CN108171193B - Polarized SAR (synthetic aperture radar) ship target detection method based on super-pixel local information measurement - Google Patents
Polarized SAR (synthetic aperture radar) ship target detection method based on super-pixel local information measurement Download PDFInfo
- Publication number
- CN108171193B CN108171193B CN201810013991.5A CN201810013991A CN108171193B CN 108171193 B CN108171193 B CN 108171193B CN 201810013991 A CN201810013991 A CN 201810013991A CN 108171193 B CN108171193 B CN 108171193B
- Authority
- CN
- China
- Prior art keywords
- sliding window
- superpixel
- pixel
- center
- super
- 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
- 238000001514 detection method Methods 0.000 title claims abstract description 45
- 238000005259 measurement Methods 0.000 title claims abstract description 26
- 238000000034 method Methods 0.000 claims abstract description 44
- 230000011218 segmentation Effects 0.000 claims abstract description 37
- 239000013598 vector Substances 0.000 claims abstract description 32
- 238000013507 mapping Methods 0.000 claims abstract description 5
- 239000011159 matrix material Substances 0.000 claims description 47
- 238000000354 decomposition reaction Methods 0.000 claims description 32
- 238000011524 similarity measure Methods 0.000 claims description 11
- 238000003064 k means clustering Methods 0.000 claims description 5
- 238000012706 support-vector machine Methods 0.000 claims description 5
- OAICVXFJPJFONN-UHFFFAOYSA-N Phosphorus Chemical compound [P] OAICVXFJPJFONN-UHFFFAOYSA-N 0.000 claims description 4
- 230000004927 fusion Effects 0.000 claims description 3
- 239000004576 sand Substances 0.000 claims description 3
- 238000002474 experimental method Methods 0.000 description 15
- 230000010287 polarization Effects 0.000 description 11
- 238000012360 testing method Methods 0.000 description 5
- 238000007796 conventional method Methods 0.000 description 3
- 230000002087 whitening effect Effects 0.000 description 3
- 230000001427 coherent effect Effects 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 238000012549 training Methods 0.000 description 2
- 238000012935 Averaging Methods 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 230000002708 enhancing effect Effects 0.000 description 1
- 238000010206 sensitivity analysis Methods 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/10—Terrestrial scenes
- G06V20/13—Satellite images
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/26—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V2201/00—Indexing scheme relating to image or video recognition or understanding
- G06V2201/07—Target detection
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Theoretical Computer Science (AREA)
- Astronomy & Astrophysics (AREA)
- Remote Sensing (AREA)
- Radar Systems Or Details Thereof (AREA)
Abstract
The invention discloses a polarized SAR (synthetic aperture radar) ship target detection method based on superpixel local information measurement, which mainly solves the problem of low target detection rate in a complex scene, and has the scheme that: 1. performing superpixel segmentation on an original image to obtain superpixel segmentation results under different scales; 2. calculating three super-pixel-level-based difference metrics by utilizing a sliding window model for the segmented result; 3. converting the super-pixel level based dissimilarity measure into a pixel level based dissimilarity measure; 4. mapping the difference measurement vector of the pixel level into a difference measurement value by utilizing kernel fisher judgment to obtain the difference measurement value of each pixel point; 5. and classifying the difference metric value of each pixel point by using a linear SVM classifier, determining the category of each pixel point, and performing automatic target detection. The method improves the target detection performance in a complex scene, realizes the automatic detection process, and can be used for subsequent ship target identification, identification and classification.
Description
Technical Field
The invention belongs to the technical field of radar target detection, and mainly relates to a polarized SAR ship target detection method which can be used for subsequent ship target identification, identification and classification.
Background
The synthetic aperture radar SAR utilizes a microwave remote sensing technology, is not influenced by weather and day and night, has all-weather and all-day working capability, and has the characteristics of multiple frequency bands, multiple polarization, variable visual angle, penetrability and the like. At present, the SAR is widely applied to the fields of military reconnaissance, geological survey, topographic mapping and charting, disaster prediction, marine application, scientific research and the like, and has wide research and application prospects. Due to the remarkable advantage of being able to obtain complete polarization information, polarization SAR is rapidly becoming one of the important directions for SAR development. The ship target detection based on the polarized SAR image is an important application field of the polarized SAR.
To date, many methods have been proposed to achieve target detection using polarized SAR data, such as polarized whitening filters, polarized notch filters, and reflective symmetric filters. In addition, some polarization parameters and discrimination features are proposed to enhance the difference between the potential target and the local clutter. Polarization entropy and polarizability have also been used for ship target detection, for example. Recently, a ship target detection method based on the scattering mechanism distribution characteristics of the superpixel and the local scattering mechanism difference of the regression kernel is also proposed. These methods, while enhancing ship and sea surface contrast to some extent, also have some potential to be affected by complex sea state and signal-to-clutter ratio changes.
