CN110751196B - Oil-like drop attachment identification method in oil-water two-phase flow transparent pipe wall - Google Patents
Oil-like drop attachment identification method in oil-water two-phase flow transparent pipe wall Download PDFInfo
- Publication number
- CN110751196B CN110751196B CN201910966266.4A CN201910966266A CN110751196B CN 110751196 B CN110751196 B CN 110751196B CN 201910966266 A CN201910966266 A CN 201910966266A CN 110751196 B CN110751196 B CN 110751196B
- Authority
- CN
- China
- Prior art keywords
- sample
- oil
- sample set
- current
- edge
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/13—Edge detection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/22—Matching criteria, e.g. proximity measures
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
- G06F18/232—Non-hierarchical techniques
- G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
- G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- 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
- G06V10/56—Extraction of image or video features relating to colour
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Probability & Statistics with Applications (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
The invention belongs to the technical field of petroleum engineering and measurement, and particularly relates to a method for identifying oil-like drop attachments in a transparent pipe wall of an oil-water two-phase flow. 1. Carrying out initial image processing on the oil-water two-phase flow PTV image; 2. extracting effective characteristics; 3. randomly sampling A; 4. randomly selecting a sample attribute V; 5. recursively performing step 4 in each subspace; 6. repeating 3-5 times until t isolated trees are generated; 7. dividing A into an abnormal sample set D, a dense sample set C and a normal sample set B; 8. calculating the distance from the sample point to the clustering center in the union of the normal sample set B and the dense sample set C; 9. recalculating an oil drop edge clustering center and an oil drop attachment edge clustering center under the current classification result; 10. repeat 8 and 9 until one of the specified conditions is met; 11. and calculating the distance from each sample in the abnormal sample set D to the current oil drop edge cluster center and the current oil drop attachment edge cluster center respectively.
Description
The technical field is as follows:
the invention belongs to the technical field of petroleum engineering and measurement, and particularly relates to a method for identifying oil-like drop attachments in a transparent pipe wall of an oil-water two-phase flow.
Background art:
the accurate measurement of the oil phase flow velocity of the oil-water two-phase flow is always a difficult problem in oil field logging, and the measuring method gradually tends to adopt an undisturbed and non-contact image analysis measuring method from a logging instrument with mechanical parts. The PTV is an effective method for researching the flow characteristics of the fluid, and characteristic points in the PTV image are extracted for matching, so that the flow velocity is measured. The method has high requirements on the shot multiphase flow image, and the quality of the image directly relates to the measurement precision of the oil-water two-phase flow parameters. In the process of measuring the oil phase flow velocity of the oil-water two-phase flow, after a plurality of experiments, oil drop attachment exists on the wall of the oil well, the main components of the oil drop attachment are oil and water stains adhered to the wall of the oil well due to unclean wall, the oil drop attachment is recorded while a multi-phase flow image is collected, and the oil drop attachment is used as interference information in the image and causes great interference on the measurement and analysis of the oil phase flow velocity of the oil-water two-phase flow.
PTV oil-water two-phase flow image pixels can be divided into 5 types: the pixel belongs to the background, the pixel belongs to the interior of oil drops, the pixel belongs to the edge of the oil drops, the pixel belongs to the interior of the pipe wall oil drop attachment and the pixel belongs to the edge of the pipe wall oil drop attachment. The problem of identifying the wall oil drop attachments can be converted into the problem of image clustering, so that the problem is solved by combining a clustering algorithm with an image processing method. The edge pixels in the image are involved in the image matching process, so that only the edge pixels need to be clustered.
The K-means algorithm is the most common clustering division algorithm, has the characteristics of simple principle, unsupervised performance and iterative optimization, and is valued by numerous scholars. Although the K-means algorithm has the advantages of simple implementation and high execution efficiency, the clustering effect of the K-means algorithm is greatly influenced by the selected initial central point and the outliers in the samples. The initial clustering center point selected randomly may cause the clustering result to fall into a local optimal solution, and the outliers in the sample data may cause the clustering center to deviate from the optimal position, so that the optimal clustering effect cannot be achieved, and therefore, the initial clustering center and the outliers in the data must be preprocessed before clustering.
The isolated forest algorithm is an unsupervised data anomaly detection algorithm, and compared with other algorithms, the method does not use distance or density for detection, so that large calculation amount during detection based on a distance and density method is avoided. In an isolated forest algorithm, the abnormal degree of a sample in data is judged based on a cutting idea, outlier data can be divided independently by cutting the outlier data for a few times because the outlier data is separated from other data points, dense data is just opposite, the abnormal degree of all samples can be obtained by utilizing the characteristic, and the linear time complexity is low. The isolated forest can judge the separating degree of the sample in the data, so that the traditional K-means algorithm is improved by adopting the isolated forest, the data are divided into three categories according to the abnormal score of the data under the algorithm, the high score is used as an outlier, and the clustering process is not involved; the low score is taken as a dense point, and the initial clustering center is selected; and taking other points as common normal data to participate in the iterative clustering process.
The invention content is as follows:
the invention aims to solve the problem of measurement accuracy reduction caused by serious interference of transparent tube wall oil drop attachments when PTV is adopted to measure the oil phase flow velocity of oil-water two-phase flow, and provides an image preprocessing method for the transparent tube wall oil drop attachments of the oil-water two-phase flow. The method solves the problem that when PTV is used for measuring the oil phase flow velocity of the oil-water two-phase flow, the attachment of transparent tube wall oil drops seriously interferes the measurement. Extracting effective characteristics of a plurality of oil-water two-phase flow images by adopting an image clustering method; the isolated forest algorithm is utilized to overcome the defect that the traditional K-means algorithm is greatly influenced by initial clustering center selection and outliers, the image clustering effect is enhanced, and the oil drop attachment identification accuracy is improved.
