N2291/00—Indexing codes associated with group G01N29/00
- G01N2291/02—Indexing codes associated with the analysed material
- G01N2291/023—Solids
- G01N2291/0232—Glass, ceramics, concrete or stone
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N2291/00—Indexing codes associated with group G01N29/00
- G01N2291/02—Indexing codes associated with the analysed material
- G01N2291/028—Material parameters
- G01N2291/0289—Internal structure, e.g. defects, grain size, texture
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Remote Sensing (AREA)
- Radar, Positioning & Navigation (AREA)
- Immunology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Chemical & Material Sciences (AREA)
- Analytical Chemistry (AREA)
- Biochemistry (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Pathology (AREA)
- Signal Processing (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Algebra (AREA)
- Mathematical Analysis (AREA)
- Mathematical Optimization (AREA)
- Mathematical Physics (AREA)
- Pure & Applied Mathematics (AREA)
- Acoustics & Sound (AREA)
- Computer Networks & Wireless Communication (AREA)
- Measurement Of Mechanical Vibrations Or Ultrasonic Waves (AREA)
Abstract
The invention discloses an intelligent detection device and a detection method for concrete pavement void, which comprises the steps that a microphone array collects aliasing sounds of a stimulation device hitting a detected pavement, the DOA direction of the aliasing sounds is judged through a sound source DOA direction estimation algorithm, a clear target sound sample set is obtained by combining an adaptive beam forming algorithm, sub-band filtering is carried out through a noise reduction module to obtain a logarithmic Mel frequency spectrum diagram of a signal, a void detection model is trained on one part of the logarithmic Mel frequency spectrum diagram, and the other part of the logarithmic Mel frequency spectrum diagram is substituted into the void identification model to detect the void accuracy. The method adopts a microphone array acquisition mode to carry out time, space and frequency three-domain combined processing on the space sampling information, realizes the positioning, noise suppression and signal enhancement of the sound source, integrates a convolutional neural network model and a cyclic neural network model, carries out accurate and intelligent judgment on the road surface void in an interference environment, and has the advantages of high recognition rate, simplicity in operation, rapidness, accuracy, economy, no damage and low cost.
Description
Technical Field
The invention belongs to the technical field of road bed and pavement detection, and particularly relates to an intelligent concrete pavement void detection device and a detection method thereof.
Background
In practical application, the in-service concrete pavement is affected by severe environments such as vehicle-mounted load, high temperature, surface water accumulation and the like for a long time, various damages such as void and the like of the pavement are directly caused, and under repeated load, if the pavement cannot be identified and repaired in time, the service life is shortened, and further the driving safety is damaged.
The currently common void detection methods mainly comprise: the manual observation method is mainly used for subjectively judging according to experience, the identification accuracy is greatly influenced by human factors, different people can obtain completely opposite conclusions, and great errors exist; the deflection detection method is used on the premise that the road surface plate angle, the deflection value and the plate data are known, so that the method is only suitable for individual road sections, and in addition, professional technicians are required to operate and maintain instruments, and the detection cost is high; the ground penetrating radar detection method is restricted by dielectric constant, frequency and detection depth of a medium, manual interpretation and detection of the image are difficult, and a radar detection system is high in maintenance cost and expensive in price; the existing sound vibration method is difficult to judge and read manually in an interference environment, and the detection accuracy is not ideal enough.
Disclosure of Invention
The invention provides an intelligent concrete pavement void detection device based on a sound vibration detection method, which has the advantages of higher timeliness, higher stability and lower cost, and still keeps excellent accuracy in a strong interference environment.
In order to achieve the purpose, the technical scheme of the invention is as follows:
an intelligent detection device for concrete pavement void, which comprises a support plate, a computer control system, a display screen and an excitation device; a through groove is formed in the middle of the supporting plate, two sides of the through groove are connected with the excitation device, the top of one side of the supporting plate is connected with the computer control system, and the bottom of the supporting plate is provided with a roller; the top of the computer control system is connected with a display screen; the top of the display screen is provided with a voice broadcast device;
the excitation device comprises a rotating shaft, a motor, an elastic steel bar and a steel ball; the middle parts of the two sides of the through groove are hinged with two ends of a rotating shaft, and one end of the rotating shaft is connected with a motor arranged on the outer side of the supporting plate; the rotating shafts are uniformly distributed with rotating discs; elastic steel bars are uniformly distributed on the outer ring of the turntable and connected with steel balls, and the steel balls rotate to the lowest end of the turntable and are lower than the rollers;
the both sides that lead to the groove in backup pad middle part upwards extend, install the bracing piece between the extension department, bracing piece equipartition microphone, and the quantity of microphone is no less than 4, is on a parallel with the installation of excitation device.
As a further technical improvement, the display screen is internally provided with a storage module, a display module and an alarm module which are sequentially connected, and the alarm module is connected with a voice broadcast device.
As a further technical improvement, the microphone comprises a microphone array pickup module and a microphone array signal receiving module; and the microphone array pickup module is connected with the microphone array signal receiving module.
As a further technical improvement, the computer control system comprises an excitation device control module, a sound source direction estimation module, a noise reduction module, a feature extraction module, a void detection model training module, a void detection model detection module and a Beidou GNSS positioning module; a sound source DOA direction estimation algorithm is arranged in the sound source direction estimation module; the noise reduction module is internally provided with a self-adaptive beam forming algorithm; the excitation device control module, the microphone array pickup module, the microphone array signal receiving module, the sound source direction estimation module, the noise reduction module, the feature extraction module, the void detection model training module, the void detection model detection module and the Beidou GNSS positioning module are connected in sequence; the excitation device control module is connected with the motor; the sound source direction estimation module is connected with the microphone; the display screen is respectively connected with the air-out detection model detection module and the Beidou GNSS positioning module.
Microphone array pickup module for adopt the excitation device to strike the sound on surveyed road surface, microphone array pickup module is even linear array, including at least 4 MEMS microphones, be on a parallel with the installation of excitation device, microphone array pickup module includes: the A/D conversion sub-module is used for converting the collected 4 or more paths of sound analog signals into digital signals; the demodulation and pretreatment submodule is used for preprocessing, demodulating and filtering the digital signal output by the A/D conversion submodule. The demodulation and preprocessing sub-module is used for preprocessing the digital signal, wherein the preprocessing is to adopt a self-adaptive subband spectrum entropy endpoint detection algorithm to detect a sound signal so as to extract a useful sound segment under a noise background; then, 4 paths of digital signals with the sound sampling rate of 16KB and 16bit above are obtained through demodulation and filtering processing; and a signal receiving submodule is arranged in the microphone array signal receiving module and is used for receiving the processed signals and transmitting the processed signals to a computer control system.