Since the existing detection methods are basically unsupervised detection methods, the determination of the detection threshold is an important task. There are three main ways to determine the detection threshold. The first approach is to apply a constant false alarm rate to the polarization statistics to calculate a detection threshold, which depends largely on the accuracy of clutter statistical modeling and parameter estimation. The second method, based on sensitivity analysis of certain parameters, selects detection thresholds empirically, which is inconvenient for different polarization systems. The third method is to determine the detection threshold by using a clustering method, but the clustering process of this method must be performed in a local area, and the detection threshold determined by this method may cause a lot of false alarms to be detected subsequently when no ship exists.
The above conventional method has the following two main disadvantages: firstly, the ship is easily influenced by complex sea conditions and signal-to-clutter ratio changes, so that the contrast between ships and sea clutter is reduced; secondly, it is difficult to obtain an accurate detection threshold, which brings great inconvenience to the polarized target detection system.
Disclosure of Invention
The invention aims to provide a polarized SAR ship target detection method based on superpixel local information measurement aiming at the defects of the existing polarized SAR ship target detection method, so as to enhance the contrast of ships and sea clutter, realize automatic target detection and improve the detection performance under complex conditions.
In order to achieve the purpose, the technical scheme of the invention comprises the following steps:
(1) respectively carrying out multi-scale superpixel segmentation on a given polarized SAR image I to obtain superpixel segmentation results of the polarized SAR image under 4 scales: s1,S2,S3,S4;
(2) For the result S after super pixel segmentationkAnd respectively calculating three super-pixel-level-based difference metrics by using a super-pixel sliding window model: likelihood ratio metricRiemann distance metricScatter component similarity metricWherein k 1., 4, represents 4 scales, representing the current scale segmentation result SkJ is 1, Mi,MiRepresenting by super-pixelsThe number of superpixels on the sliding window boundary which is the center;
(3) converting the three super-pixel-level-based disparity metrics in (2) into pixel-level-based disparity metric vectorsObtaining a difference measurement vector D of each pixel point after fusion under different segmentation scaless:
Wherein,super-pixel segmentation result S for each pixel pointkThe difference measurement vector s is 1, the other words, and P is the number of all pixel points in the polarized SAR image I;
(4) selecting a Gaussian kernel function, and fusing the difference measurement vector D obtained in the step (3) under different segmentation scales by using kernel fisher discriminant analysis algorithm KFDAsMapping into the final difference metric Dis of each pixel points;
(5) Utilizing a support vector machine classifier SVM to carry out final difference metric Dis of each pixel point mapped in the step (4)sAnd classifying, outputting the category of each pixel point, and realizing automatic target detection.
The invention has the following advantages:
1. because the difference measurement of three super-pixel levels is calculated and the difference of the super-pixel levels is converted into the difference measurement of the pixel levels, compared with the traditional polarimetric SAR image ship target detection method, the method not only reduces the influence of coherent speckles, but also enhances the contrast of ships and sea clutter;
2. the invention adopts a supervision and classification method to carry out automatic target detection, thereby ensuring good detection performance under complex conditions.
Drawings
FIG. 1 is a flow chart of an implementation of the present invention;
FIG. 2 is a measured polarimetric SAR image used in experiment 1 of the present invention;
FIG. 3 is a measured polarimetric SAR image used in experiment 2 of the present invention;
FIG. 4 is a detection result image of the R5 region and the R5 region in FIG. 2 under each detection algorithm;
fig. 5 is a detection result image of the R6 region and the R6 region in fig. 3 under each detection algorithm.
Detailed Description
The embodiments and effects of the present invention will be further described in detail with reference to the accompanying drawings:
referring to fig. 1, the implementation steps of the present invention include the following:
step 1, respectively carrying out multi-scale superpixel segmentation on a given polarized SAR image I.
The superpixel segmentation algorithm comprises the following steps: the method comprises the following steps of performing super-pixel segmentation on a polarized SAR image I by using the improved SLIC algorithm according to different scales according to the SLIC algorithm, the Turbo pixel algorithm, the Normalized-cuts algorithm, the Y.Wang and the like, wherein the improved SLIC algorithm is provided for the polarized SAR image I, and the method is realized as follows:
when the scale is 6, performing superpixel segmentation on the given polarized SAR image I by using an improved SLIC algorithm to obtain a segmented result S1;
When the scale is 9, performing superpixel segmentation on the given polarized SAR image I by using an improved SLIC algorithm to obtain a segmented result S2;
When the scale is 12, performing superpixel segmentation on the given polarized SAR image I by using an improved SLIC algorithm to obtain a segmented result S3;
When the scale is 15, the given polarized SAR image I is superpixed by using a modified SLIC algorithmDividing to obtain divided result S4。
Step 2, dividing the result S of the super pixelkThree superpixel-level-based dissimilarity measures are respectively calculated by using a superpixel sliding window model, wherein k is 1.
