CN113093198A - Acoustic imaging detection method based on multi-scale Markov random field - Google Patents
Acoustic imaging detection method based on multi-scale Markov random field Download PDFInfo
- Publication number
- CN113093198A CN113093198A CN202110258326.4A CN202110258326A CN113093198A CN 113093198 A CN113093198 A CN 113093198A CN 202110258326 A CN202110258326 A CN 202110258326A CN 113093198 A CN113093198 A CN 113093198A
- Authority
- CN
- China
- Prior art keywords
- scale
- markov random
- random field
- pixel
- acoustic imaging
- 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
Links
Images
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
- G01S15/00—Systems using the reflection or reradiation of acoustic waves, e.g. sonar systems
- G01S15/88—Sonar systems specially adapted for specific applications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
- G01S15/00—Systems using the reflection or reradiation of acoustic waves, e.g. sonar systems
- G01S15/88—Sonar systems specially adapted for specific applications
- G01S15/89—Sonar systems specially adapted for specific applications for mapping or imaging
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/29—Graphical models, e.g. Bayesian networks
- G06F18/295—Markov models or related models, e.g. semi-Markov models; Markov random fields; Networks embedding Markov models
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/11—Region-based segmentation
Abstract
The invention discloses an acoustic imaging detection method based on a multi-scale Markov random field, which comprises the following steps: firstly, constructing a multi-scale acoustic imaging data Markov random field model; secondly, combining a multi-scale Markov random field model and a level set model to construct a multi-scale Markov random field-level set combined model; and finally, the acoustic imaging detection is realized through the convergence of the multi-scale Markov random field-level set combined model, and the substantial defects of low resolution and high noise of the acoustic imaging information in the detection process are overcome.
Description
Technical Field
The invention relates to an acoustic imaging detection method, in particular to an acoustic imaging detection method combining a multi-scale Markov random field model and a level set model.
Background
With the development of electronic computers, digitizing circuits and highly integrated microelectronic devices, acoustic imaging technology has improved dramatically. The acoustic imaging determines the position of a sound source by measuring the difference of signal phases of sound waves reaching each microphone in a certain space range and the amplitude of the sound source, and simultaneously displays the distribution of the sound source in the space in an image mode, wherein the strength of the sound is expressed by color and brightness in the image. Acoustic imaging is a measurement technique based on microphone arrays.
With the increasing demand for ocean exploration, the development of acoustic imaging exploration has been dramatically advanced as a key technology in ocean development. The development of the acoustic imaging technology improves the speed and the resolution of underwater target detection, and provides an important technical basis for realizing the detection and the identification of the underwater target. However, due to the complexity of the underwater acoustic environment, the technical level of the noise elimination material, the imaging nonlinearity of the sonar equipment and other limiting conditions, the acquired sonar image has low signal-to-noise ratio and the characteristics of poor target imaging conditions and weak characteristics in the image, and the characteristics cause great difficulty in the work of detection, identification and the like of the target in the sonar image.
An algorithm with a good sonar image segmentation effect is Markov (MRF) random field model segmentation. The Markov random field method is based on MRF model and Bayes theory, it links the uncertainty description with the prior knowledge, and determines the segmented target function according to the optimal criterion in the statistical decision and theoretical estimation by observing the image, and solves the maximum possible distribution of the function meeting these conditions, thus converting the segmentation problem into the optimal problem, the biggest advantage is that the model parameter is less and the space constraint force is strong. The markov random field model emphasizes the spatial constraint relationship of the image, and considers adjacent pixels in the image to be relevant and dependent. System for passing through domain of relationship between any pixel point and other pixel points in imageWhere S is the set of all pixel points of the image, and N isiIs the field of pixel point i, whereinThe pixel point satisfies two conditions: (1) the neighborhood of a pixel does not contain itself i.e.(2) The fields have mutual charactersTherefore, the markov random field can well suppress noise, but for images with poor imaging quality, the uneven intensity is also a big problem, and the pixel-based markov random field is far from enough, so that a region-based markov random field appears. The method takes the region as the minimum unit, better solves the problem of uneven intensity, but inevitably loses the detail information of the image while resisting noise and uneven intensity.
