CN113269210A - Method for detecting floating frequency of liquid level meter pointer based on image processing - Google Patents

Method for detecting floating frequency of liquid level meter pointer based on image processing Download PDF

Info

Publication number
CN113269210A
CN113269210A CN202110808501.2A CN202110808501A CN113269210A CN 113269210 A CN113269210 A CN 113269210A CN 202110808501 A CN202110808501 A CN 202110808501A CN 113269210 A CN113269210 A CN 113269210A
Authority
CN
China
Prior art keywords
liquid level
frequency
time
neural network
image
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.)
Granted
Application number
CN202110808501.2A
Other languages
Chinese (zh)
Other versions
CN113269210B (en
Inventor
刘云川
殷姣
郑光胜
郑侃
杨正川
叶明�
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Beijing Baolong Hongrui Technology Co.,Ltd.
Chongqing Hongbao Technology Co.,Ltd.
Sichuan hongbaorunye Engineering Technology Co.,Ltd.
Original Assignee
Beijing Baolong Hongrui Technology Co ltd
Chongqing Qingyun Petroleum Engineering Technology Co ltd
Sichuan Hongbaorunye Engineering Technology Co ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Beijing Baolong Hongrui Technology Co ltd, Chongqing Qingyun Petroleum Engineering Technology Co ltd, Sichuan Hongbaorunye Engineering Technology Co ltd filed Critical Beijing Baolong Hongrui Technology Co ltd
Priority to CN202110808501.2A priority Critical patent/CN113269210B/en
Publication of CN113269210A publication Critical patent/CN113269210A/en
Application granted granted Critical
Publication of CN113269210B publication Critical patent/CN113269210B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/40—Extraction of image or video features
    • G06V10/46—Descriptors for shape, contour or point-related descriptors, e.g. scale invariant feature transform [SIFT] or bags of words [BoW]; Salient regional features
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/80—Analysis of captured images to determine intrinsic or extrinsic camera parameters, i.e. camera calibration
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/30—Subject of image; Context of image processing
    • G06T2207/30204—Marker
    • G06T2207/30208—Marker matrix
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/40—Extraction of image or video features
    • G06V10/46—Descriptors for shape, contour or point-related descriptors, e.g. scale invariant feature transform [SIFT] or bags of words [BoW]; Salient regional features
    • G06V10/478—Contour-based spectral representations or scale-space representations, e.g. by Fourier analysis, wavelet analysis or curvature scale-space [CSS]

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • General Engineering & Computer Science (AREA)
  • Evolutionary Computation (AREA)
  • Computing Systems (AREA)
  • Biomedical Technology (AREA)
  • General Health & Medical Sciences (AREA)
  • Computational Linguistics (AREA)
  • Biophysics (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Molecular Biology (AREA)
  • Health & Medical Sciences (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Evolutionary Biology (AREA)
  • Multimedia (AREA)
  • Measurement Of Levels Of Liquids Or Fluent Solid Materials (AREA)

Abstract

The disclosure is directed to the field of image processing and wavelet transformation, and discloses a method for detecting floating frequency of a liquid level indicator based on image processing. The method comprises the following steps: shooting a pointer of the liquid level meter and identifying the reading to form a liquid level time domain graph of the liquid level changing along with time; discrete wavelet transform the time domain graph of the liquid level to obtain a time-frequency graph of the liquid level; obtaining frequency signals based on the time-frequency diagram, quantizing the amplitudes of some frequency signals into zero, and taking the rest frequency signals as the frequency signals after denoising; and updating the time-frequency graph according to the denoised frequency signal, and performing wavelet inverse transformation on the new time-frequency graph to obtain a new time-domain graph, so as to obtain the denoised liquid level data and further detect the floating frequency of the pointer of the liquid level meter. Therefore, the pointer floating can be quickly and accurately detected and early warned through pure image processing instead of manual monitoring.