The exciting device control module controls the frequency of striking the soil-mixed pavement; the sound source direction estimation module is used for estimating the position directions of a target signal and an interference signal, the sound source direction estimation module is provided with a sound source DOA direction estimation algorithm, and the sound source DOA direction estimation algorithm is an improved MUSIC algorithm and specifically comprises the following steps:
1) taking the first array element of the linear array as a reference, the signals received by the array are:
X(n)＝AS(t)+U(t)； (1)
2) order: y (n) ═ TNX*(n)； (2)
Wherein: t isNIs an N × N switching matrix with elements of the negative diagonal being 1 and the remainder being 0, X*(N) is the conjugate matrix of X (N), N being the number of microphones of the microphone array;
3) and (3) solving a covariance matrix of Y (n) according to (1) and (2): rYY＝E[Y(n)YH(n)]＝ARSSAH+δ2I； (3)
4) Calculating the total covariance
5) According to the corrected total covariance
Carrying out direction estimation according to a traditional MUSIC algorithm to obtain direction information of target sound and interference;
the noise reduction module is used for carrying out noise reduction and reverberation removal processing according to the target signal and the interference direction information output by the sound source direction estimation module, and inhibiting the signal in the interference direction to obtain a clear target signal; the noise reduction module is provided with a self-adaptive beam forming algorithm, namely a GSC beam forming algorithm, and comprises the following specific steps: according to the direction information of a target sound source and an interference sound source obtained by a sound source DOA direction estimation algorithm, namely an improved MUSIC algorithm, calculating the optimal weighting vector of the direction of the target signal, and finally obtaining a clear target sound sample set; the characteristic extraction module is a logarithmic Mel frequency spectrum characteristic extraction module of the target sound sample set and is used for extracting logarithmic Mel frequency spectrum characteristics of the target sound sample set output by the noise reduction module;
the system comprises a void detection model training module, a data processing module and a data processing module, wherein the void detection model training module is used for training a void detection model by utilizing logarithmic Mel frequency spectrum characteristics of a target sound sample set, and the void detection model consists of a convolutional neural network and a bidirectional Bi-LSTM cyclic neural network; GLU gate control linear units are introduced into the convolution layers of the convolution neural network, and the convolution neural network is a bidirectional Bi-LSTM network; the target sound sample set comprises a void sample, a non-void sample and a reference sample; the method is characterized in that various types of sound characteristics of the void state are stored in the void state sound data parameter stock, the optimal characteristic parameters are obtained through the training and testing of a void detection model, the sound sample set is a reference sample, a void sample and a non-void sample which are labeled and classified, and the reference sample and the void sample have similar characteristics:
the method for detecting the target sound sample set characteristic training void comprises the following steps:
1) building the convolution-cycle network model;
2) inputting the 3/4 log mel frequency spectrum diagram of the sound sample set into the convolutional neural network to obtain first vector data of the reference sample, the void sample and the non-void sample;
3) inputting first vector data of a reference sample, a void sample and a non-void sample into the bidirectional Bi-LSTM recurrent neural network respectively to obtain second vector data;
4) inputting second vector data of the reference sample, the void sample and the non-void sample into a softmax classifier respectively to obtain third vector data;
5) and calculating a triple loss value according to the third vector data of the void sample, the non-void sample and the reference sample, and optimizing the void detection model according to the triple loss value.
6) Establishing a void detection template, substituting the residual 1/4 logarithmic Mel frequency spectrum diagram into the void identification model, and detecting the void accuracy;
the void detection model training module comprises: the device comprises a feature prediction first sub-module, a feature prediction second sub-module, a feature prediction third sub-module, a ternary loss value calculation operator module and a training end judgment sub-module;
the feature prediction first submodule is used for inputting the logarithmic Mel frequency spectrum features of the 3/4 sound sample set into a convolution cyclic neural network, the convolution neural network comprises a plurality of convolution layers and pooling layers, furthermore, an activation function of each convolution layer adopts a GLU gated cyclic linear unit, and the final output of the convolution cyclic neural network is a feature prediction first vector of a reference sample, a void sample and a non-void sample;
the characteristic prediction second sub-module is used for inputting the characteristic prediction first vector of the reference sample, the void sample and the non-void sample into the bidirectional Bi-LSTM recurrent neural network to obtain a characteristic prediction second vector;
the feature prediction third sub-module is used for enabling the feature prediction second vectors of the reference sample, the void sample and the non-void sample to act on the softmax activation function of the classification layer to obtain a feature prediction third vector;
the ternary loss value operator module is used for predicting a third vector according to the characteristics of the reference sample, the void sample and the non-void sample and calculating a ternary loss value;
further, the ternary loss value operator module comprises: a first calculating unit, a second calculating unit and a third calculating unit; the first calculating unit is used for calculating the Euclidean distance between feature prediction third vectors of the reference sample and the void sample to obtain a first distance; the second calculation unit is used for calculating the Euclidean distance between feature prediction third vectors of the reference sample and the non-null sample to obtain a second distance; the third calculation unit is used for calculating a ternary loss value according to the first distance, the second distance and a preset minimum distance constant;
a training end judgment submodule, configured to judge whether to end the training of the void detection model according to the ternary loss value and a preset loss minimum threshold, where the training end judgment submodule specifically includes: a continuous training unit and an end training unit; when the ternary loss value is larger than a preset loss minimum threshold value, the continuous training unit adjusts the model parameters of the convolutional neural network and the bidirectional Bi-LSTM circulating network and continues training; when the ternary loss value is smaller than a preset loss minimum threshold value, the training ending unit ends training to obtain a void detection model, and substitutes the residual 1/4 logarithmic Mel frequency spectrograms into the void recognition model to detect the void accuracy;
the detection module of the void detection model is used for inputting the logarithmic Mel frequency spectrum characteristics of the sound signal to be detected into the void detection model to obtain a classification result and probability;
the Beidou GNSS positioning module is used for accurately acquiring the geographic information of the current detection position; the display module is used for visually displaying the detection result and the position information; the alarm module sends an instruction to the voice broadcaster to send alarm information to prompt a detector when the detection result is in an empty state; and the storage module is used for storing all data and programs of the detection result.