Disclosure of Invention
The purpose of the invention is as follows: aiming at the problem that the traditional single-scale Markov random field model is insufficient in noise and image non-uniform suppression in acoustic imaging detection, the invention aims to provide an acoustic imaging detection method based on a multi-scale Markov random field, which can better segment images with poor imaging quality.
The technical scheme is as follows: an acoustic imaging detection method based on a multi-scale Markov random field comprises the following steps:
(4) constructing a multi-scale Markov random field model of acoustic imaging data under a pixel scale and a region scale;
(5) combining a multi-scale Markov random field model with a level set model to construct a multi-scale Markov random field-level set combined model;
(6) and detecting the target by the convergence of the multi-scale Markov random field-level set combined model to realize acoustic imaging detection.
Further, in the step (1), the following Markov random field model under multiple scales is constructed:
P(X|Y)=P(YP|X)P(YR|X)P(X)
wherein X is the tag field of the acoustic imaging,y is the observation field of the acoustic imagingPIs the observation field at the pixel scale, YRFor field of observation at the regional scale, P (Y)P|X),P(YR| X) is the likelihood probability of the label at the pixel scale and the region scale, respectively; p (x) is the probability distribution of the acoustic imaging tag field;
P(YP| X) is modeled as:
wherein D is YPDimension of (2), YPRepresenting the grey scale, mu, of pixel information for the pixel informationPIs the average value, σ, of the pixel gray scale informationPIs the variance of the pixel gray scale information;
P(YR| X) is modeled as:
wherein D' is YRDimension (d); mu.sRIs the average value, σ, of the region informationRIs the variance of the region information, YRIs area information;
wherein p isRAs a ratio of the area of the divided region to the area of the whole image, NRFor the number of neighborhoods around the partition area, pTThe ratio of the area of the Tth neighborhood to the area of the whole image is set;
p (X) is modeled as:
wherein Z is a normalization constant;s is the image size, S is the pixel point in the image, xsDemarcating x-labeled and s-positioned pixel points, NsIs the neighborhood size of the pixel s, ytIs a label on a pixel with t in the neighborhood of pixel s, V (x)s,yt) Is a potential energy function of the pixel point s and the pixel point t in the neighborhood,beta is a preset constant.
Further, in the step (2), the following multi-scale Markov random field-level set joint model is constructed:
E=EUMRF+Ereg
wherein E isUMRFFor multi-scale Markov random field model energy terms, EregIs a regular term;
EUMRFmodeling is as follows:
wherein the content of the first and second substances,for the level set function, H (-) is the Heaviside function, Xout、XinRespectively the pixels outside and inside the profile of the horizontal set;
Eregmodeling is as follows:
wherein ν and η are weight adjustment coefficients, p is a potential function:and p(s) is equal to or more than p (1) for all s;
in the end of this process,
further, in the step (3), a gradient descent method is adopted to converge the multi-scale Markov random field-level set combined model, and the specific steps are as follows:
wherein t is an evolution time variable,is a regular term gradient function used for controlling the convergence process of the multi-scale Markov random field-level set combined model.
Has the advantages that:
compared with the prior art, the invention has the following remarkable progress: 1. the advantages and the disadvantages of a Markov random field model are integrated, the substantive defects of low resolution and high noise of acoustic imaging information in the detection process are overcome, images with poor imaging quality can be better segmented, and more detailed information is kept in the segmentation result; 2. aiming at the images with high noise and uneven intensity, the problem that the traditional single-scale Markov random field model is insufficient in suppression of noise and image unevenness in acoustic imaging detection is solved.
Drawings
FIG. 1 is a flow chart of a first embodiment of the present invention;
fig. 2 is an original sonar image of a second embodiment of the present invention;
fig. 3 shows the final segmentation result of the second embodiment of the present invention.
Detailed Description
The technical scheme of the invention is further clarified by the following specific embodiments in combination with the attached drawings.