Description

Method for detecting floating frequency of liquid level meter pointer based on image processing
Technical Field
The disclosure belongs to the field of image processing and wavelet transformation, and particularly relates to a method for detecting floating frequency of a liquid level indicator based on image processing.
Background
In the oil-gas exploration process, the on-site observation of overflow is the key point of the well control work of the oil-gas exploration field, the monitoring of a liquid level meter is important work content, manual monitoring consumes manpower, and the adoption of robot identification enables the collected data to easily generate various interference noises, causes the problems of poor identification effect, even incapability of identifying and the like, and seriously influences the liquid level monitoring. How to use an image processing technology to directly detect the liquid level according to the liquid level meter image under the condition of overcoming the related interference noise becomes an urgent problem to be solved.
The above information disclosed in this background section is only for enhancement of understanding of the background of the disclosure and therefore it may contain information that does not form the prior art that is already known in this country to a person of ordinary skill in the art.
Disclosure of Invention
Aiming at the problems in the prior art, the method for detecting the floating frequency of the pointer of the liquid level meter based on image processing and wavelet transformation is provided in the disclosure, and liquid level data collected in practical application comprises high-frequency mechanical vibration of the environment besides low-frequency liquid level floating change. The wavelet analysis filtering and floating frequency detection method can realize rapid and accurate prediction and early warning of abnormal data.
The purpose of the disclosure is realized by the following technical scheme:
the method for detecting the floating frequency of the liquid level indicator based on image processing and wavelet transformation comprises the following steps:
s100: shooting a pointer of the liquid level meter by using a camera in the image liquid level acquisition module to obtain a liquid level meter image containing the pointer, and identifying a reading in the liquid level meter image through a deep neural network;
s200: forming a first time domain graph of liquid level changing along with time according to liquid level meter images and readings obtained at different moments;
s300: performing discrete wavelet transform on the first time domain graph to obtain a first time frequency graph, and obtaining a first frequency signal containing amplitude of time and frequency based on the first time frequency graph;
s400: quantizing the amplitude of the first type frequency signal with the duration time lower than a first threshold value into zero, and quantizing the amplitude of the second type frequency signal with the amplitude higher than a second threshold value into zero, so as to realize the denoising of the first frequency signal, and taking the rest frequency signals as denoised second frequency signals;
s500: and updating the first time-frequency graph according to the second frequency signal to obtain a denoised second time-frequency graph, performing wavelet inverse transformation on the second time-frequency graph to obtain a second time-frequency graph so as to obtain denoised liquid level data, and further detecting the floating frequency of the pointer of the liquid level meter when corresponding liquid level data in a period of time is detected according to the denoised liquid level data and the denoised second frequency signal.
Preferably, the first and second liquid crystal materials are,
in step S100, the identifying the reading in the liquid level meter image through the deep neural network includes the following steps:
s101: manually marking the position of an interest point of each acquired original liquid level meter image for training to obtain each marked image;
s102: and taking each marked image as input to construct an auxiliary neural network to assist the deep neural network in learning and training the liquid level meter image.
Preferably, the first and second liquid crystal materials are,
step S102 is to construct an auxiliary neural network to assist the deep neural network in learning and training the liquid level gauge image, specifically by:
s1021: taking each marked image as input, selecting a proper middle layer from the deep neural network, and obtaining the output of the middle layer;
s1022: establishing an auxiliary neural network formed by convolution functions;
s1023: inputting the output of the middle layer and the corresponding attitude estimation matrix of each original liquid level meter image into an auxiliary neural network;
s1024: and combining the outputs of the auxiliary neural network and the deep neural network, and jointly inputting the outputs into a loss function of the deep neural network to assist the deep neural network in learning and training the liquid level meter image.
Preferably, the first and second liquid crystal materials are,
the attitude estimation matrix in step S1023 is obtained by:
s10231: calibrating the camera, and solving the intrinsic parameters of the camera, wherein the intrinsic parameters comprise: the image optical axis principal point, the focal lengths in the X direction and the Y direction, the tangential distortion coefficient and the radial distortion coefficient;
s10232: the attitude estimation matrix is further solved as follows:
solving an attitude estimation matrix [ R | t ] (X = M [ R | t ])/X),
wherein M is an internal parameter of the camera, X is a world coordinate system, and X is an image pixel coordinate of a known shot object; r, t are the rotation vector and the translation vector of the attitude estimation matrix, respectively.
Preferably, the first and second liquid crystal materials are,
the solving of the intrinsic parameters of the camera in step S10231 is to solve the intrinsic parameters of the camera according to the zhangzhengyou scaling method and the checkerboard.
Preferably, the first and second liquid crystal materials are,
the grid is 10cm by 10 cm.
Preferably, the first and second liquid crystal materials are,
the deep neural network selects ResNet 50.
Preferably, the first and second liquid crystal materials are,
the auxiliary neural network selects ResNet 18.
Preferably, the first and second liquid crystal materials are,
the convolution function is Conv (input, w), where input represents the input and w represents the weight.
Preferably, the first and second liquid crystal materials are,
after the step S100 is executed and before the step S200 is executed, the method further includes the following steps:
s1000: the method comprises the steps that a laser ranging sensor is used for obtaining the height of a current liquid level indicating rod of a liquid level meter; and taking the current liquid level indicator rod height as: the reference of the reading in the level gauge image is identified by the deep neural network in step S100.
In addition, the present disclosure also discloses a method for detecting floating frequency of a pointer of a liquid level meter, which comprises the following steps:
the method comprises the following steps of firstly, acquiring a time domain graph of liquid level changing along with time;
secondly, performing discrete wavelet transformation on the time domain graph to obtain a time frequency graph from the time domain graph;
thirdly, obtaining a frequency signal containing time and frequency amplitude values based on the time-frequency diagram, and denoising by adopting time filtering and frequency threshold filtering, wherein the rest frequency signal is used as a denoised frequency signal;
the fourth step, updating the time-frequency graph by using the denoised frequency signal, and performing wavelet inverse transformation on the updated time-frequency graph to obtain a denoised time-domain graph;
and fifthly, obtaining denoised liquid level data by using the denoised time domain diagram, and further detecting the floating frequency of a liquid level meter pointer in a period of time when corresponding liquid level data is detected according to the denoised liquid level data and the denoised frequency signal.
Preferably, in the first step, the image liquid level acquisition module shoots the liquid level meter, and the image recognition obtains a reading.
Preferably, in the first step, the laser ranging sensor obtains the current level indicating rod height of the liquid level meter to identify the reading.
Preferably, for a discrete wavelet transform:
the basis functions of the continuous wavelet are:
Figure 520834DEST_PATH_IMAGE001
，
wherein, FbAs a bandwidth parameter, FcIs the center frequency, i is the imaginary unit, t is the time unit,
and performing extension and translation on the basis function to obtain a family function:
Figure 120890DEST_PATH_IMAGE003
，
wherein a is a scale, namely a telescopic parameter, and b is a translation parameter;
the method comprises the following steps of discretizing a scale parameter a and a translation parameter b of a definition formula of continuous wavelet transformation, wherein a discretized wavelet sequence is as follows:
Figure 111980DEST_PATH_IMAGE004
，