The detection method of the intelligent detection device for concrete pavement void comprises the following steps:
s1: collecting aliasing sounds of a tested road surface hit by an exciting device under different environments by using a microphone array, converting M analog signals output by the microphone array in a pickup mode into M digital signals with preset bit rate and bit positions, carrying out end point detection, and then intercepting end point signals to obtain digital signals of multi-path aliasing sounds;
s2: transmitting the digital signals of the multiple paths of aliasing sounds of S1 to a sound source direction estimation module, performing sound source DOA direction estimation on the digital signals of the multiple paths of aliasing sounds of S1 by a sound source DOA direction estimation algorithm of the sound source direction estimation module, and performing enhancement and denoising processing on the digital signals of the multiple paths of aliasing sounds after the direction estimation by combining with an adaptive beam forming algorithm of a noise reduction module to obtain a clear target sound sample set;
s3: obtaining logarithmic Mel frequency spectrum characteristics of the target sound sample set of S2;
s4: establishing a sound data parameter base in a void state, and training a void detection model according to the characteristics of the target sound sample set of S3, wherein the void detection model consists of a convolutional neural network and a cyclic neural network;
s5: extracting a logarithmic Mel frequency spectrum diagram of a sound sample to be detected, carrying out intelligent identification by using a trained void detection model, and judging the void state of the detected sound sample;
s6: and displaying the identification detection result through a display screen, storing and broadcasting the voice.
As a further technical improvement, the sound samples comprise a void sample, a non-void sample and a reference sample; the sound data parameter stock of the void state is stored with various void state sound characteristics, and the optimal characteristic parameters are obtained through the void detection model training test.
As a further technical improvement, a GLU gating linear unit is introduced into a convolution layer of the convolutional neural network, and the convolutional neural network is a bidirectional Bi-LSTM network; the method for detecting the target sound sample set characteristic training void comprises the following steps:
1) building the convolution-cycle network model;
2) inputting the 3/4 log mel frequency spectrum diagram of the sound sample set into the convolutional neural network to obtain first vector data of the reference sample, the void sample and the non-void sample;
3) inputting first vector data of a reference sample, a void sample and a non-void sample into the bidirectional Bi-LSTM recurrent neural network respectively to obtain second vector data;
4) inputting second vector data of the reference sample, the void sample and the non-void sample into a softmax classifier respectively to obtain third vector data;
5) and calculating a triple loss value according to the third vector data of the void sample, the non-void sample and the reference sample, and optimizing the void detection model according to the triple loss value.
6) And establishing a void detection template, and substituting the residual 1/4 logarithmic Mel frequency spectrogram into the void identification model to detect the void accuracy.
As a further technical improvement, the triple loss is calculated according to the third vector data of the void sample, the non-void sample and the reference sample, wherein the method comprises the following steps:
solving the Euclidean distance of the third vector data of the void sample and the reference sample to obtain a first distance;
solving the Euclidean distance of the third vector data of the non-void sample and the reference sample to obtain a second distance;
calculating a difference between the first distance and the second distance; and determining a triple loss value according to the difference value and a preset minimum distance.
As a further technical improvement, the method for optimizing the void detection model according to the triplet loss value comprises the following steps:
if the ternary loss value is smaller than a preset loss threshold value, ending the training to obtain a final void detection model;
and if the ternary loss value is greater than the preset loss threshold value, adjusting the model parameters, and continuing training until the ternary loss value is less than the preset loss threshold value.
Compared with the prior art, the invention has the following beneficial effects:
1. the invention drives the steel balls on the turntable to hit the tested road surface by driving the rotating shaft, different sounds are fed back from different emptying states of the road surface, and the linear array arrangement is adopted by the microphone to realize the linear pickup of sound signals; the sound source direction estimation module is added, the improved MUSIC algorithm is utilized to judge the directions of the expected signal and the interference signal, and then the GSC beam forming algorithm is utilized to realize spatial filtering according to the determined target sound and the interference direction information, so that the noise with directivity is well inhibited, and the clear target sound signal is extracted.
2. The invention adopts a logarithmic Mel frequency spectrogram as the frequency change of the description signal, and effectively identifies and classifies the void sound.
3. The void detection model is formed by cascading a convolutional neural network and a cyclic neural network, and the log Mel frequency spectrum structure information of various void state sounds is fully learned by utilizing the convolutional neural network; the cyclic neural network introduces context information of the sound segments for modeling, so that the model has strong generalization capability, and higher detection rate and stronger robustness are achieved.
4. According to the method, the GLU gate control linear unit is introduced into the convolutional layer to serve as an activation function, the contribution value of each frame of feature to detection classification is adjusted, irrelevant features are abandoned, the extraction of distinguishing features is facilitated, and model convergence and training are simpler.
5. The invention adopts Bi-LTSM as a cyclic convolution neural network, simultaneously obtains time information in both front and back directions, introduces future information and improves the detection performance in an interference environment.
Drawings
Fig. 1 is a schematic front view of an excitation device.
Fig. 2 is a schematic side view of the excitation device.
FIG. 3 is a schematic view of the cross-sectional structure A-A.
FIG. 4 is a flow chart of the steps of the detection method of the present invention.
Fig. 5 is a block diagram of the structure of the present invention.
Wherein, each icon and the corresponding part name are as follows:
1-supporting plate, 11-roller, 2-computer control system, 3-display screen, 4-voice broadcaster, 5-excitation device, 51-rotating shaft, 52-motor, 53-steel ball, 6-supporting rod, 7-microphone.
Detailed Description
In order to make the aforementioned objects, features and advantages of the present invention comprehensible, embodiments accompanied with figures are described in detail below.