The three superpixel-level-based dissimilarity metrics include a likelihood ratio metricRiemann distance metricScatter component similarity metricWherein Representing the current scale segmentation result SkJ is 1, Mi,MiRepresenting by super-pixelsThe number of superpixels on the sliding window boundary which is the center;
the three superpixel-based disparity metrics are calculated as follows:
2a) segmenting the result S at each scalekNext, a likelihood ratio metric based on the superpixel level is calculated
2a1) At the super-pixel segmentation result SkIn the method, a superpixel sliding window model is used for calculating the superpixel in the center of a sliding windowAnd superpixels on sliding window boundaryLikelihood ratio statistics
Where L (-) is a likelihood function, N is a view, NiRepresenting a superpixelNumber of pixels in, NjRepresenting a superpixelNumber of pixels in (1), parameter Andis an intermediate variable, estimated by the following formula:
wherein,representing center superpixel of sliding windowThe first ofiA coherence matrix of the individual pixels of the image,representing superpixels on sliding window boundariesThe first ofjA coherence matrix of the individual pixels of the image,to representAndthe coherence matrix of the ith pixel in (a);
2a2) detecting statistics from likelihood ratios calculated in step 2a1)Calculating center superpixel of sliding windowAnd superpixels on sliding window boundaryLikelihood ratio metric of
Wherein, lnQcIs a constant parameter with the value of-150;
2b) segmenting the result S at each scalekNext, a super-pixel level based Riemann distance metric is calculated
2b1) Respectively calculating the center superpixel of the sliding window by using the following formulaEquivalent coherence matrixAnd superpixels at the sliding window boundaryEquivalent coherence matrix
Wherein p represents the center superpixel of the sliding windowPixel of (2), NiTo representThe number of pixels in (1) | · non-woven phosphorFDenotes the F norm, TpTo representQ represents a superpixel at the boundary of the sliding windowPixel of (2), NjTo representNumber of pixels in, TqTo representA coherence matrix of the pixel q in (a);
2b2) at the super-pixel segmentation result SkIn the method, a superpixel sliding window model is used for calculating the superpixel in the center of a sliding windowAnd superpixels on sliding window boundaryRiemann distance statistic
Wherein tr (-) represents the trace of the matrix;
2b3) from the Riemann distance statistic obtained in step 2b2)Calculating center superpixel of sliding windowAnd superpixels on sliding window boundaryRiemann distance measurement
Wherein h is a parameter having a value of 0.5;
2c) segmenting the result S at each scalekNext, a scatter component similarity metric based on the superpixel level is calculated
representing center superpixel of sliding windowCoherence matrixThe surface of (a) scatters the power,
representing center superpixel of sliding windowCoherence matrixThe secondary scattered power of (a) is,
representing center superpixel of sliding windowCoherence matrixThe volume of (a) is used to scatter power,
wherein, | - | represents the absolute value, β1Is a center superpixel of a sliding windowCoherence matrixSurface scattering parameter, alpha, obtained by a polarized target decomposition algorithm1Is a center superpixel of a sliding windowCoherence matrixThe secondary scattering parameters obtained by the polarized target decomposition algorithm,is a center superpixel of a sliding windowCoherence matrixThe surface scattering decomposition coefficient obtained by the polarized target decomposition algorithm,is a center superpixel of a sliding windowCoherence matrixThe secondary scattering decomposition coefficient obtained by the polarized target decomposition algorithm,is a center superpixel of a sliding windowCoherence matrixObtaining a volume scattering decomposition coefficient through a polarized target decomposition algorithm;
representing sliding window boundary superpixelsCoherence matrixThe surface of (a) scatters the power,
representing sliding window boundary superpixelsCoherence matrixThe secondary scattered power of (a) is,
representing sliding window boundary superpixelsCoherence matrixThe volume of (a) is used to scatter power,
wherein, beta2Is a sliding window boundary superpixelCoherence matrixSurface scattering parameter, alpha, obtained by a polarized target decomposition algorithm2Is a sliding window boundary superpixelCoherence matrixThe secondary scattering parameters obtained by the polarized target decomposition algorithm,superpixels for sliding window boundariesCoherence matrixThe surface scattering decomposition coefficient obtained by the polarized target decomposition algorithm,superpixels for sliding window boundariesCoherence matrixThe secondary scattering decomposition coefficient obtained by the polarized target decomposition algorithm,superpixels for sliding window boundariesCoherence matrixObtaining a volume scattering decomposition coefficient through a polarized target decomposition algorithm;
2c3) at the super-pixel segmentation result SkIn the method, a superpixel sliding window model is used for calculating the superpixel in the center of a sliding windowAnd sliding window boundary superpixelScattering power vector similarity parameter r ofij:
Wherein the superscript H is a conjugate transpose, | · |. non-woven phosphor2Is the norm of L2, |, represents the absolute value;