the discretization formula of the scale and translation parameters a and b in continuous wavelet change is as follows:
Figure 607552DEST_PATH_IMAGE005
，
a0is the extension step size, where the power of j represents the scale reduction multiple to a; when j is not equal to 0, a is changed from a0 j-1Become a0 jWhen a is enlarged by a0Multiple, i.e. the centre frequency of the wavelet is reduced by a0Multiple, bandwidth is also reduced by a0Doubling; k represents the uniform sampling rate for b; z represents an integer set;
the wavelet coefficients of different positions and different scales are solved through discrete wavelet transform, and the formula is as follows:
Figure 619371DEST_PATH_IMAGE006
，
wherein the content of the first and second substances,
Figure 849495DEST_PATH_IMAGE007
is composed of
Figure 660587DEST_PATH_IMAGE008
A complex conjugate function of; dt is the data sampling time; cj,kFor the calculation of the resulting wavelet coefficients for each frequency and scale, f (t) is a function of the original signal.
When in reconstruction, each wavelet function is directly multiplied and accumulated by the coefficient, and the formula is as follows:
Figure 495819DEST_PATH_IMAGE009
。
it can be understood that when the time domain graph is subjected to discrete wavelet transform, f (t) is time domain data;
when the time-frequency diagram is subjected to wavelet inverse transformation to obtain the liquid level data after denoising, the liquid level data obtained through inverse transformation is: f (t) obtained by performing inverse transformation on the time-frequency diagram.
In the method, if the liquid level value of the time domain exceeds the threshold value, an alarm signal is sent out, and if the frequency of the effective frequency of the frequency domain is abnormally floated and exceeds the threshold value, a higher-level emergency signal is sent out.
Advantageous effects
According to the method, the time domain graph and the frequency domain graph are obtained through wavelet processing, after the frequency signals with continuous time lower than a preset time range are quantized to be zero, the amplitude of the main frequency signal higher than a high-frequency threshold value is quantized to be zero, the rest frequency signals are used as the liquid level signal frequency, interference caused by noise such as high-frequency vibration is avoided, and the abnormal data can be quickly and accurately predicted and early warned.
The foregoing description is only an overview of the technical solutions of the present disclosure, and in order to make the technical solutions of the present disclosure more clearly understood, the technical solutions of the present disclosure are implemented to the extent that those skilled in the art can implement the technical solutions according to the description, and the above and other objects, features, and advantages of the present disclosure can be more clearly understood, the following description is given by way of example of the specific embodiments of the present disclosure.
Drawings
Various additional advantages and benefits of the present disclosure will become apparent to those of ordinary skill in the art upon reading the following detailed description of the preferred embodiments. The drawings are only for purposes of illustrating the preferred embodiments and are not to be construed as limiting the disclosure. It is apparent that the drawings described below are only some embodiments of the disclosure, and that other drawings may be derived from those drawings by one of ordinary skill in the art without inventive effort. Also, like parts are designated by like reference numerals throughout the drawings.
In the drawings:
FIG. 1(a) is original hydraulic data, which illustrates a liquid level data simulation diagram for wavelet transform in an embodiment of the present disclosure, the original data being formed by 1.0-20.0 s of 10Hz sine + 0.9Hz +17.5Hz sine +0.02 white noise;
FIG. 1(b) is a wavelet time-frequency diagram illustrating the time-frequency diagram of FIG. 1(a) after wavelet transform, with the abscissa being time, the ordinate being frequency, and the pixel values being wavelet coefficient values at a certain point in time for the frequency;
FIGS. 2(a) and 2(b) are schematic diagrams of reconstructed time domain data of FIGS. 1(a) and 1(b) subjected to data filtering, wherein FIG. 2(a) illustrates a graph of raw level readings over time; FIG. 2(b) illustrates the denoised liquid level data obtained by wavelet reconstruction;
fig. 3 is a schematic flow chart in one embodiment of the present disclosure.
The present disclosure is further explained below with reference to the drawings and examples.
Detailed Description
Specific embodiments of the present disclosure will be described in more detail below with reference to fig. 1(a) to 3. While specific embodiments of the disclosure are shown in the drawings, it should be understood that the disclosure may be embodied in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.
It should be noted that certain terms are used throughout the description and claims to refer to particular components. As one skilled in the art will appreciate, various names may be used to refer to a component. This specification and claims do not intend to distinguish between components that differ in name but not function. In the following description and in the claims, the terms "include" and "comprise" are used in an open-ended fashion, and thus should be interpreted to mean "include, but not limited to. The description which follows is a preferred embodiment of the disclosure, but is made for the purpose of illustrating the general principles of the disclosure and not for the purpose of limiting the scope of the disclosure. The scope of the present disclosure is to be determined by the terms of the appended claims.
To facilitate an understanding of the embodiments of the present disclosure, the following detailed description is to be considered in conjunction with the accompanying drawings, and the drawings are not to be construed as limiting the embodiments of the present disclosure.
In one embodiment, the present disclosure discloses a method for detecting a floating frequency of a liquid level indicator pointer based on image processing and wavelet transform, the method comprising the steps of:
s100: shooting a pointer of the liquid level meter by using a camera in the image liquid level acquisition module to obtain a liquid level meter image containing the pointer, and identifying a reading in the liquid level meter image through a deep neural network;
s200: forming a first time domain graph of liquid level changing along with time according to liquid level meter images and readings obtained at different moments;
s300: performing discrete wavelet transform on the first time domain graph to obtain a first time frequency graph, and obtaining a first frequency signal containing amplitude of time and frequency based on the first time frequency graph;
s400: quantizing the amplitude of the first type frequency signal with the duration time lower than a first threshold value into zero, and quantizing the amplitude of the second type frequency signal with the amplitude higher than a second threshold value into zero, so as to realize the denoising of the first frequency signal, and taking the rest frequency signals as denoised second frequency signals;
s500: and updating the first time-frequency graph according to the second frequency signal to obtain a denoised second time-frequency graph, performing wavelet inverse transformation on the second time-frequency graph to obtain a second time-frequency graph so as to obtain denoised liquid level data, and further detecting the floating frequency of the pointer of the liquid level meter when corresponding liquid level data in a period of time is detected according to the denoised liquid level data and the denoised second frequency signal.
In the embodiment, after the frequency signal with the continuous time lower than the preset time range is quantized to be zero, the amplitude of the main frequency signal higher than the high-frequency threshold value is quantized to be zero, the rest frequency signal is used as the denoised frequency signal, and meanwhile, the floating frequency of the liquid level meter pointer is detected by combining the denoised liquid level data. The method avoids the interference caused by noise such as obvious external disturbance, high-frequency vibration and the like, and realizes the rapid and accurate prediction and early warning of abnormal data. It can be understood that, according to the corresponding relationship between the time domain diagram and the liquid level data, the liquid level data after denoising is obtained exactly according to the second time domain diagram, i.e. according to the new time domain diagram after denoising.
In another embodiment of the present invention, the substrate is,
in step S100, the identifying the reading in the liquid level meter image through the deep neural network includes the following steps:
s101: manually marking the position of an interest point of each acquired original liquid level meter image for training to obtain each marked image;
s102: and taking each marked image as input to construct an auxiliary neural network to assist the deep neural network in learning and training the liquid level meter image.