Example (b):
as shown in fig. 1-5, the intelligent detection device for concrete pavement void of the embodiment comprises a support plate 1, a computer control system 2, a display screen 3 and an excitation device 5; the middle part of the supporting plate 1 is hollow, an exciting device 5 is arranged at the hollow part, the bottom of the supporting plate 1 is provided with a roller 11 for moving, and the top of one side of the supporting plate 1 is connected with a computer control system 2; the top of the computer control system 2 is connected with a display screen 3; the display screen 3 is internally provided with a storage module, a display module and an alarm module, the top of the display screen is provided with a voice broadcast device 4, and the storage module, the display module, the alarm module and the voice broadcast device 4 are electrically connected in sequence;
the excitation device 5 comprises a rotating shaft 51, a motor 52 and a steel ball 53; the middle parts of two sides of the hollow part of the supporting plate 1 are hinged with two ends of a rotating shaft 51, and one end of the rotating shaft 51 is connected with a motor 52 arranged on the outer side of the supporting plate 1; the rotating shafts 51 are uniformly distributed with rotating discs; the outer ring of the turntable is connected with a steel ball 53 through an elastic steel bar, and the steel ball 53 rotates to the lowest end position and is lower than the roller 11;
two sides of the hollow part in the supporting plate 1 extend upwards, supporting rods 6 are arranged among the extending parts, microphones 7 with downward detection ends are uniformly distributed on the supporting rods 6, and the number of the microphones 7 is not less than 4 and is arranged in parallel to the exciting device 5; the microphone 7 comprises a microphone array pickup module and a microphone array signal receiving module;
the computer control system 2 comprises an excitation device control module, a sound source direction estimation module, a noise reduction module, a feature extraction module, a void detection model training module, a void detection model detection module and a Beidou GNSS positioning module;
the excitation device control module, the microphone array pickup module, the microphone array signal receiving module, the sound source direction estimation module, the noise reduction module, the feature extraction module, the void detection model training module, the void detection model detection module and the Beidou GNSS positioning module are electrically connected in sequence; the excitation device control module is electrically connected with the motor 52; the microphone array signal receiving module is electrically connected with a microphone 7; the display screen 3 respectively with take off and detect model detection module and big dipper GNSS orientation module electric connection.
Microphone array pickup module for adopt the excitation device to strike the sound on surveyed road surface, microphone array pickup module is even linear array, including at least 4 MEMS microphones, be on a parallel with the installation of excitation device, microphone array pickup module includes: the A/D conversion sub-module is used for converting the collected 4 or more paths of sound analog signals into digital signals; the demodulation and pretreatment submodule is used for preprocessing, demodulating and filtering the digital signal output by the A/D conversion submodule. The demodulation and preprocessing submodule 2022 preprocesses the digital signal, wherein the preprocessing is to detect a sound signal by adopting an adaptive subband spectral entropy endpoint detection algorithm so as to extract a useful sound segment under a noise background; then, 4 paths of digital signals with the sound sampling rate of 16KB and 16bit above are obtained through demodulation and filtering processing;
the sound source direction estimation module is used for estimating the position directions of a target signal and an interference signal, the sound source direction estimation module is provided with a sound source DOA direction estimation algorithm, and the sound source DOA direction estimation algorithm is an improved MUSIC algorithm and specifically comprises the following steps:
1) taking the first array element of the linear array as a reference, the signals received by the array are:
X(n)＝AS(t)+U(t)； (1)
2) order: y (n) ═ TNX*(n)； (2)
Wherein: t isNIs an N × N switching matrix with elements of the negative diagonal being 1 and the remainder being 0, X*(N) is the conjugate matrix of X (N), N being the number of microphones of the microphone array;
3) and (3) solving a covariance matrix of Y (n) according to (1) and (2): rYY＝E[Y(n)YH(n)]＝ARSSAH+δ2I； (3)
4) Calculating the total covariance
5) According toCorrected total covariance
Carrying out direction estimation according to a traditional MUSIC algorithm to obtain direction information of target sound and interference;
the noise reduction module is used for carrying out noise reduction and reverberation removal processing according to the target signal and the interference direction information output by the sound source direction estimation module, and inhibiting the signal in the interference direction to obtain a clear target signal; the noise reduction module is provided with a self-adaptive beam forming algorithm, namely a GSC beam forming algorithm, and comprises the following specific steps: and calculating the optimal weighted vector of the target signal direction according to the direction information of the target sound source and the interference sound source obtained by the sound source DOA direction estimation algorithm, namely the improved MUSIC algorithm, and finally obtaining a clear digital signal of the target direction.
The characteristic extraction module is a target sample logarithmic Mel frequency spectrum characteristic extraction module and is used for extracting logarithmic Mel frequency spectrum characteristics of the sound digital signals output by the noise reduction module;
the system comprises a training module of a void detection model, a model analysis module and a model analysis module, wherein the training module of the void detection model is used for training the void detection model by utilizing the logarithmic Mel frequency spectrum characteristics of 3/4 sound sample sets, the void detection model consists of a convolutional neural network and a bidirectional Bi-LSTM cyclic neural network, the sound sample sets are labeled and classified reference samples, void samples and non-void samples, and the reference samples and the void samples have similar characteristics:
the void detection model training module comprises: the device comprises a feature prediction first sub-module, a feature prediction second sub-module, a feature prediction third sub-module, a ternary loss value calculation operator module and a training end judgment sub-module;
the feature prediction first submodule is used for inputting the logarithmic Mel frequency spectrum features of the sound sample set into a convolution cyclic neural network, the convolution neural network comprises a plurality of convolution layers and a pooling layer, furthermore, an activation function of each convolution layer adopts a GLU gated cyclic linear unit, and the final output of the convolution cyclic neural network is a feature prediction first vector of a reference sample, a void sample and a non-void sample;
the characteristic prediction second sub-module is used for inputting the characteristic prediction first vector of the reference sample, the void sample and the non-void sample into the bidirectional Bi-LSTM recurrent neural network to obtain a characteristic prediction second vector;
the feature prediction third sub-module is used for enabling the feature prediction second vectors of the reference sample, the void sample and the non-void sample to act on the softmax activation function of the classification layer to obtain a feature prediction third vector;
and the ternary loss value operator module is used for predicting a third vector according to the features of the reference sample, the void sample and the non-void sample and calculating a ternary loss value.
Further, the ternary loss value operator module comprises: a first calculating unit, a second calculating unit and a third calculating unit; the first calculating unit is used for calculating the Euclidean distance between feature prediction third vectors of the reference sample and the void sample to obtain a first distance; the second calculation unit is used for calculating the Euclidean distance between feature prediction third vectors of the reference sample and the non-null sample to obtain a second distance; and the third calculation unit calculates a ternary loss value according to the first distance, the second distance and a preset minimum distance constant.