The technical scheme adopted by the invention is as follows: a method for identifying oil-like drop attachments in an oil-water two-phase flow transparent pipe wall comprises the following steps:
the method comprises the following steps: carrying out initial image processing on the collected multiple oil-water two-phase flow PTV images, wherein the processing comprises filtering, denoising and sharpening, and adopting a Canny operator to carry out oil drop and oil drop attachment edge detection;
step two: randomly selecting an image as a current frame image, and respectively extracting the color difference in the neighborhood of the edge pixels of oil drops and oil drop attachment, the gray value difference of the edge pixels of multi-frame images of the oil drops and the oil drop attachment and the motion displacement difference of the oil drops and the oil drop attachment as effective characteristics according to the detected edge;
step three: randomly sampling a characteristic data set A, wherein the characteristic data set A is a set of color difference characteristics in the neighborhood of oil drop and oil drop attachment edge pixels of the current frame image, multi-frame image edge pixel gray value difference characteristics of the oil drop and oil drop attachment, and motion displacement difference characteristics of the oil drop and the oil drop attachment, and A is { X { (X)1,X2,…,Xi},XiThe sample is the ith sample in A, i is the edge pixel serial number, i is1, 2, …, and L is the number of edge pixels detected in the image; extracting N samples as a subsample set AsubN is the size of the extracted subsample set, and the isolated tree, lambda, is initially built1Is a sequence number of the isolated tree layer, λ1Has a value range of lambda 11,2, …, l, l is the height of the isolated tree, o is the node number, the value range of o isIn this step lambda1Is 1;
step four: randomly select a sample attribute V at AsubIn the V attribute value of (1)Generating a cutting point q, wherein the value of q is between the maximum value and the minimum value of the sample V attribute value, and under the V attribute, taking q as the cutting point AsubPartitioning into left node sample setsAnd right node sample setRandomly selecting left sample attributesAnd right sample attributesThe current left sample attribute is denoted VleftThe current right sample attribute is denoted as VrightThe current left node sample set is denoted GleftThe current right node sample set is denoted GrightWherein λ is1Is the sequence number of the isolated tree layer, and omicron is the node sequence number. At this time, the process of the present invention,
step five: lambda [ alpha ]1Is increased by 1, at the current left node sample set GleftCurrent left sample property VleftRandomly generating a left cut point q in the valuesleft,qleftIs located at the current left sample attribute VleftBetween the maximum and minimum of the values, at the current left sample property VleftAt a value of qleftSet G of current left node samples for left cut pointleftPartitioning into left node subsamplesRight node subsample setAt the current right node sample set GrightCurrent right sample property VrightRandomly generating a right cut point q in the valueright,qrightIs located at the current right sample attribute VrightBetween the maximum and minimum of the values, at the current right sample property VrightAt a value of qrightSet of current right node samples G for right cut pointrightPartitioning into left node sample subsetsRight node sample subsetRandomly selecting left sample attributesAnd right sample attributesUpdating current left sample attributesUpdating current right sample attributesCurrent left node sample setUpdating a current right node sample setWherein tau is1Is the horizontal sequence number, tau, of the left node subsample set2Is the horizontal sequence number, tau, of the right node subsample set3For the left node sample subset horizontal sequence number, τ4Is the horizontal sequence number of the right node sample subset;
step six: repeating the step five until any one of the following conditions (1) is met and G on the current left node sample setleftAnd the current right node sample set GrightOnly one sample point in; (2) current left node sample set GleftAll samples in the same feature and the current right node sample set GrightAll the samples have the same characteristics; (3) the isolated tree reaches a limited height l;
step seven: repeating the third step and the sixth step until t isolated trees are generated, inputting each sample in the characteristic data set A into the t isolated trees, and calculating the abnormal score S of the sample;
step eight: dividing the characteristic data set A into an abnormal sample set D, a dense sample set C and a normal sample set B, and selecting an initial clustering center C of oil drop edges in the dense sample set C1Initial clustering center c of oil drop attachment edge2;
Step nine: calculating the distance d (X) between the sample point and the cluster center in the union of the normal sample set B and the dense sample set C,c′j) And clustering the union sample points of the normal sample set B and the dense sample set C to obtain the union sample serial number X of the normal sample set B and the dense sample set CThe first sample in the union of the normal sample set B and the dense sample set C, j is the serial number of the cluster center, and j is1, 2, C'1Is the current drop edge cluster center, c'2Clustering the edge of the current oil drop attachment;
step ten: recalculating the oil drop edge clustering center c' under the classification result of the step nine1Oil drop attachment edge clustering center c2;
Step eleven: and repeating the ninth step and the tenth step until one of the specified conditions is met, wherein the specified conditions comprise: (1) reaching the specified iteration times; (2) the change of the clustering target function value is less than a threshold value tr;
step twelve: calculating the current oil drop edge clustering center c 'of each sample in the abnormal sample set D'1And c 'is the edge clustering center of current oil drop attachment'2And dividing the samples into cluster classes to which the cluster centers closest to the samples belong according to the distance.
Further, in the second step, the extracted image features are as follows:
(1) oil drop and oil drop attachment edge pixel neighborhood interior color difference characteristics:
descriptor f of color difference characteristics in oil drop and oil drop attachment edge pixel neighborhoodi c1Is defined as follows
fi c1=ωHσH+ωSσS+ωVσV
Wherein f isi c1Is a descriptor of the color difference characteristics in the neighborhood of the oil drop and oil drop attachment edge pixel of the ith edge pixel, omegaHIs a weight of the color difference on the H color component, and ωH<1,ωSIs a weight of the color difference on the S color component, and ωS<1,ωVIs a weight of the color difference on the V color component, and ωV<1,σHFor the color difference on the color component of H,σSfor the color difference on the S color component,σVfor the color difference on the V color component,ii is the pixel neighborhood x coordinate, jj is the pixel neighborhood y coordinate, M1Is the maximum value of the x coordinate of the pixel neighborhood, M2Is the maximum value of the y-coordinate of the pixel neighborhood,is the H color component value of the pixel at (ii, jj) within the neighborhood,is the S color component value of the pixel at (ii, jj) within the neighborhood,is the V color of the pixel at (ii, jj) in the neighborhoodComponent value, muHIs the average value, mu, of the H color components of the pixels in the neighborhoodSIs the average value, mu, of the S color components of the pixels in the neighborhoodVIs the average of the V color components of the pixels in the neighborhood;
(2) oil drops and oil drop attachment multi-frame image edge pixel gray value difference characteristics:
descriptor f of multi-frame image edge pixel gray value difference characteristic of oil drop and oil drop-like attachmenti c2The definition is as follows:
wherein f isi c2Is a descriptor of the gray value difference characteristic of oil drops of the ith edge pixel and oil drop attachment multi-frame image edge pixels, ncIs the number of frames selected for the frame,is the gray value of the ith edge pixel in the current frame image, k is the serial number of the image frame number,the gray value of the pixel at the position corresponding to the current frame image in the kth frame image is obtained;
(3) the oil drop and the oil drop attachment motion displacement difference characteristics are as follows:
the descriptor f of the motion displacement difference characteristici m1The definition is as follows:
wherein f isi m1Is a descriptor of the difference characteristic of the motion displacement of oil drops and oil drop attachment of the ith edge pixel, nαIn order to use the number of the query windows with different sizes, w is the serial number of the query window used, w is1, 2, …, nα,lwThe displacement obtained by adopting query windows with different sizes is calculated according to the following formula:
wherein u iswIs the displacement, v, of the pixel in the x direction obtained by cross-correlation calculation using the w-th query windowwIs the displacement of the pixel in the y direction obtained by cross-correlation calculation using the w-th query window:
whereinIs the coordinate of the x direction of the cross-correlation matrix,. phi.wAdopting a gray scale map f of an inquiry area covered by a w inquiry window for a current frame image'wA grey-scale map of the query region covered with the w-th query window is used for the next frame of image,is fwThe average pixel value within the corresponding region,is f'wThe average pixel value in the corresponding area, M is the x-direction pixel coordinate of the query area covered by the query window, and M is1, 2w,MwThe maximum value of the pixel coordinates in the x direction of the query area covered by the query window, N is the pixel coordinates in the y direction of the query area covered by the query window, and N is1, 2w,NwQuery region y-direction pixels covered by query windowThe maximum value of the coordinates.