With this embodiment, it is considered that the final result of the image recognition is to output the coordinates of the respective interest points on the image. However, if the floating frequency of the pointer can be detected as per the previous embodiment, the deep neural network learns the image and directly outputs two-dimensional coordinates for optimization learning without adopting the optimization manner of the present embodiment, which obviously means that the previous embodiment is a very non-linear process, and the loss function for optimization is less constrained to the weights in the neural network during optimization learning. In this regard, those skilled in the art will appreciate. In order to solve this problem, in this embodiment, an intermediate state is finally constructed by constructing an auxiliary neural network according to the collected images and the positions of the artificially labeled interest points, so as to assist the training and learning of the deep neural network. That is, the auxiliary neural network is used to assist the training and learning of the deep neural network, compared to the previous embodiment, which is another technical contribution of the present disclosure.
In another embodiment of the present invention, the substrate is,
step S102 is to construct an auxiliary neural network to assist the deep neural network in learning and training the liquid level gauge image, specifically by:
s1021: taking each marked image as input, selecting a proper middle layer from the deep neural network, and obtaining the output of the middle layer;
s1022: establishing an auxiliary neural network formed by convolution functions;
s1023: inputting the output of the middle layer and the corresponding attitude estimation matrix of each original liquid level meter image into an auxiliary neural network;
s1024: and combining the outputs of the auxiliary neural network and the deep neural network, and jointly inputting the outputs into a loss function of the deep neural network to assist the deep neural network in learning and training the liquid level meter image.
It is to be appreciated that the present embodiment, in connection with the previous embodiment, is a further exemplary illustration of the technical contributions related to the ancillary neural network aspects of the present disclosure.
In another embodiment of the present invention, the substrate is,
after the step S100 is executed and before the step S200 is executed, the method further includes the following steps:
s1000: the method comprises the steps that a laser ranging sensor is used for obtaining the height of a current liquid level indicating rod of a liquid level meter; and taking the current liquid level indicator rod height as: the reference of the reading in the level gauge image is identified by the deep neural network in step S100.
It can be understood that since S200: forming a time domain graph of the liquid level changing along with time according to the liquid level meter images and the readings obtained at different moments; this means, this embodiment will be with the current liquid level of the liquid level table that laser rangefinder sensor obtained and instruct the pole height, has equallyd to an extra basis of "reading", and this disclosure can make the neural network of depth have clear external reference more, and this reference is the value of the higher precision of laser rangefinder. This facilitates more accurate readings by the deep neural network of the present disclosure.
It should be noted that the laser ranging can measure more than 100 times per second, and one or more laser ranging sensors can be used to obtain one or more references as needed, or even an average value reference can be formed by averaging a plurality of reference values.
Furthermore, in another embodiment, the present disclosure discloses a method for detecting a level gauge pointer floating frequency based on image processing and wavelet transform, the method comprising the steps of:
the method comprises the following steps of firstly, acquiring a time domain graph of liquid level changing along with time;
secondly, performing discrete wavelet transformation on the time domain graph to obtain a time frequency graph from the time domain graph;
thirdly, obtaining a frequency signal containing time and frequency amplitude values based on the time-frequency diagram, and denoising by adopting time filtering and frequency threshold filtering, wherein the rest frequency signal is used as a denoised frequency signal;
the fourth step, updating the time-frequency graph by using the denoised frequency signal, and performing wavelet inverse transformation on the updated time-frequency graph to obtain a denoised time-domain graph;
and fifthly, obtaining denoised liquid level data by using the denoised time domain diagram, and further detecting the floating frequency of a liquid level meter pointer in a period of time when corresponding liquid level data is detected according to the denoised liquid level data and the denoised frequency signal.
With respect to the above embodiments, it reveals the key technical idea of the present disclosure, and widens the scope of the present disclosure compared with the first embodiment.
Among these, temporal filtering is due to: the inventor finds that when the drilling platform works continuously, the drilling platform generates a discontinuous certain frequency signal in a low-frequency range, and the discontinuous frequency signal is quantized to be zero;
the frequency filtering is due to: the inventor finds that the vibration of a power system is one of the long-term continuous noises of a hydraulic pressure meter and a liquid level meter when the drilling platform works. For example, the drill pipe rotates at a speed of tens of revolutions per minute, and the inventor finds that the frequency of the maximum amplitude of the power system during the drilling operation is 17.5Hz, so that 40% of offset range can be reserved during frequency filtering, the high-frequency threshold value is set to be 10.5Hz, the amplitude of the main frequency signal higher than the high-frequency threshold value is quantized to be zero, and the rest frequency signal is taken as the liquid level signal frequency.
Referring to fig. 3, in another embodiment, the intermediate layer output is the output of C5:7 × 2048.
For the related embodiments described above, the present disclosure effectively reduces the fitting difficulty during the training of the relevant model, and simultaneously improves the robustness of the model. In the same test set, after the images are used as a training set and the training and optimizing methods are adopted, the map @0.5 precision of the model is 2.76% higher than that of the model which is not adopted. It is further noted that, when the attitude estimation matrix is solved by the PNP based on the rannsac algorithm, even if the average error of the PNP solution is found to be about 5% in the actual scene verification, the difficulty of fitting during the training of the relevant model is effectively reduced and the robustness of the model is improved, so that the subsequent actual scene verification is not affected by the error of the attitude estimation matrix.
In the preferred embodiment of the method, if the liquid level value in the time domain exceeds the threshold value, an alarm signal is sent out, and if the frequency fluctuation abnormality of the effective frequency in the frequency domain exceeds the threshold value, an emergency signal is sent out.
It should be noted that the above embodiments are key to the core concept and technical contribution of the present disclosure.
The following examples are more specific alternatives, are conventional in the art and are not critical to the present disclosure, nor are they intended to represent novel technical contributions to the following examples.
In another embodiment of the present invention, the substrate is,
the attitude estimation matrix in step S1023 is obtained by:
s10231: calibrating the camera, and solving the intrinsic parameters of the camera, wherein the intrinsic parameters comprise: the image optical axis principal point, the focal lengths in the X direction and the Y direction, the tangential distortion coefficient and the radial distortion coefficient;
s10232: the attitude estimation matrix is further solved as follows:
solving an attitude estimation matrix [ R | t ] (X = M [ R | t ])/X),
wherein M is an internal parameter of the camera, X is a world coordinate system, and X is an image pixel coordinate of a known shot object; r, t are the rotation vector and the translation vector of the attitude estimation matrix, respectively.
In another embodiment, the solving of the intrinsic parameters of the camera in step S10231 is based on the Zhang friend calibration method and the checkerboard.
Further, in another embodiment, the attitude estimation matrix [ R | t ] is solved by PNP based on the Ranssac algorithm.
In another embodiment of the present invention, the substrate is,
step S1021 includes:
and solving the internal parameters of the camera according to the Zhangyingyou calibration method and the checkerboard.
In short, the Zhangyingyou scaling method is taken as a means in the prior art, namely: the internal parameters of the camera are solved by shooting the pixel coordinates of checkerboard with known sizes in different directions and different positions in an image coordinate system.
In another embodiment of the present invention, the substrate is,
the grid is 10cm by 10 cm.