A training end judgment submodule, configured to judge whether to end the training of the void detection model according to the ternary loss value and a preset loss minimum threshold, where the training end judgment submodule specifically includes: a continuous training unit and an end training unit; when the ternary loss value is larger than a preset loss minimum threshold value, the continuous training unit adjusts the model parameters of the convolutional neural network and the bidirectional Bi-LSTM circulating network and continues training; and when the ternary loss value is smaller than a preset loss minimum threshold value, the training ending unit ends the training to obtain an empty detection model, and substitutes the residual 1/4 logarithmic Mel frequency spectrograms into the empty recognition model to detect the empty accuracy.
And the void detection model detection module is used for inputting the logarithmic Mel frequency spectrum characteristics of the sound signal to be detected into the void detection model to obtain a classification result and probability.
And the Beidou GNSS positioning module is used for accurately acquiring the geographic information of the current detection position.
And the display module is used for visually displaying the detection result and the position information.
Alarm module works as detection result for the state of coming to nothing, alarm module sends the instruction to the voice broadcast ware, sends alarm information, suggestion measurement personnel.
And the storage module is used for storing all data and programs of the detection result.
The detection method of the embodiment comprises the following steps:
s1: collecting sounds of concrete pavements in different void states hit by different excitation devices by using a microphone array in the environments of signal to noise ratios of 5dB, 10dB and 25dB, wherein the microphone array is a uniform linear array and at least comprises 4 digital microphones; detecting the sound signal by adopting an adaptive subband spectrum entropy endpoint detection algorithm to extract a useful sound segment under a noise background; after demodulation and filtering processing, 4 paths of digital signals with the sampling rate of 16KB and 16bit are obtained;
s2: estimating the direction of the sound signal, and performing noise reduction by combining a GSC beam forming algorithm to obtain a clear target sound sample set;
the method for estimating the direction of the sound signal comprises the following steps:
1) taking the first array element of the linear array as a reference, the signals received by the array are:
X(n)＝AS(t)+U(t)； (1)
2) order: y (n) ═ TNX*(n)； (2)
Wherein: t isNIs an N × N switching matrix with elements of the negative diagonal being 1 and the remainder being 0, X*(N) is the conjugate matrix of X (N), N being the number of microphones of the microphone array;
3) and (3) solving a covariance matrix of Y (n) according to (1) and (2): rYY＝E[Y(n)YH(n)]＝ARSSAH+δ2I； (3)
4) Calculating the total covariance
5) According to the corrected total covariance
And carrying out direction estimation according to the traditional MUSIC algorithm to obtain the direction information of the target sound and the interference.
The adaptive beamforming algorithm is a noise reduction method for a GSC beamforming algorithm as follows:
1) obtaining direction signals of a target sound source and an interference sound source according to the sound source DOA direction estimation algorithm;
2) the direction signal is adopted to pass through the optimal weighting vector W0 of the conjugation of the direction vector, and then the output y (t) is calculated;
3) solving to obtain a clear target sound digital signal after filtering;
s3: the characteristics of the sound digital signal sample set are obtained, a logarithmic Mel spectrogram is extracted as the characteristics in the embodiment, the logarithmic Mel spectrogram calculates sound frame by frame, the transient dynamic characteristics of a sound source can be captured, frequency response similar to human auditory perception is mapped, and the characteristics of a sound signal can be well represented; the extraction method comprises the following steps:
dividing the sound digital signal into 2s time length segments, carrying out framing and windowing, converting the time domain signal into frequency domain characteristics through FFT (fast Fourier transform), and compressing through a Mel scale filter and a logarithmic dynamic range to obtain a logarithmic Mel spectrogram.
S4: training a null-out detection model according to the logarithmic Mel frequency spectrogram of the target sound sample set, wherein the null-out detection model consists of a convolutional neural network and a bidirectional Bi-LSTM recurrent neural network; the method can understandably train the void detection model according to the logarithmic Mel frequency spectrogram of the target sound sample set, namely train the convolutional neural network and the cyclic neural network, fully learn the expression relation between the characteristics, and continuously adjust the parameters of the network model to enable the network to accurately identify the void state.
Further, in the embodiment of the present invention, step S4 includes the sub-steps of S401, S402, S403, S404, S405;
the substep S401 is to extract a first vector feature of the sound sample set;
specifically, a log-mel frequency spectrum diagram of a sound sample set is input into a convolutional neural network, and first vector features of a reference sample, a void sample and a non-void sample are obtained. It is understood that the convolutional neural network comprises a plurality of convolutional layers, which introduce the GLU-gated linear unit, and the output expression of the convolutional layers is:
wherein X is a characteristic vector input into the current convolution layer, W and V are different convolution kernels, b and c are offset values, and sigma is a sigmoid, relu and other linear activation functions.
The substep S402 is to extract a second vector feature based on the first vector feature;
specifically, the first vector features of the reference sample, the null sample and the non-null sample obtained in the substep S401 are input to a bidirectional Bi-LSTM recurrent neural network, and the second vector feature is obtained. It is understood that the bidirectional Bi-LSTM circulation network may be plural, but the application result shows that one bidirectional Bi-LSTM unit works best, so that the present embodiment preferably includes one bidirectional Bi-LSTM unit as the circulation network; the size of the second vector feature is determined by the number of Bi-directional Bi-LSTM loop networks.
The substep S403 is to calculate a third vector feature according to the second vector feature;
and inputting the second vector features of the reference sample, the void sample and the non-void sample into a softmax classifier, and enabling a softmax function to act on the second vector features to obtain third vector features of the reference sample, the void sample and the non-void sample.
The substep S404 is to calculate a ternary loss value based on the third vector feature of the reference sample, the positive sample and the negative sample, and includes the following steps:
1): calculating the Euclidean distance of the third vector characteristics of the reference sample and the void sample to obtain a first distance, wherein the formula is as follows:
wherein f (a)m,iThe ith vector value of the mth reference sample, f (p)m,iIs the ith vector value of the mth positive sample, and N is the third vector feature length.
2): calculating the Euclidean distance of the third vector characteristics of the reference sample and the non-void sample to obtain a second distance, wherein the formula is as follows:
wherein, f (n)m,iIs the ith vector value of the mth non-void sample.