Further, the three-step neutron sample set AsubExpressed as:
whereinIs a color difference feature set in the neighborhood of the edge pixel, i' is a subsample set AsubThe inner edge pixel number, i' is1, 2, …, N is the decimated sub-sample set size,as a subsample set AsubThe color difference feature in the edge pixel neighborhood of the ith' edge pixel,for a multi-frame image edge pixel gray value difference feature set,as a subsample set AsubThe multi-frame image edge pixel gray value difference characteristic of the ith' edge pixel,in order to move the set of difference features of displacement,as a subsample set AsubMotion displacement difference feature of the ith' edge pixel in the subsample set AsubInner ith' edge pixel sample
Further, the sample attribute in step four, V ∈ { F }c1,Fc2,Fm1And putting the samples with the V attribute value smaller than the cutting point q in the left node sample setPutting samples with V attribute values larger than the cut point q into a right node sample set
Further, in the fifth step, the current left node sample set G is usedleftIn, the current left sample attribute VleftValue less than left cutting point qleftIs placed in the left node subsample setIn, the current left sample attribute VleftValue greater than left cutting point qleftIs placed in the right node subsample setAt the current right node sample set GrightIn, the current right sample attribute VrightValue less than right cutting point qrightIs placed in the left node sample subsetAttribute V of current right samplerightValue greater than right cut point qrightIs placed in the right node sample subset
Further, the function of the condition (3) l in the sixth step and the size N of the extracted subsample set is as follows:
l=ceiling(log2N)
ceiling (log) in the formula2N) represents log2And N is rounded up.
Further, in the seventh step, the anomaly score S of the feature data set a is calculated according to the following method:
S=[s(X1,N),s(X2,N),...,s(XL,N)]T
wherein s (X)iN) is a sample XiAbnormality score E (h (X) of (1)i) Is the average h (X) of t isolated treesi) Value h (X)i) For path length, i.e. in an orphan tree, from AsubTo XiThe number of edges experienced, (N) is the average of the path lengths given the decimated sub-sample set size N, defined as follows:
wherein H is the sum of the tones, and the calculation formula is as follows:
H(N-1)=ln(N-1)+0.5772156649。
further, the abnormality score is larger than the specified value Tr in the eighth stepHIs marked as an outlier sample, and the outlier score is less than a specified value TrLThe samples of (2) are marked as dense samples, so that the original sample set A is divided into an abnormal sample set D, a dense sample set C and a normal sample set B:
wherein Xis1Is the sample of the dense sample set C, is1 is the sample number of the dense sample set C, Xis2Is the sample of the abnormal sample set D, is2 is the sample number of the abnormal sample set D, N*Represents a positive integer set, sgn is a step function;
at this time, oil drop edge cluster center c'1=c1Oil droplet-like deposit edge cluster center c'2=c2。
Further, the distance d (X) in the step nine,c′j) Is represented as follows:
d(X,c′j)=||X,c′j||2
and clustering the union concentrated sample points of the normal sample set B and the dense sample set C according to the following formula:
wherein ζRepresents the first sample XThe category is the union concentrated sample number, zeta, of the normal sample set B and the dense sample set C∈{1,2},ζζ 1, represents an oil drop edge class2, the oil drop attachment edge-like class is represented; n issubThe total number of samples in the union set of the normal sample set B and the dense sample set C,represents d (X),c′j) The value of j at the minimum is taken.
Further, in the step ten, the oil drop edge cluster center c ″ is recalculated according to the following formula1Oil drop attachment edge clustering center c2:
Where j is the serial number of the cluster center, and j is1, 2. At this time, the current drop edge cluster center c'1′=c″1Current oil drop attachment edge cluster center c'2=c″2;
Further, the condition (2) clustering objective function in the step eleven is expressed as follows:
wherein the function of the coefficientLambda is the serial number of the clustering center,represents d (X),c′λ) Taking the value of lambda at the minimum value;
further, in the step twelve, the samples in the abnormal sample set D are classified according to the following formula:
η is the sample number in the abnormal sample set D, XηIs the η th sample in the abnormal sample set D, ζηRepresents the η th sample XηClass to which it belongs, ζη∈{1,2},ζηζ 1, represents an oil drop edge classηThe term "2" indicates an oil drop attachment edge-like substance.