The first embodiment is as follows:
as shown in fig. 1, an acoustic imaging detection method based on a multi-scale markov random field includes the following steps:
the method comprises the following steps: and constructing a multi-scale Markov random field model. The model is as follows:
P(X|Y)=P(YP|X)P(YR|X)P(X)
wherein X is a tag field of acoustic imaging, Y is an observation field of acoustic imaging, and Y isPIs the observation field at the pixel scale, YRFor field of observation at the regional scale, P (Y)P|X),P(YR| X) is the likelihood probability of the label at the pixel scale and the region scale, respectively;
first, likelihood function P (Y) based on pixelP| X) and region-based likelihood function P (Y)RI X), the modeling adopts Gaussian distribution, and the method comprises the following steps:
wherein the content of the first and second substances,
(1) d, D' are each YPAnd YRDimension (d);
(2)YPrepresenting a gray scale of the pixel information for the pixel information; y isRFor region information, defined by the following equation:
wherein p isRAs a ratio of the area of the divided region to the area of the whole image, NRAs a region near the divided region, pTIs the proportion of the area of one region T in the field to the whole image area.
(3) μ and σ are the mean and standard deviation of different regions of the image, divided according to level set region. Mu.sPAnd σPIs the average value of the pixel characteristics, muPAnd σPIs the average of the regional features.
WhereinRespectively the mean values of the inside and outside of the segmentation region under the pixel scale,respectively the variance inside and outside the segmentation region at the pixel scale, whereinAre respectively the average values inside and outside the segmentation region under the region scale,the variance of the inside and outside of the segmentation region under the region scale is respectively, mean { } is a mean function, and default { } is a variance function.
Secondly, according to Hammersley-Clifford theorem, a one-to-one correspondence relationship exists between the Markov random field and the Gibbs random field, namely if the random field F has Markov property, the random field F has Gibbs distribution; conversely, if the random field F has a Gibbs distribution, then the random field F has a Markov property. Thus, p (x) modeling employs a gibbs distribution:
wherein Z is a normalization constant;
(2) U (x) is a function of energy:
s is an input image of size NxM, S ═ SiI ═ 1,2, L, nxm }, s is a pixel point in the image, x issDemarcating x-labeled and s-positioned pixel points, NsIs the neighborhood size of the pixel s, ytThe labels on the pixel points with the position t in the neighborhood of the pixel points s; v (x)s,yt) Is a potential energy function of the pixel point s and the pixel point t in the neighborhood,β is a predetermined constant, typically 1.
Step two: and establishing a multi-scale Markov random field-level set combined model.
Combining a multi-scale markov random field model with a level set model:
E=EUMRF+Ereg
wherein E isUMRFFor multi-scale Markov random field model energy terms, EregIs a regular term;
EUMRFmodeling is as follows:
wherein the content of the first and second substances,for the level set function, H (-) is the Heaviside function, Xout,XinPixels outside and inside the horizontal set contour;
Eregmodeling is as follows:
wherein ν and η are weight adjustment coefficients, p is a potential function:and p(s) is equal to or more than p (1) for all s;
finally, modeling the multi-scale Markov random field-level set combined model as follows:
step three: and imaging detection is realized through convergence of a multi-scale Markov random field-level set combined model.
Adopting a gradient descent method to perform convergence calculation on the multi-scale Markov random field-level set combined model:
wherein t is an evolution time variable,the method is a regular term gradient function and is used for controlling the convergence process of the multi-scale Markov random field-level set combined model.
Example two:
taking the sonar image as an example, referring to fig. 2, the pixel feature Y in the sonar image is first extractedPAnd region feature YR:
YPProcessing is not needed, and the gray value of the sonar image is directly obtained;
wherein p isRAs a ratio of the area of the divided region to the area of the whole image, NRAs a region near the divided region, pTIs the proportion of the area of one region T in the field to the whole image area. The initial segmentation roughly divides the sonar image in an over-segmentation mode.
And respectively substituting the extracted features into likelihood functions to obtain:
where D is the dimension of Y, σ is the standard deviation of Y, and μ is the mean of Y.
The prior probability adopts Gibbs distribution:
Since the sonar image is divided into three regions, namely a target region, a highlight region and a background region, the level set needs to adopt a two-phase level set. Finally, the multi-scale Markov random field-level set combined model for image segmentation is constructed as follows:
wherein phi is1And phi2Is a level set function; h (-) is the Heaviside function; eregThe regularization term is used for controlling the regularization of the convergence of the level set function; v and mu are weight coefficients; e (X)1|Y)、E(X2|Y)、E(X3|Y)、E(X4Y) are respectively a target area, a shadow area, a background area and other four areas in the sonar image,andis a regular term.