3): calculating a ternary loss value according to the first distance, the second distance and a preset minimum distance alpha, wherein the formula is as follows:
substep S405 ends the training of the void detection model based on the ternary loss value, specifically:
if LOSS is less than or equal to mu, finishing the training of the void detection model; wherein mu is a preset loss value;
and if LOSS is more than mu, modifying the parameters of the convolutional neural network model and the bidirectional Bi-LSTM recurrent neural network, and continuing training until LOSS is less than or equal to mu.
S5: detecting and classifying the sound samples to be detected by using the trained void detection model; and (5) after the sound sample to be detected is processed in the steps S2-S3, inputting the sound sample to be detected to the trained void detection model, completing feature comparison, and outputting the class and probability of the sample.
S6: and the display screen stores and displays the detection result of the sample, and simultaneously carries out voice broadcast reminding in a linkage manner.
It is noted that, in this document, relational terms such as "first" and "second," and the like, may be used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Also, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising an … …" does not exclude the presence of other identical elements in a process, method, article, or apparatus that comprises the element.
The foregoing are merely exemplary embodiments of the present invention, which enable those skilled in the art to understand or practice the present invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other embodiments without departing from the spirit or scope of the invention. Thus, the present invention is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.
Claims (10)
1. The utility model provides a concrete road surface intelligent detection device that comes to nothing which characterized in that: the device comprises a support plate (1), a computer control system (2), a display screen (3) and an excitation device (5); a through groove is formed in the middle of the supporting plate (1), two sides of the through groove are connected with the excitation device (5), the top of one side of the supporting plate (1) is connected with the computer control system (2), and the bottom of the supporting plate (1) is provided with a roller (11); the top of the computer control system (2) is connected with a display screen (3); a voice broadcasting device (4) is installed at the top of the display screen (3);
the excitation device (5) comprises a rotating shaft (51), a motor (52), an elastic steel bar and a steel ball (53); the middle parts of two sides of the through groove are hinged with two ends of a rotating shaft (51), and one end of the rotating shaft (51) is connected with a motor (52) arranged on the outer side of the supporting plate (1); the rotating shafts (51) are uniformly distributed with rotating discs; elastic steel bars are uniformly distributed on the outer ring of the turntable, the elastic steel bars are connected with steel balls (53), and the steel balls (53) rotate to the lowest end of the turntable and are lower than the roller (11);
the both sides that lead to the groove in backup pad (1) middle part upwards extend, install bracing piece (6) between the extension, bracing piece (6) equipartition microphone (7), and the quantity of microphone (7) is no less than 4, is on a parallel with the installation of excitation device (5).
2. The intelligent detection device of concrete road surface coming to nothing of claim 1, characterized in that: the display screen (3) is internally provided with a storage module, a display module and an alarm module, the storage module, the display module and the alarm module are sequentially connected, and the alarm module is connected with the voice broadcast device (4).
3. The intelligent detection device of concrete road surface coming to nothing of claim 1, characterized in that: the microphone (7) comprises a microphone array pickup module and a microphone array signal receiving module; and the microphone array pickup module is connected with the microphone array signal receiving module.
4. The intelligent detection device of concrete road surface coming to nothing of claim 3, characterized in that: the computer control system (2) comprises an excitation device control module, a sound source direction estimation module, a noise reduction module, a feature extraction module, a void detection model training module, a void detection model detection module and a Beidou GNSS positioning module; a sound source DOA direction estimation algorithm is arranged in the sound source direction estimation module; the noise reduction module is internally provided with a self-adaptive beam forming algorithm;
the excitation device control module, the microphone array pickup module, the microphone array signal receiving module, the sound source direction estimation module, the noise reduction module, the feature extraction module, the void detection model training module, the void detection model detection module and the Beidou GNSS positioning module are connected in sequence; the excitation device control module is connected with a motor (52); the sound source direction estimation module is connected with a microphone (7); the display screen (3) is respectively connected with the air-out detection model detection module and the Beidou GNSS positioning module.
5. The detection method of the concrete pavement void intelligent detection device as claimed in claims 1-4, characterized by comprising the following steps:
s1: collecting aliasing sounds of a tested road surface hit by an exciting device under different environments by using a microphone array, converting M analog signals output by the microphone array in a pickup mode into M digital signals with preset bit rate and bit positions, carrying out end point detection, and then intercepting end point signals to obtain digital signals of multi-path aliasing sounds;
s2: transmitting the digital signals of the multiple paths of aliasing sounds of S1 to a sound source direction estimation module, performing DOA direction estimation on the digital signals of the multiple paths of aliasing sounds of S1 by a sound source DOA direction estimation algorithm of the sound source direction estimation module, and performing enhancement and denoising processing on the digital signals of the multiple paths of aliasing sounds after direction estimation by combining with an adaptive beam forming algorithm of a denoising module to obtain a clear target sound sample set;
s3: obtaining logarithmic Mel frequency spectrum characteristics of the target sound sample set of S2;
s4: establishing a sound data parameter base in a void state, and training a void detection model according to the characteristics of the target sound sample set of S3, wherein the void detection model consists of a convolutional neural network and a cyclic neural network;
s5: extracting a logarithmic Mel frequency spectrum diagram of a sound sample to be detected, carrying out intelligent identification by using a trained void detection model, and judging the void state of the detected sound sample;
s6: and displaying the identification detection result through a display screen, storing and broadcasting the voice.
6. The detection method of the intelligent detection device for concrete pavement void according to claim 5, characterized in that: the sound source DOA direction estimation algorithm is an improved MUSIC algorithm, and specifically comprises the following steps:
1) taking the first array element of the linear array as a reference, the signals received by the array are:
X(n)＝AS(t)+U(t)； (1)
2) order: y (n) ═ TNX*(n)； (2)
Wherein: t isNIs an N × N switching matrix with elements of the negative diagonal being 1 and the remainder being 0, X*(N) is the conjugate matrix of X (N), N being the number of microphones of the microphone array;
3) and (3) solving a covariance matrix of Y (n) according to (1) and (2): rYY＝E[Y(n)YH(n)]＝ARSSAH+δ2I； (3)
4) Calculating the total covariance
5) According to the corrected total covariance
Carrying out direction estimation according to a traditional MUSIC algorithm to obtain direction information of target sound and interference;
the self-adaptive beam forming algorithm is a GSC beam forming algorithm, and the optimal weighting vector of the target signal direction is calculated according to the direction information of the target sound source and the interference sound source obtained by the sound source DOA direction estimation algorithm, so that the clear digital signal of the target direction is obtained finally.