The invention has the beneficial effects that: in order to solve the problem of measurement accuracy reduction caused by serious interference of transparent tube wall oil drop attachments when PTV is adopted to measure the oil phase flow velocity of the oil-water two-phase flow, a method for preprocessing an image of the oil drop attachments on the wall of the transparent tube of the oil-water two-phase flow is provided. The method solves the problem that when PTV is used for measuring the oil phase flow velocity of the oil-water two-phase flow, the attachment of transparent tube wall oil drops seriously interferes the measurement. Extracting effective characteristics of a plurality of oil-water two-phase flow images by adopting an image clustering method; the isolated forest algorithm is utilized to overcome the defect that the traditional K-means algorithm is greatly influenced by initial clustering center selection and outliers, the image clustering effect is enhanced, and the oil drop attachment identification accuracy is improved. Its main advantage is as follows:
(1) the invention adopts an image clustering method to extract a plurality of characteristics of the oil-water two-phase flow picture for clustering, thereby effectively identifying the edges of the wall oil drop attachments of the transparent pipe;
(2) the method adopts the improved K-means based on the isolated forest, and compared with other algorithms, the method does not use the distance or the density for detection, avoids large calculation amount when the detection is carried out based on the distance and density method, and has lower linear time complexity;
(3) the invention can effectively reduce the interference of the transparent tube wall oil drop attachment on PTV characteristic matching, thereby improving the measurement precision of the oil phase flow velocity of the oil-water two-phase flow.
Description of the drawings:
FIG. 1 is a diagram illustrating the result of image feature matching under the influence of oil drop attachment in the first embodiment;
FIG. 2 is a graph showing the result of the improved K-means image clustering identification and elimination of the oil-water two-phase flow picture with the oil drop-like attachment in the first embodiment;
fig. 3 is a measurement result diagram of the application of the identification and elimination of oil drop attachments of the transparent tube wall of the oil-water two-phase flow based on the traditional K-means and the identification and elimination of oil drop attachments of the transparent tube wall of the oil-water two-phase flow based on the improved K-means to the measurement of the oil phase flow velocity PTV of the oil-water two-phase flow of the vertical well in the first embodiment.
The specific implementation mode is as follows:
example one
Referring to the drawings, fig. 1 shows the image feature matching result under the influence of oil drop attachment, and it can be seen that many oil drop attachments are mistakenly matched as oil drops; FIG. 2 is the result of improved K-means image clustering identification and elimination of an oil-water two-phase flow picture with oil drop-like attachments; FIG. 3 is a measurement result obtained by applying recognition and elimination of oil drop attachments of transparent tube walls of oil-water two-phase flow based on traditional K-means and recognition and elimination of oil drop attachments of transparent tube walls of oil-water two-phase flow based on improved K-means to oil phase flow velocity PTV measurement of oil phase of oil-water two-phase flow of a vertical well respectively. The specific implementation mode is as follows:
firstly, oil drops are used as tracer particles, an LED backlight light source illuminates an area to be measured, and a high-speed camera is adopted to continuously shoot a plurality of vertical well oil-water two-phase flow pictures;
secondly, performing initial image processing on the collected multiple oil-water two-phase flow PTV images, wherein the initial image processing comprises filtering, denoising and sharpening, and performing oil drop and oil drop attachment edge detection by adopting a Canny operator;
randomly selecting an image as a current frame image, and respectively extracting color difference in the neighborhood of edge pixels of oil drops and oil drop attachment, gray value difference of multi-frame image edge pixels of the oil drops and the oil drop attachment and motion displacement difference of the oil drops and the oil drop attachment as effective characteristics according to the detected edge;
randomly sampling a characteristic data set A, wherein the characteristic data set A is a set of color difference characteristics in the neighborhood of oil drop and oil drop attachment edge pixels of image edge pixels, multi-frame image edge pixel gray value difference characteristics of oil drop and oil drop attachment, and motion displacement difference characteristics of oil drop and oil drop attachment, and A is { X ═ X { (X)1,X2,…,Xi},XiThe sample is the ith sample in A, i is the edge pixel serial number, i is1, 2, …, and L is the number of edge pixels detected in the image; extracting N samples as a subsample set AsubN is the size of the extracted subsample set, and the isolated tree, lambda, is initially built1Is a sequence number of the isolated tree layer, λ1Has a value range of lambda 11,2, …, l, l is the height of the isolated tree, o is the node number, the value range of o isIn this step lambda1Is 1;
fifthly, randomly selecting a sample attribute V at AsubRandomly generating a cutting point q in the V attribute values, wherein the value of q is between the maximum value and the minimum value of the sample V attribute values, and under the V attribute, taking q as a cutting point to obtain an AsubPartitioning into left node sample setsAnd right node sample setRandomly selecting left sample attributesAnd right sample attributesThe current left sample attribute is denoted VleftThe current right sample attribute is denoted as VrightThe current left node sample set is denoted GleftThe current right node sample set is denoted GrightWherein λ is1Is the number of the isolated tree layer, o%Is the node sequence number. At this time, the process of the present invention,
six, lambda1Is increased by 1, at the current left node sample set GleftCurrent left sample property VleftRandomly generating a left cut point q in the valuesleft,qleftIs located at the current left sample attribute VleftBetween the maximum and minimum of the values, at the current left sample property VleftAt a value of qleftSet G of current left node samples for left cut pointleftPartitioning into left node subsamplesRight node subsample setAt the current right node sample set GrightCurrent right sample property VrightRandomly generating a right cut point q in the valueright,qrightIs located at the current right sample attribute VrightBetween the maximum and minimum of the values, at the current right sample property VrightAt a value of qrightSet of current right node samples G for right cut pointrightPartitioning into left node sample subsetsRight node sample subsetRandomly selecting left sample attributesAnd right sample attributesUpdating the currentLeft sample PropertiesUpdating current right sample attributesCurrent left node sample setUpdating a current right node sample setWherein tau is1Is the horizontal sequence number, tau, of the left node subsample set2Is the horizontal sequence number, tau, of the right node subsample set3For the left node sample subset horizontal sequence number, τ4Is the horizontal sequence number of the right node sample subset;
seventhly, repeating six steps until any one of the following conditions (1) is met and G on the current left node sample set is obtainedleftAnd the current right node sample set GrightOnly one sample point in; (2) current left node sample set GleftAll samples in the same feature and the current right node sample set GrightAll the samples have the same characteristics; (3) the isolated tree reaches a limited height l;
eighthly, repeating four to seven steps until t isolated trees are generated, traversing (inputting) each sample in the characteristic data set A to the t isolated trees, and calculating the abnormal score S of the sample;
ninthly, dividing the characteristic data set A into an abnormal sample set D, a dense sample set C and a normal sample set B, and selecting an initial clustering center C of oil drop edges in the dense sample set C1Initial clustering center c of oil drop attachment edge2;