Therefore, the multi-scale Markov random field-level set combined model provided by the invention is obtained.
And finally, carrying out convergence calculation on the multi-scale Markov random field-level set combined model by adopting a gradient descent method:
according to the iterative equation phik+1=φk+ΔφkAnd obtaining a segmentation result after multiple iterations of Δ t.
Thus, an energy function evolution process is completed, and sonar image segmentation is realized, as shown in fig. 3.
Claims (5)
1. An acoustic imaging detection method based on a multi-scale Markov random field is characterized by comprising the following steps:
(1) constructing a multi-scale Markov random field model of acoustic imaging data under a pixel scale and a region scale;
(2) combining a multi-scale Markov random field model with a level set model to construct a multi-scale Markov random field-level set combined model;
(3) and detecting the target by the convergence of the multi-scale Markov random field-level set combined model to realize acoustic imaging detection.
2. The method for acoustic imaging detection based on multi-scale markov random fields according to claim 1, wherein in step (1), the following multi-scale markov random field model is constructed:
P(X|Y)=P(YP|X)P(YR|X)P(X)
wherein X is a tag field of acoustic imaging, Y is an observation field of acoustic imaging, and Y isPIs the observation field at the pixel scale, YRFor field of observation at the regional scale, P (Y)P|X),P(YR| X) is the likelihood probability of the label at the pixel scale and the region scale, respectively; p (x) is the probability distribution of the acoustic imaging tag field;
P(YP| X) is modeled as:
wherein D is YPDimension of (2), YPRepresenting the grey scale, mu, of pixel information for the pixel informationPIs the average value, σ, of the pixel gray scale informationPIs the variance of the pixel gray scale information;
P(YR| X) is modeled as:
wherein D' is YRDimension (d); mu.sRIs the average value, σ, of the region informationRIs the variance of the region information, YRIs area information;
wherein p isRAs a ratio of the area of the divided region to the area of the whole image, NRFor the number of neighborhoods around the partition area, pTThe ratio of the area of the Tth neighborhood to the area of the whole image is set;
p (X) is modeled as:
wherein Z is a normalization constant;s is the image size, S is the pixel point in the image, xsDemarcating x-labeled and s-positioned pixel points, NsIs the neighborhood size of the pixel s, ytIs a label on a pixel with t in the neighborhood of pixel s, V (x)s,yt) Is a potential energy function of the pixel point s and the pixel point t in the neighborhood,beta is a preset constant.
3. The method for acoustic imaging detection based on multi-scale markov random field according to claim 2, wherein in step (2), the following multi-scale markov random field-level set joint model is constructed:
E=EUMRF+Ereg
wherein E isUMRFFor multi-scale Markov random field model energy terms, EregIs a regular term;
EUMRFmodeling is as follows:
wherein the content of the first and second substances,for the level set function, H (-), is the Heaviside function,Xout、Xinrespectively the pixels outside and inside the profile of the horizontal set;
Eregmodeling is as follows:
wherein ν and η are weight adjustment coefficients, p is a potential function:and p(s) is equal to or more than p (1) for all s;
in the end of this process,
4. the method for acoustic imaging detection based on multi-scale Markov random fields as claimed in claim 3, wherein in step (3), the multi-scale Markov random field-level set joint model is converged by gradient descent.