7. The detection method of the intelligent detection device for concrete pavement void according to claim 5, characterized in that: the sound samples comprise void samples, non-void samples and reference samples; the sound data parameter stock of the void state is stored with various void state sound characteristics, and the optimal characteristic parameters are obtained through the void detection model training test.
8. The detection method of the intelligent detection device for concrete pavement void according to claim 5, characterized in that: the convolution layer of the convolution neural network introduces a GLU gate control linear unit as an activation function, and the convolution neural network is a bidirectional Bi-LSTM network; the method for detecting the target sound sample set characteristic training void comprises the following steps:
1) building the convolution-cycle network model;
2) inputting the logarithmic Mel frequency spectrum characteristics of the 3/4 sound sample set into the convolutional neural network to obtain the data of characteristic prediction first vectors of a void sample, a non-void sample and a reference sample;
3) inputting first vector data of a void sample, a non-void sample and a reference sample into the bidirectional Bi-LSTM recurrent neural network respectively to obtain data of a feature prediction second vector;
4) inputting second vector data of the void sample, the non-void sample and the reference sample into a softmax classifier respectively to obtain feature prediction third vector data;
5) and calculating a triple loss value according to the third vector data of the void sample, the non-void sample and the reference sample, and optimizing the void detection model according to the triple loss value.
6) And establishing a void detection template, and substituting the residual 1/4 logarithmic Mel frequency spectrogram into the void identification model to detect the void accuracy.
9. The detection method of the intelligent detection device for concrete pavement void according to claim 5, characterized in that: calculating the triplet loss according to the third vector data of the void sample, the non-void sample and the reference sample, wherein the method comprises the following steps:
solving the Euclidean distance between the third vector data of the void sample and the reference sample to obtain a first distance;
solving the Euclidean distance between the third vector data of the non-void sample and the reference sample to obtain a second distance;
calculating a difference between the first distance and the second distance; and determining a triple loss value according to the difference value and a preset minimum distance.
10. The detection method of the intelligent detection device for concrete pavement void according to claim 8, characterized in that: optimizing said void detection model based on said triplet loss values, comprising the steps of:
if the ternary loss value is smaller than a preset loss threshold value, ending the training to obtain a final void detection model;
and if the ternary loss value is greater than the preset loss threshold value, adjusting the model parameters, and continuing training until the ternary loss value is less than the preset loss threshold value.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011463253.4A CN112557510A (en) | 2020-12-11 | 2020-12-11 | Concrete pavement void intelligent detection device and detection method thereof |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011463253.4A CN112557510A (en) | 2020-12-11 | 2020-12-11 | Concrete pavement void intelligent detection device and detection method thereof |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN112557510A true CN112557510A (en) | 2021-03-26 |
Family
ID=75062946
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011463253.4A Pending CN112557510A (en) | 2020-12-11 | 2020-12-11 | Concrete pavement void intelligent detection device and detection method thereof |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112557510A (en) |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113095258A (en) * | 2021-04-20 | 2021-07-09 | 深圳力维智联技术有限公司 | Directional signal extraction method, system, device and storage medium |
| CN113138033A (en) * | 2021-06-01 | 2021-07-20 | 大连理工大学 | Deep learning-based quantitative evaluation method for cavity of inner surface layer of concrete filled steel tube |
| CN113219063A (en) * | 2021-04-12 | 2021-08-06 | 北京航空航天大学 | Concrete structure internal defect imaging system and method based on sound vibration signal deep learning |
| CN113375063A (en) * | 2021-06-07 | 2021-09-10 | 国家石油天然气管网集团有限公司西气东输分公司 | Intelligent monitoring method and system for natural gas pipeline leakage |
| CN113658608A (en) * | 2021-08-27 | 2021-11-16 | 中关村科学城城市大脑股份有限公司 | Method and device for monitoring surface water, storage medium and electronic equipment |
| CN113763986A (en) * | 2021-09-07 | 2021-12-07 | 山东大学 | Air conditioner indoor unit abnormal sound detection method based on sound classification model |
| CN113988142A (en) * | 2021-12-27 | 2022-01-28 | 中南大学 | Tunnel lining cavity acoustic identification method based on convolutional neural network |
| WO2024145052A1 (en) * | 2022-12-30 | 2024-07-04 | Baker Hughes Oilfield Operations Llc | Characterization system and method for cement evaluation through-tubing |
Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20030164044A1 (en) * | 2001-12-17 | 2003-09-04 | Gayle James Edward | Ultraphonics array system |
| CN201306192Y (en) * | 2008-11-04 | 2009-09-09 | 刘廷国 | Tunnel lining hollow detection device |
| US20130289896A1 (en) * | 2011-01-20 | 2013-10-31 | Northeastern University | Real-time pavement profile sensing system using air-coupled surface wave |
| CN103413113A (en) * | 2013-01-15 | 2013-11-27 | 上海大学 | Intelligent emotional interaction method for service robot |
| CN104091593A (en) * | 2014-04-29 | 2014-10-08 | 苏州大学 | Speech endpoint detection algorithm adopting perceptual speech spectrum structure boundary parameters |
| CN204925039U (en) * | 2015-08-31 | 2015-12-30 | 吉林大学 | Graphite electrode nondestructive test device |
| CN109558512A (en) * | 2019-01-24 | 2019-04-02 | 广州荔支网络技术有限公司 | A kind of personalized recommendation method based on audio, device and mobile terminal |
| CN210037720U (en) * | 2019-02-26 | 2020-02-07 | 王利学 | Device for detecting cement concrete pavement and void of highway bridge |
| CN111551326A (en) * | 2020-05-28 | 2020-08-18 | 广西交科集团有限公司 | Displacement monitoring method for settlement foundation layered casting concrete beam |