Ten, calculating the distance d (X) between the sample point and the cluster center in the union of the normal sample set B and the dense sample set C,c′j) And clustering the union sample points of the normal sample set B and the dense sample set C to obtain the union sample serial number X of the normal sample set B and the dense sample set CIs the union of normal sample set B and dense sample set CThe first sample, j is the serial number of the cluster center, j is1, 2, c'1Is the current drop edge cluster center, c'2Clustering the edge of the current oil drop attachment;
eleven, recalculating the oil drop edge clustering center c' under the ten-step classification result1Oil drop attachment edge clustering center c2;
Twelve, repeating ten to eleven until one of the specified conditions is met, wherein the specified conditions comprise: (1) reaching the specified iteration times; (2) the change of the clustering target function value is less than a threshold value tr;
thirteen, calculating each sample in the abnormal sample set D to the current oil drop edge clustering center c'1And c 'is the edge clustering center of current oil drop attachment'2According to the distance, dividing the sample into cluster classes to which the cluster centers closest to the sample belong;
fourteen, processing the pixels identified as the oil drop attachment edges under the current clustering result, and removing the effect of the pixels in the PTV velocity measurement as shown in figure 2;
fifteen, identifying transparent tube wall oil drop attachments of the oil-water two-phase flow based on the traditional K-means and identifying transparent tube wall oil drop attachments of the oil-water two-phase flow based on the improved K-means are applied to the PTV measurement of the oil phase flow velocity of the oil-water two-phase flow of the vertical well respectively, the measurement result is shown in figure 3, under the working conditions of different water contents and different flow velocities, the PTV measurement precision based on the traditional K-means is 92.2%, the PTV measurement precision based on the improved K-means is 94.8%, and the PTV measurement precision based on the improved K-means is higher than that of the traditional K-means.
Claims (10)
1. A method for identifying oil-like drop attachments in an oil-water two-phase flow transparent pipe wall is characterized by comprising the following steps: the identification method comprises the following steps:
the method comprises the following steps: carrying out initial image processing on the collected multiple oil-water two-phase flow PTV images, wherein the processing comprises filtering, denoising and sharpening, and adopting a Canny operator to carry out oil drop and oil drop attachment edge detection;
step two: randomly selecting an image as a current frame image, and respectively extracting the color difference in the neighborhood of the edge pixels of oil drops and oil drop attachment, the gray value difference of the edge pixels of multi-frame images of the oil drops and the oil drop attachment and the motion displacement difference of the oil drops and the oil drop attachment as effective characteristics according to the detected edge;
step three: randomly sampling a characteristic data set A, wherein the characteristic data set A is a set of color difference characteristics in the neighborhood of oil drop and oil drop attachment edge pixels of the current frame image, multi-frame image edge pixel gray value difference characteristics of the oil drop and oil drop attachment, and motion displacement difference characteristics of the oil drop and the oil drop attachment, and A is { X { (X)1,X2,…,Xi},XiThe sample is the ith sample in A, i is the edge pixel serial number, i is1, 2, …, and L is the number of edge pixels detected in the image; extracting N samples as a subsample set AsubN is the size of the extracted subsample set, and the isolated tree, lambda, is initially built1Is a sequence number of the isolated tree layer, λ1Has a value range of lambda11,2, …, l, l is the height of the isolated tree, o is the node number, the value range of o isIn this step lambda1Is 1;
step four: randomly select a sample attribute V at AsubRandomly generating a cutting point q in the V attribute values, wherein the value of q is between the maximum value and the minimum value of the sample V attribute values, and under the V attribute, taking q as a cutting point to obtain an AsubPartitioning into left node sample setsAnd right node sample setRandomly selecting left sample attributesAnd right sample genusProperty of (2)The current left sample attribute is denoted VleftThe current right sample attribute is denoted as VrightThe current left node sample set is denoted GleftThe current right node sample set is denoted GrightWherein λ is1Is the sequence number of the isolated tree layer, and omicron is the node sequence number; at this time, the process of the present invention,
step five: lambda [ alpha ]1Is increased by 1, at the current left node sample set GleftCurrent left sample property VleftRandomly generating a left cut point q in the valuesleft,qleftIs located at the current left sample attribute VleftBetween the maximum and minimum of the values, at the current left sample property VleftAt a value of qleftSet G of current left node samples for left cut pointleftPartitioning into left node subsamplesRight node subsample setAt the current right node sample set GrightCurrent right sample property VrightRandomly generating a right cut point q in the valueright,qrightIs located at the current right sample attribute VrightBetween the maximum and minimum of the values, at the current right sample property VrightAt a value of qrightSet of current right node samples G for right cut pointrightPartitioning into left node sample subsetsRight node sampleSubset of the bookRandomly selecting left sample attributesAnd right sample attributesUpdating current left sample attributesUpdating current right sample attributesCurrent left node sample setUpdating a current right node sample setWherein tau is1Is the horizontal sequence number, tau, of the left node subsample set2Is the horizontal sequence number, tau, of the right node subsample set3For the left node sample subset horizontal sequence number, τ4Is the horizontal sequence number of the right node sample subset;
step six: repeating the step five until any one of the following conditions (1) is met and G on the current left node sample setleftAnd the current right node sample set GrightOnly one sample point in; (2) current left node sample set GleftAll samples in the same feature and the current right node sample set GrightAll the samples have the same characteristics; (3) the isolated tree reaches a limited height l;
step seven: repeating the third step and the sixth step until t isolated trees are generated, inputting each sample in the characteristic data set A into the t isolated trees, and calculating the abnormal score S of the sample;
step eight: will specially beThe characteristic data set A is divided into an abnormal sample set D, a dense sample set C and a normal sample set B, and an oil drop edge initial clustering center C is selected from the dense sample set C1Initial clustering center c of oil drop attachment edge2;
Step nine: calculating the distance d (X) between the sample point and the cluster center in the union of the normal sample set B and the dense sample set C,c′j) And clustering the union sample points of the normal sample set B and the dense sample set C to obtain the union sample serial number X of the normal sample set B and the dense sample set CThe first sample in the union of the normal sample set B and the dense sample set C, j is the serial number of the cluster center, and j is1, 2, C'1Is the current drop edge cluster center, c'2Clustering the edge of the current oil drop attachment;
step ten: recalculating the oil drop edge clustering center c' under the classification result of the step nine1Oil drop attachment edge clustering center c2;
Step eleven: and repeating the ninth step and the tenth step until one of the specified conditions is met, wherein the specified conditions comprise: (1) reaching the specified iteration times; (2) the change of the clustering target function value is less than a threshold value tr;
step twelve: calculating the current oil drop edge clustering center c 'of each sample in the abnormal sample set D'1And c 'is the edge clustering center of current oil drop attachment'2And dividing the samples into cluster classes to which the cluster centers closest to the samples belong according to the distance.