2c4) according to the scattering power vector similarity parameter r obtained in the step 2c3)ijCalculating the center superpixel of the sliding windowAnd sliding window boundary superpixelScatter component similarity measure ofThe value is calculated by the following formula:
step 3, converting the three types of difference metrics based on the super-pixel level in the step (2) into difference metric vectors based on the pixel levelObtaining a difference measurement vector D of each pixel point after fusion under different segmentation scaless;
3a) Performing superpixel on the center of the sliding window by using a K-means clustering algorithmAnd sliding window boundary superpixelLikelihood ratio metric ofj=1,...,MiClustering into two categories, using the mean of likelihood ratio metrics in the majority of categories as the center superpixel of the sliding windowFinal likelihood ratio metric
Wherein,representing a value of a superpixel likelihood ratio metric, c, in a current majority class1=1,...,ML,MLRepresenting likelihood ratio metric in current clusterThe number of superpixels of the majority class;
3b) performing superpixel on the center of the sliding window by using a K-means clustering algorithmAnd sliding window boundary superpixelRiemann distance measurementClustering into two classes, and taking the mean of Riemann distance measures in the majority of classes as the center superpixel of the sliding windowFinal Riemann distance metric
Wherein,representing a super-pixel Riemann distance metric, c, in a current majority class2=1,...,MR,MRRepresenting the number of the superpixels of most types of the Riemann distance metric value in the current cluster;
3c) performing superpixel on the center of the sliding window by using a K-means clustering algorithmAnd sliding window boundary superpixelScatter component similarity measure ofCluster toIn both categories, the mean of the scatter component similarity measures in the majority of classes is taken as the sliding window center superpixelFinal scatter component similarity measure
Wherein,representing a similarity measure, c, of the superpixel scatter components in the current majority class3=1,...,MC,MCRepresenting the number of superpixels of a majority of classes of the similarity measurement of the scattering components in the current cluster;
3d) center superpixel of sliding windowFinal likelihood ratio metricRiemann distance metricScatter component similarity metricConstituting a vector as a center superpixel of the sliding windowOf a vector of dissimilarity measures
3e) Center superpixel of sliding windowOf a vector of dissimilarity measuresIs distributed to the pixel inside to obtain the difference metric vector of the pixel pWhereinRepresenting a superpixelA pixel of (1);
3f) repeating the steps 3a) to 3e) through a sliding window model to obtain a difference measurement vector of the pixel point in the superpixel at the center of each sliding windows 1, P represents the number of pixel points in the polarized SAR image I;
3g) segmenting the different superpixel segmentation results SkAveraging the difference metric of each pixel to obtain a final difference metric vector of each pixel s:
step 4, selecting a Gaussian kernel function, and fusing the difference measurement vector D obtained in the step 3 under different segmentation scales by using a kernel fisher discriminant analysis algorithm KFDAsMapping into the final difference metric Dis of each pixel points。
Step 5, utilizing a linear support vector machine classifier SVM to carry out final treatment on each pixel point mapped in the step (4)Difference metric DissAnd classifying, outputting the category of each pixel point, and realizing automatic target detection.
The effects of the present invention can be further illustrated by the following experimental data:
experiment 1:
1.1) experimental scenario:
the data used in this experiment was a C-band RadarSat-2 fully polarized dataset with a resolution of 12 m × 8 m, i.e., range direction × azimuth direction, obtained in tokyo bay on 8/4/2010, and an angle of incidence of 35 degrees, as shown in fig. 2.
The training data is the R4 region image in fig. 2, the test data is the R5 region image in fig. 2 as shown in fig. 4(a), there are a total of 38 potential targets in fig. 4(a), the strong targets are circled by rectangles, and the weak targets are circled by circles.
1.2) experimental parameters:
the penalty coefficient V of the linear support vector machine SVM is 50, and the kernel parameter g in the kernel fisher discriminant analysis algorithm KFDA is 2.
1.3) contents of the experiment:
FIG. 4(a) was tested according to the present invention, and the results are shown in FIG. 4 (b);
the experiment of fig. 4(a) was performed using a conventional polarization whitening filter detector PWF, and the result is shown in fig. 4 (c);
fig. 4(a) was subjected to an experiment using a conventional polarization notch filter PNF, and the result is shown in fig. 4 (d);
the experiment of FIG. 4(a) was performed using a conventional reflection symmetric filter RSF, and the result is shown in FIG. 4 (e);
fig. 4(a) was tested with the existing significance-based detector SD-LSMDRK, and the results are shown in fig. 4 (f);
fig. 4(a) was tested with a prior art detector SPD based on the distribution characteristics of the superpixel scattering mechanism, and the results are shown in fig. 4 (g).