-
2020
- 2020-12-11 CN CN202011463253.4A patent/CN112557510A/en active Pending
Patent Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20030164044A1 (en) * | 2001-12-17 | 2003-09-04 | Gayle James Edward | Ultraphonics array system |
| CN201306192Y (en) * | 2008-11-04 | 2009-09-09 | 刘廷国 | Tunnel lining hollow detection device |
| US20130289896A1 (en) * | 2011-01-20 | 2013-10-31 | Northeastern University | Real-time pavement profile sensing system using air-coupled surface wave |
| CN103413113A (en) * | 2013-01-15 | 2013-11-27 | 上海大学 | Intelligent emotional interaction method for service robot |
| CN104091593A (en) * | 2014-04-29 | 2014-10-08 | 苏州大学 | Speech endpoint detection algorithm adopting perceptual speech spectrum structure boundary parameters |
| CN204925039U (en) * | 2015-08-31 | 2015-12-30 | 吉林大学 | Graphite electrode nondestructive test device |
| CN109558512A (en) * | 2019-01-24 | 2019-04-02 | 广州荔支网络技术有限公司 | A kind of personalized recommendation method based on audio, device and mobile terminal |
| CN210037720U (en) * | 2019-02-26 | 2020-02-07 | 王利学 | Device for detecting cement concrete pavement and void of highway bridge |
| CN111551326A (en) * | 2020-05-28 | 2020-08-18 | 广西交科集团有限公司 | Displacement monitoring method for settlement foundation layered casting concrete beam |
Non-Patent Citations (6)
| Title |
|---|
| 刘华平等: "一种改进的自适应子带谱熵语音端点检测方法", 《系统仿真学报》, no. 5, pages 1366 - 1371 * |
| 吴云飞等: "基于自适应子带谱熵和短时平均幅度的心音端点检测", 《计算机与数字工程》, vol. 46, no. 9, pages 1740 - 1743 * |
| 孙承爱等: "基于GLU-CNN和Attention-BiLSTM的神经网络情感倾向性分析", vol. 40, no. 7, pages 62 - 66 * |
| 李晓霞: "基于深度学习的音频信号砼路面脱空分类研究及应用", no. 1, pages 140 - 525 * |
| 赵拥军, 国防工业出版社 * |
| 郑舒予等: "基于改进MUSIC算法的散射中心参数提取及RCS重构", vol. 42, no. 1, pages 76 - 82 * |
Cited By (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113219063A (en) * | 2021-04-12 | 2021-08-06 | 北京航空航天大学 | Concrete structure internal defect imaging system and method based on sound vibration signal deep learning |
| CN113219063B (en) * | 2021-04-12 | 2022-11-29 | 北京航空航天大学 | Concrete structure internal defect imaging system and method based on sound vibration signal deep learning |
| CN113095258A (en) * | 2021-04-20 | 2021-07-09 | 深圳力维智联技术有限公司 | Directional signal extraction method, system, device and storage medium |
| CN113138033A (en) * | 2021-06-01 | 2021-07-20 | 大连理工大学 | Deep learning-based quantitative evaluation method for cavity of inner surface layer of concrete filled steel tube |
| CN113138033B (en) * | 2021-06-01 | 2022-03-25 | 大连理工大学 | Deep learning-based quantitative evaluation method for cavity of inner surface layer of concrete filled steel tube |
| CN113375063A (en) * | 2021-06-07 | 2021-09-10 | 国家石油天然气管网集团有限公司西气东输分公司 | Intelligent monitoring method and system for natural gas pipeline leakage |
| CN113658608A (en) * | 2021-08-27 | 2021-11-16 | 中关村科学城城市大脑股份有限公司 | Method and device for monitoring surface water, storage medium and electronic equipment |
| CN113658608B (en) * | 2021-08-27 | 2022-03-25 | 中关村科学城城市大脑股份有限公司 | Method and device for monitoring surface water, storage medium and electronic equipment |
| CN113763986A (en) * | 2021-09-07 | 2021-12-07 | 山东大学 | Air conditioner indoor unit abnormal sound detection method based on sound classification model |
| CN113763986B (en) * | 2021-09-07 | 2024-02-02 | 山东大学 | Abnormal sound detection method for air conditioner indoor unit based on sound classification model |
| CN113988142A (en) * | 2021-12-27 | 2022-01-28 | 中南大学 | Tunnel lining cavity acoustic identification method based on convolutional neural network |
| WO2024145052A1 (en) * | 2022-12-30 | 2024-07-04 | Baker Hughes Oilfield Operations Llc | Characterization system and method for cement evaluation through-tubing |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112557510A (en) | Concrete pavement void intelligent detection device and detection method thereof | |
| CN110398647B (en) | Transformer state monitoring method | |
| CN102027338B (en) | Signal judgment method, signal judgment apparatus, and signal judgment system | |
| CN111048114A (en) | Equipment and method for detecting abnormal sound of equipment | |
| CN110412467B (en) | Lithium battery fault data screening method constrained by normalized mutual information criterion | |
| CN109505741B (en) | Wind driven generator damaged blade detection method and device based on rectangular microphone array | |
| CN108648764B (en) | Rainfall measurement system based on rainwater knocking sound identification and measurement method thereof | |
| CN103546853A (en) | Speaker abnormal sound detecting method based on short-time Fourier transformation | |
| CN112727710B (en) | Wind field thunderbolt density statistical method and system based on audio signals | |
| JP2010011433A (en) | Sound source separation and display method, and system thereof | |
| CN113029327B (en) | Tunnel fan embedded foundation damage identification method based on metric attention convolutional neural network | |
| CN111830470B (en) | Combined calibration method and device, target object detection method, system and device | |
| CN111916089B (en) | Hail detection method and device based on acoustic signal characteristic analysis | |
| CN112331231A (en) | Broiler feed intake detection system based on audio technology | |
| CN112394324A (en) | Microphone array-based remote sound source positioning method and system | |
| CN109997186A (en) | A kind of device and method for acoustic environment of classifying | |
| CN110263876A (en) | A kind of patter of rain signal de-noising processing method and system | |
| CN113990303A (en) | Environmental sound identification method based on multi-resolution cavity depth separable convolution network | |
| CN117607598A (en) | Transformer fault detection method and system based on voiceprint characteristics | |
| CN104200815B (en) | A kind of audio-frequency noise real-time detection method based on correlation analysis | |
| CN117116293A (en) | Machine equipment fault diagnosis system in complex sound field environment | |
| CN108270495B (en) | Background noise extraction method and system | |
| CN116337210A (en) | Noise evaluation system for building construction operation | |
| CN111968671B (en) | Low-altitude sound target comprehensive identification method and device based on multidimensional feature space | |
| CN116124716A (en) | Water quality monitoring method and device based on proximity hyperspectral |
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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20210326 |