2. The method for identifying the oil-like drop attachment in the oil-water two-phase flow transparent pipe wall according to claim 1, wherein the method comprises the following steps: in the second step, the extracted image features are as follows:
(1) oil drop and oil drop attachment edge pixel neighborhood interior color difference characteristics:
descriptor f of color difference characteristics in oil drop and oil drop attachment edge pixel neighborhoodi c1Is defined as follows
fi c1=ωHσH+ωSσS+ωVσV
Wherein f isi c1Is a descriptor of the color difference characteristics in the neighborhood of the oil drop and oil drop attachment edge pixel of the ith edge pixel, omegaHIs a weight of the color difference on the H color component, and ωH<1,ωSIs a weight of the color difference on the S color component, and ωS<1,ωVIs a weight of the color difference on the V color component, and ωV<1,σHFor the color difference on the color component of H,σSfor the color difference on the S color component,σVfor the color difference on the V color component,ii is the pixel neighborhood x coordinate, jj is the pixel neighborhood y coordinate, M1Is the maximum value of the x coordinate of the pixel neighborhood, M2Is the maximum value of the y-coordinate of the pixel neighborhood,is the H color component value of the pixel at (ii, jj) within the neighborhood,is the S color component value of the pixel at (ii, jj) within the neighborhood,is the V color component value, μ, of the pixel at (ii, jj) within the neighborhoodHIs the average value, mu, of the H color components of the pixels in the neighborhoodSIs the average value, mu, of the S color components of the pixels in the neighborhoodVIs the average of the V color components of the pixels in the neighborhood;
(2) oil drops and oil drop attachment multi-frame image edge pixel gray value difference characteristics:
descriptor f of multi-frame image edge pixel gray value difference characteristic of oil drop and oil drop-like attachmenti c2The definition is as follows:
wherein f isi c2Is a descriptor of the gray value difference characteristic of oil drops of the ith edge pixel and oil drop attachment multi-frame image edge pixels, ncIs the number of frames selected for the frame,is the gray value of the ith edge pixel in the current frame image, k is the serial number of the image frame number,the gray value of the pixel at the position corresponding to the current frame image in the kth frame image is obtained;
(3) the oil drop and the oil drop attachment motion displacement difference characteristics are as follows:
the descriptor f of the motion displacement difference characteristici m1The definition is as follows:
wherein f isi m1Is a descriptor of the difference characteristic of the motion displacement of oil drops and oil drop attachment of the ith edge pixel, nαIn order to use the number of the query windows with different sizes, w is the serial number of the query window used, w is1, 2, …, nα,lwThe displacement obtained by adopting query windows with different sizes is calculated according to the following formula:
wherein u iswAdopts the w-th inquiry windowDisplacement, v, of pixel in x direction calculated by cross correlationwIs the displacement of the pixel in the y direction obtained by cross-correlation calculation using the w-th query window:
whereinIs the coordinate of the x direction of the cross-correlation matrix,. phi.wAdopting a gray scale map f of an inquiry area covered by a w inquiry window for a current frame image'wA grey-scale map of the query region covered with the w-th query window is used for the next frame of image,is fwThe average pixel value within the corresponding region,is f'wThe average pixel value in the corresponding area, M is the x-direction pixel coordinate of the query area covered by the query window, and M is1, 2w,MwThe maximum value of the pixel coordinates in the x direction of the query area covered by the query window, N is the pixel coordinates in the y direction of the query area covered by the query window, and N is1, 2w,NwAnd the maximum value of the pixel coordinate in the y direction of the query area covered by the query window.
3. An oil according to claim 1The method for identifying oil-like drop attachments in the wall of the water two-phase flow transparent pipe is characterized by comprising the following steps of: the three-step neutron sample set AsubExpressed as:
whereinIs a color difference feature set in the neighborhood of the edge pixel, i' is a subsample set AsubThe inner edge pixel number, i' is1, 2, …, N is the decimated sub-sample set size,as a subsample set AsubThe color difference feature in the edge pixel neighborhood of the ith' edge pixel,for a multi-frame image edge pixel gray value difference feature set,as a subsample set AsubThe multi-frame image edge pixel gray value difference characteristic of the ith' edge pixel,in order to move the set of difference features of displacement,as a subsample set AsubMotion displacement difference feature of the ith' edge pixel in the subsample set AsubInner ith' edge pixel sample
4. The method for identifying the oil-like drop attachments in the transparent tube wall of the oil-water two-phase flow according to claim 1, wherein the sample attribute V ∈ { F in the fourth step isc1,Fc2,Fm1And putting the samples with the V attribute value smaller than the cutting point q in the left node sample setPutting samples with V attribute values larger than the cut point q into a right node sample setWhereinIs a color difference feature set in the neighborhood of the edge pixel, i' is a subsample set AsubThe inner edge pixel number, i' is1, 2, …, N is the decimated sub-sample set size,as a subsample set AsubThe color difference feature in the edge pixel neighborhood of the ith' edge pixel,for a multi-frame image edge pixel gray value difference feature set,as a subsample set AsubThe multi-frame image edge pixel gray value difference characteristic of the ith' edge pixel,in order to move the set of difference features of displacement,as a subsample set AsubMotion displacement difference feature of the ith' edge pixel in the subsample set AsubInner ith' edge pixel sample
5. The method for identifying the oil-like drop attachment in the oil-water two-phase flow transparent pipe wall according to claim 1, wherein the method comprises the following steps: in the fifth step, the current left node sample set GleftIn, the current left sample attribute VleftValue less than left cutting point qleftIs placed in the left node subsample setIn, the current left sample attribute VleftValue greater than left cutting point qleftIs placed in the right node subsample setAt the current right node sample set GrightIn, the current right sample attribute VrightValue less than right cutting point qrightIs placed in the left node sample subsetAttribute V of current right samplerightValue greater than right cut point qrightIs placed in the right node sample subset。
6. The method for identifying the oil-like drop attachment in the oil-water two-phase flow transparent pipe wall according to claim 1, wherein the method comprises the following steps: in the sixth step, the functional relation between the condition (3) l and the size N of the extracted subsample set is as follows:
l=ceiling(log2N)
ceiling (log) in the formula2N) represents log2And N is rounded up.