As can be seen from fig. 4(b) -4(g), the method of the present invention detected all targets and had few false alarms, which demonstrates the effectiveness of the method for suppressing sea clutter. For the traditional methods PWF, PNF and RSF, more false alarms exist; the PWF detects fewer target pixels than the method of the present invention, the PNF does not detect target 3 in fig. 4(a), and the RSF does not detect targets 1-4 in fig. 4 (a); although the SD-LSMDRK and the SPD have similar detection performance to the method, the method can automatically determine the threshold value, and the two methods cannot automatically determine the threshold value.
The results of the above test 1 are shown in table 1:
TABLE 1 test results of different methods
Different methods | The invention | PWF | PNF | RSF | SD-LSMDRK | SPD |
Ntd | 38 | 38 | 37 | 34 | 38 | 38 |
Nfa | 6 | >6 | 12 | >6 | 5 | 12 |
N in Table 1tdIndicating the number of detected objects, NfaIndicating the number of false objects.
As can be seen from Table 1, the present invention detects all the targets, and the number of the false targets is also small, so that compared with the conventional method, the present invention improves the detection performance of the algorithm to a certain extent.
Experiment 2:
2.1) experimental scenario:
the data used in this experiment was a C-band RadarSat-2 fully polarized dataset with a resolution of 12 meters × 8 meters, i.e., distance direction × azimuth direction, obtained at 23 days 6/2011 in chinese staving harbor at an angle of incidence of 30 degrees. As shown in fig. 3.
The training data is the R7 region image in fig. 3, the test data is the R6 region image in fig. 3 as shown in fig. 5(a), a total of 46 potential targets are shown in fig. 5(a), the strong targets are circled by rectangles, and the weak targets are circled by circles.
2.2) experimental parameters:
the penalty coefficient V of the linear support vector machine SVM is 50, and the kernel parameter g in the kernel fisher discriminant analysis algorithm KFDA is 2.
2.3) contents of the experiment:
FIG. 5(a) was tested according to the present invention, and the results are shown in FIG. 5 (b);
the experiment of fig. 5(a) was performed using a conventional polarization whitening filter detector PWF, and the result is shown in fig. 5 (c);
fig. 5(a) was subjected to an experiment using a conventional polarization notch filter PNF, and the result is shown in fig. 5 (d);
the experiment of FIG. 5(a) with the conventional reflection symmetric filter RSF shows the result of FIG. 5 (e);
FIG. 5(a) was tested with the existing significance-based detector SD-LSMDRK, and the results are shown in FIG. 5 (f);
fig. 5(a) was tested with a prior art detector SPD based on the distribution characteristics of the superpixel scattering mechanism, and the results are shown in fig. 5 (g).
As can be seen from fig. 5(b) -5(g), the method of the present invention detected all targets and had few false alarms, which demonstrates the effectiveness of the method for suppressing sea clutter. And more false alarms exist for the traditional methods PWF, PNF, RSF, SD-LSMDRK and SPD. The number of target pixels detected by PWF, PNF, RSF, SD-LSMDRK and SPD is less than that of the target pixels detected by the method, and the detection performance of the algorithm is improved to a certain extent.
The results of the above experiment 2 are shown in table 2:
TABLE 2 test results of different methods
Different methods | The invention | PWF | PNF | RSF | SD-LSMDRK | SPD |
Ntd | 46 | 45 | 43 | 42 | 45 | 42 |
Nfa | 3 | >3 | >3 | >3 | 5 | 4 |
As can be seen from table 2, only the present invention detects all the targets, and the number of the false targets is also the least, so compared with the conventional method, the present invention improves the detection performance of the algorithm to a certain extent.
In conclusion, the polarized SAR ship target detection method based on the superpixel local information measurement enhances the contrast ratio of ships and sea clutter, inhibits the influence of coherent spots, increases the difference between the ships and the sea clutter, and utilizes supervision and classification to perform automatic target detection, thereby improving the performance of a ship detection algorithm.