In another embodiment of the present invention, the substrate is,
the deep neural network selects ResNet 50.
For this example, the process is typically carried out by Python. This is a routine choice in the art.
In another embodiment of the present invention, the substrate is,
the auxiliary neural network selects ResNet 18.
It will be appreciated that this is again a routine choice in the art.
In another embodiment of the present invention, the substrate is,
the convolution function is Conv (input, w), where input represents the input and w represents the weight.
It will be appreciated that this is again a routine choice in the art.
In another embodiment, the loss function is selected as a mean square error function.
In another embodiment of the present invention, the substrate is,
in the discrete wavelet transform,
the basis functions of the continuous wavelet are:
Figure 599910DEST_PATH_IMAGE001
，
wherein, FbAs a bandwidth parameter, FcIs the center frequency, i is the imaginary unit, t is the time unit,
and performing extension and translation on the basis function to obtain a family function:
Figure 509278DEST_PATH_IMAGE003
，
wherein a is a scale, namely a telescopic parameter, and b is a translation parameter;
the method comprises the following steps of discretizing a scale parameter a and a translation parameter b of a definition formula of continuous wavelet transformation, wherein a discretized wavelet sequence is as follows:
Figure 949748DEST_PATH_IMAGE004
，
the discretization formula of the scale and translation parameters a and b in continuous wavelet change is as follows:
Figure 241053DEST_PATH_IMAGE005
，
a0is the extension step size, where the power of j represents the scale reduction multiple to a; when j is not equal to 0, a is changed from a0 j-1Become a0 jWhen a is enlarged by a0Multiple, i.e. the centre frequency of the wavelet is reduced by a0Multiple, bandwidth is also reduced by a0Doubling; k represents the uniform sampling rate for b; z represents an integer set;
the wavelet coefficients of different positions and different scales are solved through discrete wavelet transform, and the formula is as follows:
Figure 773665DEST_PATH_IMAGE006
，
wherein the content of the first and second substances,
Figure 425095DEST_PATH_IMAGE007
is composed of
Figure 234919DEST_PATH_IMAGE008
A complex conjugate function of; dt is the data sampling time; cj,kFor the calculation of the resulting wavelet coefficients for each frequency and scale, f (t) is a function of the original signal.
When in reconstruction, each wavelet function is directly multiplied and accumulated by the coefficient, and the formula is as follows:
Figure 759441DEST_PATH_IMAGE009
。
it can be understood that when the time domain graph is subjected to discrete wavelet transform, f (t) is time domain data;
when the time-frequency diagram is subjected to wavelet inverse transformation, since the inverse transformation is reconstruction, the liquid level data obtained by the inverse transformation is: f (t) obtained by performing inverse transformation on the time-frequency diagram.
To further understand the present disclosure, in one embodiment, the image level acquisition module uses a camera as a sensor input to observe and identify the level sensor reading as raw data to be transmitted to the subsequent module, and the data processing is divided into several steps of data screening, effective frequency detection, temporal filtering prediction, anomaly determination, and the like. Details include the following processing stages:
1.1. data screening
The module can be divided into: wavelet forward transform (decomposition), automatic frequency screening and wavelet inverse transform (reconstruction) to regenerate the signal without noise.
(1) Wavelet forward transform
FIG. 1(a) and FIG. 1(b) are time-frequency diagrams obtained by performing discrete wavelet transform on liquid level data. Wavelet transform is a time-frequency analysis method that approximates signals with a set of basis functions that attenuate rapidly.
Commonly used basis functions include Morlet, Meyer, Daubechies, Haar and the like, and cmor wavelets are selected according to actual conditions, and the expression form of the basis functions is as follows:
the wavelet basis functions are:
Figure 213905DEST_PATH_IMAGE001
，
wherein, FbAs a bandwidth parameter, FcIs the center frequency, i is the imaginary unit, t is the time unit,
performing extension and translation on the basis function to obtain a family function
Figure 333357DEST_PATH_IMAGE010
，
Figure 28780DEST_PATH_IMAGE003
，
Wherein a is scale, namely a telescopic parameter, b is a translation parameter, and tau is expressed by b;
the method comprises the following steps of discretizing a scale parameter a and a translation parameter b of a definition formula of continuous wavelet transformation, wherein a discretized wavelet sequence is as follows:
Figure 208089DEST_PATH_IMAGE004
，
the discretization formula of the scale and translation parameters a and b in continuous wavelet change is as follows:
Figure 906049DEST_PATH_IMAGE005
，
a0is the extension step size, where the power of j represents the scale reduction multiple to a; when j is not equal to 0, a is changed from a0 j-1Become a0 jWhen a is enlarged by a0Multiple, i.e. the centre frequency of the wavelet is reduced by a0Multiple, bandwidth is also reduced by a0Doubling; k represents the uniform sampling rate for b; z represents an integer set;
the wavelet coefficients of different positions and different scales are solved through discrete wavelet transform, and the formula is as follows:
Figure 487203DEST_PATH_IMAGE006
，
wherein the content of the first and second substances,
Figure 353528DEST_PATH_IMAGE007
is composed of
Figure 472662DEST_PATH_IMAGE008
A complex conjugate function of; dt is the data sampling time; cj,kTo calculate the resulting wavelet coefficients for each frequency and scale,
when in reconstruction, each wavelet function is directly multiplied and accumulated by the coefficient, and the formula is as follows:
Figure 754739DEST_PATH_IMAGE009
。
in the specific implementation process, the Mallat algorithm is used for realizing the rapid wavelet decomposition, as shown in FIG. 1(a), a group of hydraulic data simulation graphs are shown, and original data are formed by 1.0-20.0 s of 10Hz sine + 0.9Hz +17.5Hz positive chord +0.02 white noise. Fig. 1(b) is a time-frequency spectrogram obtained after wavelet transform is performed on original data, wherein the abscissa is time, the ordinate is frequency, the pixel value is a wavelet coefficient value of the frequency at a certain time point, the larger the coefficient is, the heavier the weight of the principal component of the frequency is, and it can be clearly seen that the principal components of the data frequency have 0.9Hz,10Hz,17.5Hz, and the starting point and the ending point of low-frequency data.
(2) Automatic frequency screening
According to the observation and practice of the inventor, if the peak point of a long period of time exists in the time-frequency diagram for a long period of time, the frequency components are considered to be noise caused by long-term environmental vibration such as mechanical vibration of a working platform, and the waveform of the frequency can be filtered out from frequency domain and time domain data, so that effective data screening is realized. After wavelet decomposition of previous signals, respectively calculating a high-frequency coefficient and a low-frequency coefficient, and carrying out threshold hard quantization on frequency signals with continuous time exceeding a threshold to be zero.
(3) Inverse wavelet transform
The high-frequency coefficient and the low-frequency coefficient which are subjected to threshold quantization are subjected to discrete wavelet reconstruction by using the formula, a synthesized de-noised signal is shown as the following figure 2(b), the high-frequency signal coefficient is set to be 0, and then original data are reconstructed, so that de-noised liquid level data can be obtained.
1.2 effective frequency detection
While the noise data is removed by the automatic frequency filtering, the effective data frequency and the start-stop time can be obtained from fig. 1 (b).
1.3 temporal Filtering prediction
Fig. 2(b) shows reconstructed time domain data after data screening, and it can be seen that the data are not completely superimposed sinusoidally.
Like random white noise put in simulation, actual data generally exist, so that the noise can be further filtered through a Kalman filtering algorithm to prevent error identification of abnormal data.
1.4 abnormality determination
And (3) combining the simultaneous analysis of the frequency domain and the time domain, sending an alarm signal when the liquid level value of the time domain exceeds a threshold value, and sending a further emergency signal if the frequency fluctuation abnormality of the effective frequency of the frequency domain exceeds the threshold value.
While the embodiments of the disclosure have been described above in connection with the drawings, the disclosure is not limited to the specific embodiments and applications described above, which are intended to be illustrative, instructive, and not restrictive. Those skilled in the art, having the benefit of this disclosure, may effect numerous modifications thereto and changes may be made without departing from the scope of the disclosure as set forth in the claims that follow.