5. The method for detecting acoustic imaging based on multi-scale Markov random field according to claim 4, wherein the step of converging the multi-scale Markov random field-level set combination model by gradient descent comprises the following steps:
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110258326.4A CN113093198B (en) | 2021-03-10 | 2021-03-10 | Acoustic imaging detection method based on multi-scale Markov random field |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110258326.4A CN113093198B (en) | 2021-03-10 | 2021-03-10 | Acoustic imaging detection method based on multi-scale Markov random field |
Publications (2)
Publication Number | Publication Date |
---|---|
CN113093198A true CN113093198A (en) | 2021-07-09 |
CN113093198B CN113093198B (en) | 2023-07-25 |
Family
ID=76666684
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202110258326.4A Active CN113093198B (en) | 2021-03-10 | 2021-03-10 | Acoustic imaging detection method based on multi-scale Markov random field |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN113093198B (en) |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2013041417A1 (en) * | 2011-09-23 | 2013-03-28 | Atlas Elektronik Gmbh | Method and device for extracting contours from sonar images |
CN108460773A (en) * | 2018-02-28 | 2018-08-28 | 哈尔滨工程大学 | A kind of sonar image dividing method based on biased field level set |
CN109242876A (en) * | 2018-09-10 | 2019-01-18 | 电子科技大学 | A kind of image segmentation algorithm based on markov random file |
CN111598890A (en) * | 2020-05-15 | 2020-08-28 | 河海大学 | Level set optimization method for underwater image segmentation |
-
2021
- 2021-03-10 CN CN202110258326.4A patent/CN113093198B/en active Active
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2013041417A1 (en) * | 2011-09-23 | 2013-03-28 | Atlas Elektronik Gmbh | Method and device for extracting contours from sonar images |
CN108460773A (en) * | 2018-02-28 | 2018-08-28 | 哈尔滨工程大学 | A kind of sonar image dividing method based on biased field level set |
CN109242876A (en) * | 2018-09-10 | 2019-01-18 | 电子科技大学 | A kind of image segmentation algorithm based on markov random file |
CN111598890A (en) * | 2020-05-15 | 2020-08-28 | 河海大学 | Level set optimization method for underwater image segmentation |
Non-Patent Citations (2)
Title |
---|
OMRAN SALIH 等: "Skin Lesion Segmentation Techniques Based on Markov Random Field" * |
张丽丽 等: "基于马尔可夫随机场的水下声呐图像目标检测方法" * |
Also Published As
Publication number | Publication date |
---|---|
CN113093198B (en) | 2023-07-25 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN104008538B (en) | Based on single image super-resolution method | |
CN109949349B (en) | Multi-mode three-dimensional image registration and fusion display method | |
Hosotani et al. | Image denoising with edge-preserving and segmentation based on mask NHA | |
CN113177592B (en) | Image segmentation method and device, computer equipment and storage medium | |
Pandey et al. | Segmentation of liver lesions with reduced complexity deep models | |
CN112634149A (en) | Point cloud denoising method based on graph convolution network | |
Dinesh et al. | 3D point cloud color denoising using convex graph-signal smoothness priors | |
Selvathi et al. | Phase based distance regularized level set for the segmentation of ultrasound kidney images | |
CN109741358B (en) | Superpixel segmentation method based on adaptive hypergraph learning | |
CN114119924A (en) | Three-dimensional model fuzzy texture feature saliency method based on condition generation countermeasure | |
CN111028241B (en) | Multi-scale blood vessel enhanced level set segmentation system and method | |
CN113093198A (en) | Acoustic imaging detection method based on multi-scale Markov random field | |
Skalski et al. | Left ventricle USG image segmentation using active contour model | |
Kapuriya et al. | Geometric constraints based selective segmentation for vector valued images | |
Murtaza et al. | A robust local model for segmentation based on coefficient of variation | |
CN109785331B (en) | Sonar image segmentation method based on self-adaptive pixel value constraint and MRF | |
CN113408615A (en) | Automatic ship matching method based on optical satellite remote sensing image | |
CN107292854B (en) | Gray level image enhancement method based on local singularity quantitative analysis | |
Zheng et al. | Selection of regularization parameter in GMM based image denoising method | |
Malarvel et al. | An Improved Kernel Density Estimation with adaptive bandwidth selection for Edge detection | |
CN112102334B (en) | Test data generation method based on segmentation variation | |
Ye et al. | Sonar image segmentation on Fuzzy C-Mean using local texture feature | |
Chen et al. | Research on Lossless Compression of Hyperspectral Images Based on Improved Deep Learning Algorithm | |
Yingjie et al. | Region-based image fusion using energy estimation | |
Li | Edge color difference detection of color image based on artificial intelligence technology |
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 |