Claims (5)
1. A polarized SAR ship target detection method based on super-pixel local information measurement comprises the following steps:
(1) respectively carrying out multi-scale superpixel segmentation on a given polarized SAR image I to obtain superpixel segmentation results of the polarized SAR image under 4 scales: s1,S2,S3,S4;
(2) For the result S after super pixel segmentationkAnd respectively calculating three super-pixel-level-based difference metrics by using a super-pixel sliding window model: likelihood ratio metricRiemann distance metricScatter component similarity metricWherein k 1., 4, represents 4 scales, representing the current scale segmentation result SkJ is 1, Mi,MiRepresenting by super-pixelsThe number of superpixels on the sliding window boundary which is the center; wherein solving likelihood ratio metrics based on superpixel levelThe method comprises the following steps:
2a1) at the super-pixel segmentation result SkIn the method, a superpixel sliding window model is used for calculating the superpixel in the center of a sliding windowAnd superpixels on sliding window boundaryLikelihood ratio statistics
Where L (-) is a likelihood function, N is a view, NiRepresenting a super imageVegetable extractNumber of pixels in, NjRepresenting a superpixelNumber of pixels in (1), parameterAndis an intermediate variable, estimated by the following formula:
wherein,representing center superpixel of sliding windowThe first ofiA coherence matrix of the individual pixels of the image,representing superpixels on sliding window boundariesThe first ofjA coherence matrix of the individual pixels of the image,to representAndthe coherence matrix of the ith pixel in (a);
2a2) detecting statistics from likelihood ratios calculated in step 2a1)Calculating center superpixel of sliding windowAnd superpixels on sliding window boundaryLikelihood ratio metric of
Wherein, lnQcIs a constant parameter with the value of-150;
(3) converting the three super-pixel-level-based disparity metrics in (2) into pixel-level-based disparity metric vectorsObtaining a difference measurement vector D of each pixel point after fusion under different segmentation scaless:
Wherein,super-pixel segmentation result S for each pixel pointkThe vector of dissimilarity measures of (1),.., wherein P is the number of all pixel points in the polarized SAR image I;
(4) selecting a Gaussian kernel function, and fusing the difference measurement vector D obtained in the step (3) under different segmentation scales by using kernel fisher discriminant analysis algorithm KFDAsMapping into the final difference metric Dis of each pixel points;
(5) Utilizing a support vector machine classifier SVM to carry out final difference metric Dis of each pixel point mapped in the step (4)sAnd classifying, outputting the category of each pixel point, and realizing automatic target detection.
2. The method of claim 1, wherein step (1) performs multi-scale superpixel segmentation as follows;
the superpixel segmentation algorithm comprises the following steps: the method comprises the following steps that according to different scales, an SLIC algorithm, a Turbo pixel algorithm, a Normalized-cuts algorithm and an improved SLIC algorithm for the polarized SAR image, which is proposed by Y.Wang, the improved SLIC algorithm is used for carrying out superpixel segmentation on the polarized SAR image I, and the method is realized as follows:
when the scale is 6, performing superpixel segmentation on the given polarized SAR image I by using an improved SLIC algorithm to obtain a segmented result S1;
When the scale is 9, performing superpixel segmentation on the given polarized SAR image I by using an improved SLIC algorithm to obtain a segmented result S2;
When the scale is 12, performing superpixel segmentation on the given polarized SAR image I by using an improved SLIC algorithm to obtain a segmented result S3;
When the scale is 15, performing superpixel segmentation on the given polarized SAR image I by using an improved SLIC algorithm to obtain a segmented result S4。
3. The method of claim 1, wherein step 2) finds a super-pixel level based Riemann distance metricThe method comprises the following steps of;
2b1) respectively calculating the center superpixel of the sliding window by using the following formulaEquivalent coherence matrixAnd superpixels at the sliding window boundaryEquivalent coherence matrix
Wherein p represents the center superpixel of the sliding windowPixel of (2), NiTo representThe number of pixels in (1) | · non-woven phosphorFDenotes the F norm, TpTo representQ represents a superpixel at the boundary of the sliding windowPixel of (2), NjTo representNumber of pixels in, TqTo representA coherence matrix of the pixel q in (a);
2b2) at the super-pixel segmentation result SkIn the method, a superpixel sliding window model is used for calculating the superpixel in the center of a sliding windowAnd superpixels on sliding window boundaryRiemann distance statistic
Wherein tr (-) represents the trace of the matrix;
2b3) from the Riemann distance statistic obtained in step 2b2)Calculating center superpixel of sliding windowAnd superpixels on sliding window boundaryRiemann distance measurement
Where h is a parameter with a value of 0.5.