Claims (10)

1. A method for detecting floating frequency of a liquid level indicator based on image processing and wavelet transformation is characterized by comprising the following steps:
s100: shooting a pointer of the liquid level meter by using a camera in the image liquid level acquisition module to obtain a liquid level meter image containing the pointer, and identifying a reading in the liquid level meter image through a deep neural network;
s200: forming a first time domain graph of liquid level changing along with time according to liquid level meter images and readings obtained at different moments;
s300: performing discrete wavelet transform on the first time domain graph to obtain a first time frequency graph, and obtaining a first frequency signal containing amplitude of time and frequency based on the first time frequency graph;
s400: quantizing the amplitude of the first type frequency signal with the duration time lower than a first threshold value into zero, and quantizing the amplitude of the second type frequency signal with the amplitude higher than a second threshold value into zero, so as to realize the denoising of the first frequency signal, and taking the rest frequency signals as denoised second frequency signals;
s500: and updating the first time-frequency graph according to the second frequency signal to obtain a denoised second time-frequency graph, performing wavelet inverse transformation on the second time-frequency graph to obtain a second time-frequency graph so as to obtain denoised liquid level data, and further detecting the floating frequency of the pointer of the liquid level meter when corresponding liquid level data in a period of time is detected according to the denoised liquid level data and the denoised second frequency signal.
2. The method according to claim 1, wherein the step S100 of identifying the reading in the liquid level meter image through the deep neural network comprises the steps of:
s101: manually marking the position of an interest point of each acquired original liquid level meter image for training to obtain each marked image;
s102: and taking each marked image as input to construct an auxiliary neural network to assist the deep neural network in learning and training the liquid level meter image.
3. The method according to claim 2, wherein step S102 constructs an auxiliary neural network to assist the deep neural network in learning and training the liquid level gauge image by specifically:
s1021: taking each marked image as input, selecting a proper middle layer from the deep neural network, and obtaining the output of the middle layer;
s1022: establishing an auxiliary neural network formed by convolution functions;
s1023: inputting the output of the middle layer and the corresponding attitude estimation matrix of each original liquid level meter image into an auxiliary neural network;
s1024: and combining the outputs of the auxiliary neural network and the deep neural network, and jointly inputting the outputs into a loss function of the deep neural network to assist the deep neural network in learning and training the liquid level meter image.
4. The method of claim 3, wherein the pose estimation matrix in step S1023 is obtained by:
s10231: calibrating the camera, and solving the intrinsic parameters of the camera, wherein the intrinsic parameters comprise: the image optical axis principal point, the focal lengths in the X direction and the Y direction, the tangential distortion coefficient and the radial distortion coefficient;
s10232: the attitude estimation matrix is further solved as follows:
solving an attitude estimation matrix [ R | t ] (X = M [ R | t ])/X),
wherein M is an internal parameter of the camera, X is a world coordinate system, and X is an image pixel coordinate of a known shot object; r, t are the rotation vector and the translation vector of the attitude estimation matrix, respectively.
5. The method according to claim 4, wherein the solving of the intrinsic parameters of the camera in step S10231 is based on Zhangni scaling and checkerboard.
6. The method of claim 5, wherein the checkerboard is 10cm by 10 cm.
7. The method of claim 2, wherein the deep neural network selects ResNet 50.
8. The method of claim 2, wherein the auxiliary neural network selects ResNet 18.
9. The method of claim 3, wherein the convolution function is Conv (input, w), where input represents an input and w represents a weight.
10. The method according to claim 1, wherein after the step S100 is executed and before the step S200 is executed, the method further comprises the steps of:
s1000: the method comprises the steps that a laser ranging sensor is used for obtaining the height of a current liquid level indicating rod of a liquid level meter; and taking the current liquid level indicator rod height as: the reference of the reading in the level gauge image is identified by the deep neural network in step S100.
CN202110808501.2A 2021-07-16 2021-07-16 Method for detecting floating frequency of liquid level meter pointer based on image processing Active CN113269210B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110808501.2A CN113269210B (en) 2021-07-16 2021-07-16 Method for detecting floating frequency of liquid level meter pointer based on image processing