7. The method for identifying the oil-like drop attachment in the oil-water two-phase flow transparent pipe wall according to claim 1, wherein the method comprises the following steps: in the seventh step, the abnormal score S of the feature data set A is calculated according to the following method:
S=[s(X1,N),s(X2,N),...,s(XL,N)]T
wherein s (X)iN) is a sample XiAbnormality score E (h (X) of (1)i) Is the average h (X) of t isolated treesi) Value h (X)i) For path length, i.e. in an orphan tree, from AsubTo XiThe number of edges experienced, (N) is the average of the path lengths given the decimated sub-sample set size N, defined as follows:
wherein H is the sum of the tones, and the calculation formula is as follows:
H(N-1)=ln(N-1)+0.5772156649。
8. the method for identifying the oil-like drop attachment in the oil-water two-phase flow transparent pipe wall according to claim 1, wherein the method comprises the following steps: step eight, the abnormal score is larger than the specified value TrHIs marked as an outlier sample, and the outlier score is less than a specified value TrLThe samples of (2) are marked as dense samples, so that the original sample set A is divided into an abnormal sample set D, a dense sample set C and a normal sample set B:
wherein Xis1Is the sample of the dense sample set C, is1 is the sample number of the dense sample set C, Xis2Is a sample of the abnormal sample set D, is2 is an abnormal sampleSample number of set D, N*Represents a positive integer set, sgn is a step function;
wherein s (X)iN) is a sample XiAbnormality score E (h (X) of (1)i) Is the average h (X) of t isolated treesi) Value h (X)i) For path length, i.e. in an orphan tree, from AsubTo XiThe number of edges experienced, (N) is the average of the path lengths given the decimated sub-sample set size N, defined as follows:
wherein H is the sum of the tones, and the calculation formula is as follows:
H(N-1)=ln(N-1)+0.5772156649;
at this time, oil drop edge cluster center c'1=c1Oil droplet-like deposit edge cluster center c'2=c2。
9. The method for identifying the oil-like drop attachment in the oil-water two-phase flow transparent pipe wall according to claim 1, wherein the method comprises the following steps: distance d (X) in step nine,c′j) Is represented as follows:
d(X,c′j)=||X,c′j||2
and clustering the union concentrated sample points of the normal sample set B and the dense sample set C according to the following formula:
wherein ζRepresents the first sample XThe category is the union concentrated sample number, zeta, of the normal sample set B and the dense sample set C∈{1,2},ζζ 1, represents an oil drop edge class2, the oil drop attachment edge-like class is represented; n issubThe total number of samples in the union set of the normal sample set B and the dense sample set C,represents d (X),c′j) The value of j at the minimum is taken.
10. The method for identifying the oil-like drop attachment in the oil-water two-phase flow transparent pipe wall according to claim 1, wherein the method comprises the following steps: in the step ten, the clustering center c' of the oil drop edges is recalculated according to the following formula1Oil drop attachment edge clustering center c2:
J is a serial number of a clustering center, and j is1 and 2; at this time, the current drop edge cluster center c'1=c″1Current oil drop attachment edge cluster center c'2=c″2(ii) a The condition (2) clustering objective function in the step eleven is represented as follows:
wherein the function of the coefficientLambda is the serial number of the clustering center,represents d (X),c′λ) Taking the value of lambda at the minimum value; in the step twelve, the samples in the abnormal sample set D are classified according to the following formula:
whereinη is the sample number in the abnormal sample set D, XηIs the η th sample in the abnormal sample set D, ζηRepresents the η th sample XηClass to which it belongs, ζη∈{1,2},ζηζ 1, represents an oil drop edge classη2, the oil drop attachment edge-like class is represented;
wherein ζRepresents the first sample XThe category is the union concentrated sample number, zeta, of the normal sample set B and the dense sample set C∈{1,2},ζζ 1, represents an oil drop edge class2, the oil drop attachment edge-like class is represented; n issubThe total number of samples in the union set of the normal sample set B and the dense sample set C,represents d (X),c′j) The value of j at the minimum is taken.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910966266.4A CN110751196B (en) | 2019-10-12 | 2019-10-12 | Oil-like drop attachment identification method in oil-water two-phase flow transparent pipe wall |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910966266.4A CN110751196B (en) | 2019-10-12 | 2019-10-12 | Oil-like drop attachment identification method in oil-water two-phase flow transparent pipe wall |
Publications (2)
Publication Number | Publication Date |
---|---|
CN110751196A CN110751196A (en) | 2020-02-04 |
CN110751196B true CN110751196B (en) | 2020-09-18 |
Family
ID=69278034
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910966266.4A Active CN110751196B (en) | 2019-10-12 | 2019-10-12 | Oil-like drop attachment identification method in oil-water two-phase flow transparent pipe wall |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110751196B (en) |
Families Citing this family (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN115861751B (en) * | 2022-12-06 | 2024-04-16 | 常熟理工学院 | Oil-water two-phase flow multi-oil drop identification method based on integrated characteristics |
CN117496498A (en) * | 2023-11-10 | 2024-02-02 | 常熟理工学院 | Ultrahigh water content oil-water two-phase flow oil bubble identification method |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108805865A (en) * | 2018-05-22 | 2018-11-13 | 杭州智微信息科技有限公司 | A kind of myeloplast localization method based on saturation degree cluster |
CN109410586A (en) * | 2018-12-13 | 2019-03-01 | 中南大学 | A kind of Traffic State Detection Method based on multivariate data fusion |