4. The method of claim 1, wherein step 2) evaluates a scatter component similarity metric based on superpixel levelsThe method comprises the following steps of;
representing center superpixel of sliding windowCoherence matrixThe surface of (a) scatters the power,
representing center superpixel of sliding windowCoherence matrixThe secondary scattered power of (a) is,
representing center superpixel of sliding windowCoherence matrixThe volume of (a) is used to scatter power,
wherein, | - | represents the absolute value, β1Is a center superpixel of a sliding windowCoherence matrixSurface scattering parameter, alpha, obtained by a polarized target decomposition algorithm1Is a center superpixel of a sliding windowCoherence matrixThe secondary scattering parameters obtained by the polarized target decomposition algorithm,is a center superpixel of a sliding windowCoherence momentMatrix ofThe surface scattering decomposition coefficient obtained by the polarized target decomposition algorithm,is a center superpixel of a sliding windowCoherence matrixThe secondary scattering decomposition coefficient obtained by the polarized target decomposition algorithm,is a center superpixel of a sliding windowCoherence matrixObtaining a volume scattering decomposition coefficient through a polarized target decomposition algorithm;
representing sliding window boundary superpixelsCoherence matrixThe surface of (a) scatters the power,
representing sliding window boundary superpixelsCoherence matrixThe secondary scattered power of (a) is,
representing sliding window boundary superpixelsCoherence matrixThe volume of (a) is used to scatter power,
wherein, beta2Is a sliding window boundary superpixelCoherence matrixSurface scattering parameter, alpha, obtained by a polarized target decomposition algorithm2Is a sliding window boundary superpixelCoherence matrixThe secondary scattering parameters obtained by the polarized target decomposition algorithm,superpixels for sliding window boundariesCoherence matrixThe surface scattering decomposition coefficient obtained by the polarized target decomposition algorithm,superpixels for sliding window boundariesCoherence matrixThe secondary scattering decomposition coefficient obtained by the polarized target decomposition algorithm,superpixels for sliding window boundariesCoherence matrixObtaining a volume scattering decomposition coefficient through a polarized target decomposition algorithm;
2c3) at the super-pixel segmentation result SkIn (1), calculating a sliding window by using a superpixel sliding window modelCenter super pixelAnd sliding window boundary superpixelScattering power vector similarity parameter r ofij:
Wherein the superscript H is a conjugate transpose, | · |. non-woven phosphor2Is the norm of L2, |, represents the absolute value;
2c4) according to the scattering power vector similarity parameter r obtained in the step 2c3)ijCalculating the center superpixel of the sliding windowAnd sliding window boundary superpixelScatter component similarity measure ofThe value is calculated by the following formula:
5. the method of claim 1, step (3) measuring the likelihood ratioRiemann distance metricScattering componentVolume similarity measureThe three super-pixel level-based difference metrics are converted into pixel level-based difference metric vectorsThe method comprises the following steps of;
3a) performing superpixel on the center of the sliding window by using a K-means clustering algorithmAnd sliding window boundary superpixelLikelihood ratio metric ofClustering into two categories, using the mean of likelihood ratio metrics in the majority of categories as the center superpixel of the sliding windowFinal likelihood ratio metric
Wherein,representing a value of a superpixel likelihood ratio metric, c, in a current majority class1=1,...,ML,MLRepresenting the number of the superpixels of most types of the likelihood ratio metric value in the current cluster;
3b) sliding window by using K mean value clustering algorithmCenter super pixelAnd sliding window boundary superpixelRiemann distance measurementClustering into two classes, and taking the mean of Riemann distance measures in the majority of classes as the center superpixel of the sliding windowFinal Riemann distance metric
Wherein,representing a super-pixel Riemann distance metric, c, in a current majority class2=1,...,MR,MRRepresenting the number of the superpixels of most types of the Riemann distance metric value in the current cluster;
3c) performing superpixel on the center of the sliding window by using a K-means clustering algorithmAnd sliding window boundary superpixelScatter component similarity measure ofClustering into two categories, taking the mean of the similarity measures of the scattering components in the majority of categories as the center superpixel of the sliding windowFinal scatter component similarity measure
Wherein,representing a similarity measure, c, of the superpixel scatter components in the current majority class3=1,...,MC,MCRepresenting the number of superpixels of a majority of classes of the similarity measurement of the scattering components in the current cluster;
3d) center superpixel of sliding windowFinal likelihood ratio metricRiemann distance metricScatter component similarity metricConstituting a vector as a center superpixel of the sliding windowOf a vector of dissimilarity measures