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110808501.2A CN113269210B (en) 2021-07-16 2021-07-16 Method for detecting floating frequency of liquid level meter pointer based on image processing

Publications (2)

Publication Number Publication Date
CN113269210A true CN113269210A (en) 2021-08-17
CN113269210B CN113269210B (en) 2021-09-17

Family

ID=77236708

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110808501.2A Active CN113269210B (en) 2021-07-16 2021-07-16 Method for detecting floating frequency of liquid level meter pointer based on image processing

Country Status (1)

Country Link
CN (1) CN113269210B (en)

Citations (11)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
AU2011232765A1 (en) * 2004-07-09 2011-10-27 Touchsensor Technologies, Llc Proximity sensor for level sensing in a bilge
CN102680060A (en) * 2012-06-01 2012-09-19 江南大学 Method for automatically detecting metrological characteristics of water meter based on image recognition
US20150125080A1 (en) * 2011-11-22 2015-05-07 Mayo Foundation For Medical Education And Research Image processing device and methods for performing an s-transform
CN109490614A (en) * 2018-12-11 2019-03-19 哈尔滨工程大学 One kind being based on the adjustable digital galvanometer of STM32 range
CN110800062A (en) * 2017-10-16 2020-02-14 因美纳有限公司 Deep convolutional neural network for variant classification
CN111239158A (en) * 2020-03-13 2020-06-05 苏州鑫睿益荣信息技术有限公司 Automobile instrument panel detection system and detection method based on machine vision
CN111709435A (en) * 2020-05-18 2020-09-25 杭州电子科技大学 Countermeasure sample generation method based on discrete wavelet transform
CN111798396A (en) * 2020-07-01 2020-10-20 中通服咨询设计研究院有限公司 Multifunctional image processing method based on wavelet transformation
CN112650301A (en) * 2021-01-11 2021-04-13 四川泓宝润业工程技术有限公司 Control method for guiding unmanned aerial vehicle to accurately land
CN112906602A (en) * 2021-03-04 2021-06-04 杭州电力设备制造有限公司 Automatic identification device and identification method for electricity meter of power distribution cabinet based on image processing
CN112906592A (en) * 2021-03-03 2021-06-04 新疆爱华盈通信息技术有限公司 Passenger flow volume analysis method and system and electronic equipment