CN109766905A (en) * | 2018-09-28 | 2019-05-17 | 中国人民解放军空军工程大学 | Target cluster dividing method based on Self-Organizing Feature Maps |
CN109871886A (en) * | 2019-01-28 | 2019-06-11 | 平安科技(深圳)有限公司 | Abnormal point ratio optimization method, apparatus and computer equipment based on spectral clustering |
CN109948728A (en) * | 2019-03-28 | 2019-06-28 | 第四范式(北京)技术有限公司 | The method and apparatus of the training of abnormal transaction detection model and abnormal transaction detection |
CN110046665A (en) * | 2019-04-17 | 2019-07-23 | 成都信息工程大学 | Based on isolated two abnormal classification point detecting method of forest, information data processing terminal |
Family Cites Families (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20170123931A1 (en) * | 2011-08-12 | 2017-05-04 | Nexenta Systems, Inc. | Object Storage System with a Distributed Namespace and Snapshot and Cloning Features |
CN108548817B (en) * | 2018-03-26 | 2021-08-03 | 清华大学 | Multiphase floating jet experiment generating device and oil drop bubble shadow image processing method |
-
2019
- 2019-10-12 CN CN201910966266.4A patent/CN110751196B/en active Active
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108805865A (en) * | 2018-05-22 | 2018-11-13 | 杭州智微信息科技有限公司 | A kind of myeloplast localization method based on saturation degree cluster |
CN109766905A (en) * | 2018-09-28 | 2019-05-17 | 中国人民解放军空军工程大学 | Target cluster dividing method based on Self-Organizing Feature Maps |
CN109410586A (en) * | 2018-12-13 | 2019-03-01 | 中南大学 | A kind of Traffic State Detection Method based on multivariate data fusion |
CN109871886A (en) * | 2019-01-28 | 2019-06-11 | 平安科技(深圳)有限公司 | Abnormal point ratio optimization method, apparatus and computer equipment based on spectral clustering |
CN109948728A (en) * | 2019-03-28 | 2019-06-28 | 第四范式(北京)技术有限公司 | The method and apparatus of the training of abnormal transaction detection model and abnormal transaction detection |
CN110046665A (en) * | 2019-04-17 | 2019-07-23 | 成都信息工程大学 | Based on isolated two abnormal classification point detecting method of forest, information data processing terminal |
Non-Patent Citations (6)
Title |
---|
A hybrid semi-supervised approach for financial fraud detection;Jin-Miao Liu 等;《2017 International Conference on Machine Learning and Cybernetics (ICMLC)》;20171116;第217-222页 * |
Detection of Anomalous Vital Sign of Elderly Using Hybrid K-Means Clustering and Isolation Forest;Kurnianingsih 等;《TENCON IEEE Region 10 Conference Proceedings》;20181031;第913-918页 * |
Flow Velocity Field Measurement of Vertical Upward Oil-Water Two-Phase Immiscible Flow Using the Improved DPIV Algorithm Based on ICP and MLS;Han, Lianfu 等;《APPLIED SCIENCES-BASEL》;20190831;第9卷(第16期);第1-22页 * |
基于K-medoids聚类分析的多相流相态辨识方法研究;赵鸣 等;《天津理工大学学报》;20171231;第33卷(第6期);第55-59页 * |
基于异常检测的K-means改进算法研究;薛晨杰 等;《软件导刊》;20190430;第18卷(第4期);第74-78页 * |
基于目标轮廓的附着物定位与剔除方法;栗琳 等;《光电工程》;20120531;第39卷(第5期);第45-51页 * |
Also Published As
Publication number | Publication date |
---|---|
CN110751196A (en) | 2020-02-04 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN107688808B (en) | Rapid natural scene text detection method | |
CN110533084B (en) | Multi-scale target detection method based on self-attention mechanism | |
CN107368807B (en) | Monitoring video vehicle type classification method based on visual word bag model | |
EP2087456B1 (en) | Face-based image clustering | |
CN112633382B (en) | Method and system for classifying few sample images based on mutual neighbor | |
CN110796667B (en) | Color image segmentation method based on improved wavelet clustering | |
CN110490913B (en) | Image matching method based on feature description operator of corner and single line segment grouping | |
CN110751196B (en) | Oil-like drop attachment identification method in oil-water two-phase flow transparent pipe wall | |
CN111968171A (en) | Aircraft oil quantity measuring method and system based on artificial intelligence | |
Sonawane et al. | A brief survey on image segmentation methods | |
CN109766748B (en) | Pedestrian re-recognition method based on projection transformation and dictionary learning | |
CN109191493B (en) | Target tracking method based on RefineNet neural network and sparse optical flow | |
CN107808391B (en) | Video dynamic target extraction method based on feature selection and smooth representation clustering | |
CN114417993A (en) | Scratch detection method based on deep convolutional neural network and image segmentation | |
CN108491888B (en) | Environmental monitoring hyperspectral data spectrum section selection method based on morphological analysis | |
CN115965613A (en) | Cross-layer connection construction scene crowd counting method based on cavity convolution | |
CN110490170A (en) | A kind of face candidate frame extracting method | |
CN115937540A (en) | Image Matching Method Based on Transformer Encoder | |
CN113284158A (en) | Image edge extraction method and system based on structural constraint clustering | |
CN108256572B (en) | Indoor visual feature classification method based on improved naive Bayes | |
CN107092875B (en) | Novel scene recognition method | |
CN114913334A (en) | Image denoising, segmenting and identifying method for monitoring information system of meat product processing production line | |
CN106156771B (en) | water meter reading area detection algorithm based on multi-feature fusion | |
Samarabandu et al. | Image analysis techniques for visualizing the spatial organization of DNA replication sites in the mammalian cell nucleus using multichannel confocal microscopy | |
CN115047523B (en) | Seismic data dynamic correction method based on density clustering algorithm |
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 | ||
TR01 | Transfer of patent right |
Effective date of registration: 20230322 Address after: Room 102, Building 4, MAX Science Park, No. 998, Pangnan Road, Jiangling Street, Wujiang District, Suzhou City, Jiangsu Province, 215000 Patentee after: Suzhou Hengyou New Energy Technology Co.,Ltd. Address before: 163000 99 Xuefu street, hi tech Industrial Development Zone, Daqing, Heilongjiang Patentee before: NORTHEAST PETROLEUM University |
|
TR01 | Transfer of patent right |