Patent Citations (12)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
AU2011232765A1 (en) * 2004-07-09 2011-10-27 Touchsensor Technologies, Llc Proximity sensor for level sensing in a bilge
US20150125080A1 (en) * 2011-11-22 2015-05-07 Mayo Foundation For Medical Education And Research Image processing device and methods for performing an s-transform
CN102680060A (en) * 2012-06-01 2012-09-19 江南大学 Method for automatically detecting metrological characteristics of water meter based on image recognition
CN110800062A (en) * 2017-10-16 2020-02-14 因美纳有限公司 Deep convolutional neural network for variant classification
CN110832596A (en) * 2017-10-16 2020-02-21 因美纳有限公司 Deep convolutional neural network training method based on deep learning
CN109490614A (en) * 2018-12-11 2019-03-19 哈尔滨工程大学 One kind being based on the adjustable digital galvanometer of STM32 range
CN111239158A (en) * 2020-03-13 2020-06-05 苏州鑫睿益荣信息技术有限公司 Automobile instrument panel detection system and detection method based on machine vision
CN111709435A (en) * 2020-05-18 2020-09-25 杭州电子科技大学 Countermeasure sample generation method based on discrete wavelet transform
CN111798396A (en) * 2020-07-01 2020-10-20 中通服咨询设计研究院有限公司 Multifunctional image processing method based on wavelet transformation
CN112650301A (en) * 2021-01-11 2021-04-13 四川泓宝润业工程技术有限公司 Control method for guiding unmanned aerial vehicle to accurately land
CN112906592A (en) * 2021-03-03 2021-06-04 新疆爱华盈通信息技术有限公司 Passenger flow volume analysis method and system and electronic equipment
CN112906602A (en) * 2021-03-04 2021-06-04 杭州电力设备制造有限公司 Automatic identification device and identification method for electricity meter of power distribution cabinet based on image processing

Non-Patent Citations (4)

* Cited by examiner, † Cited by third party
Title
CHENG DAI等: ""Intelligent Ammeter Reading Recognition Method Based on Deep Learning"", 《2019 IEEE 8TH JOINT INTERNATIONAL INFORMATION TECHNOLOGY AND ARTIFICIAL INTELLIGENCE CONFERENCE (ITAIC)》 *
GABRIEL SALOMON等: ""Deep Learning for Image-based Automatic Dial Meter Reading: Dataset and Baselines"", 《2020 INTERNATIONAL JOINT CONFERENCE ON NEURAL NETWORKS (IJCNN)》 *
彭昆福: ""基于深度学习的指针仪表读数识别方法研究"", 《中国优秀博硕士学位论文全文数据库(硕士) 信息科技辑》 *
徐浩等: ""基于迁移学习的配电网内部过电压识别方法"", 《中国电力》 *

Also Published As

Publication number Publication date
CN113269210B (en) 2021-09-17

Similar Documents

Publication Publication Date Title
CN105865424B (en) A kind of multispectral remote sensing inversion method and device based on nonlinear model
CN110361778B (en) Seismic data reconstruction method based on generation countermeasure network
Luo et al. On-line vibration analysis with fast continuous wavelet algorithm for condition monitoring of bearing
CN110363337B (en) Oil measuring method and system of oil pumping unit based on data driving
CN109737955A (en) A kind of attitude prediction method of wave compensation system
CN115169409B (en) Sliding window-based bridge structure self-vibration frequency identification and early warning method and equipment
CN115683431B (en) Stay cable force determination method, device and equipment based on linear tracking algorithm
CN105607288A (en) Intelligent glasses omnibearing vehicle part completeness detection method based on acoustic detection assistance
CN114596525A (en) Dynamic bridge form identification method based on computer vision
CN113160084A (en) Denoising method and device for quantum dot fluorescence image on surface of porous silicon biosensor
CN111859743A (en) Structure dynamic displacement estimation method and device based on acceleration response
Haris et al. Sounding out life in the deep using acoustic data from ships of opportunity
CN110716203B (en) Time-frequency analysis and tracking method of passive sonar target
CN113269210B (en) Method for detecting floating frequency of liquid level meter pointer based on image processing
CN110703205A (en) Ultrashort baseline positioning method based on adaptive unscented Kalman filtering
CN112881054B (en) Hoisting machinery fault diagnosis method and system
CN111999747B (en) Robust fault detection method for inertial navigation-satellite integrated navigation system
CN112698412A (en) Shaft frequency magnetic abnormal signal processing method based on magnetic buoy
CN116701840A (en) Method and system for optimizing and calculating cepstrum of mechanical vibration signal
CN116930192A (en) High-precision copper pipe defect detection method and system
CN116543317A (en) Noise reduction method for detecting bridge cable vibration frequency by unmanned aerial vehicle
CN114279311A (en) GNSS deformation monitoring method and system based on inertia
CN108664936A (en) A kind of diagnostic method and system based on mechanical disorder
CN114998624A (en) Image searching method and device
CN113836476A (en) Road flatness measuring method based on smart phone

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant
CP01 Change in the name or title of a patent holder

Address after: 610056 No.05, 11th floor, building 1, 99 mengzhuiwan Hengjie, Chenghua District, Chengdu City, Sichuan Province

Patentee after: Sichuan hongbaorunye Engineering Technology Co.,Ltd.

Patentee after: Chongqing Hongbao Technology Co.,Ltd.

Patentee after: Beijing Baolong Hongrui Technology Co.,Ltd.

Address before: 610056 No.05, 11th floor, building 1, 99 mengzhuiwan Hengjie, Chenghua District, Chengdu City, Sichuan Province

Patentee before: Sichuan hongbaorunye Engineering Technology Co.,Ltd.

Patentee before: Chongqing Qingyun Petroleum Engineering Technology Co.,Ltd.

Patentee before: Beijing Baolong Hongrui Technology Co.,Ltd.

CP01 Change in the name or title of a patent holder