3e) Center superpixel of sliding windowOf a vector of dissimilarity measuresIs distributed to the pixel inside to obtain the difference metric vector of the pixel pWhereinRepresenting a superpixelA pixel of (1);
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810013991.5A CN108171193B (en) | 2018-01-08 | 2018-01-08 | Polarized SAR (synthetic aperture radar) ship target detection method based on super-pixel local information measurement |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810013991.5A CN108171193B (en) | 2018-01-08 | 2018-01-08 | Polarized SAR (synthetic aperture radar) ship target detection method based on super-pixel local information measurement |
Publications (2)
Publication Number | Publication Date |
---|---|
CN108171193A CN108171193A (en) | 2018-06-15 |
CN108171193B true CN108171193B (en) | 2021-09-10 |
Family
ID=62517417
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810013991.5A Active CN108171193B (en) | 2018-01-08 | 2018-01-08 | Polarized SAR (synthetic aperture radar) ship target detection method based on super-pixel local information measurement |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN108171193B (en) |
Families Citing this family (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109063750B (en) * | 2018-07-17 | 2022-05-13 | 西安电子科技大学 | SAR target classification method based on CNN and SVM decision fusion |
CN110070545B (en) * | 2019-03-20 | 2023-05-26 | 重庆邮电大学 | Method for automatically extracting urban built-up area by urban texture feature density |
CN110781753A (en) * | 2019-09-27 | 2020-02-11 | 杭州电子科技大学 | Rapid detection method for polarized SAR ship target based on Gp0 hybrid model |
CN110717956B (en) * | 2019-09-30 | 2023-06-20 | 重庆大学 | L0 norm optimization reconstruction method guided by limited angle projection superpixel |
CN111123267B (en) * | 2019-10-30 | 2021-08-20 | 清华大学 | Synthetic aperture radar image ship detection method and device |
CN111027511B (en) * | 2019-12-23 | 2022-04-29 | 西安电子科技大学 | Remote sensing image ship detection method based on region of interest block extraction |
CN112528943A (en) * | 2020-12-23 | 2021-03-19 | 哈尔滨工业大学 | SAR image ocean target detection method based on entropy modulation SLIC |
CN112766221B (en) * | 2021-02-01 | 2022-06-14 | 福州大学 | Ship direction and position multitasking-based SAR image ship target detection method |
CN113989505B (en) * | 2021-11-02 | 2024-02-20 | 陕西师范大学 | SAR target detection method based on fusion multi-level fuzzy clustering soft decision result |
Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US8144361B2 (en) * | 2008-03-18 | 2012-03-27 | Konica Minolta Laboratory U.S.A., Inc. | Creation and placement of two-dimensional barcode stamps on printed documents for storing authentication information |
CN104376330A (en) * | 2014-11-19 | 2015-02-25 | 西安电子科技大学 | Polarization SAR image ship target detection method based on superpixel scattering mechanism |
CN104392228A (en) * | 2014-12-19 | 2015-03-04 | 中国人民解放军国防科学技术大学 | Unmanned aerial vehicle image target class detection method based on conditional random field model |
CN104680181A (en) * | 2015-03-09 | 2015-06-03 | 西安电子科技大学 | SAR image super-pixel segmentation method based on likelihood ratio features |
CN105469088A (en) * | 2015-11-17 | 2016-04-06 | 西北工业大学 | Object prediction area optimization method applicable to target identification |
CN105976376A (en) * | 2016-05-09 | 2016-09-28 | 电子科技大学 | High resolution SAR image target detection method based on part model |
CN106291550A (en) * | 2016-07-27 | 2017-01-04 | 西安电子科技大学 | The polarization SAR Ship Detection of core is returned based on local scattering mechanism difference |
Family Cites Families (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US10642891B2 (en) * | 2013-04-12 | 2020-05-05 | Avigilon Fortress Corporation | Graph matching by sub-graph grouping and indexing |
-
2018
- 2018-01-08 CN CN201810013991.5A patent/CN108171193B/en active Active
Patent Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US8144361B2 (en) * | 2008-03-18 | 2012-03-27 | Konica Minolta Laboratory U.S.A., Inc. | Creation and placement of two-dimensional barcode stamps on printed documents for storing authentication information |
CN104376330A (en) * | 2014-11-19 | 2015-02-25 | 西安电子科技大学 | Polarization SAR image ship target detection method based on superpixel scattering mechanism |
CN104392228A (en) * | 2014-12-19 | 2015-03-04 | 中国人民解放军国防科学技术大学 | Unmanned aerial vehicle image target class detection method based on conditional random field model |
CN104680181A (en) * | 2015-03-09 | 2015-06-03 | 西安电子科技大学 | SAR image super-pixel segmentation method based on likelihood ratio features |
CN105469088A (en) * | 2015-11-17 | 2016-04-06 | 西北工业大学 | Object prediction area optimization method applicable to target identification |
CN105976376A (en) * | 2016-05-09 | 2016-09-28 | 电子科技大学 | High resolution SAR image target detection method based on part model |
CN106291550A (en) * | 2016-07-27 | 2017-01-04 | 西安电子科技大学 | The polarization SAR Ship Detection of core is returned based on local scattering mechanism difference |
Non-Patent Citations (5)
Title |
---|
A Hierarchical Ship Detection Scheme for High-Resolution SAR Images;Yinghua Wang 等;《IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING》;20121031;第50卷(第10期);4173-4184 * |
HRWS SAR 图像舰船目标监视技术研究综述;邢相薇 等;《雷达学报》;20150206;第4卷(第1期);107-121 * |
Riemannian Distances for Signal Classification by Power Spectral Density;Kon Max Wong 等;《IEEE Journal of Selected Topics in Signal Processing》;20130831;654-669 * |
SAR 图像舰船目标识别综述;陈文婷 等;《现代雷达》;20121115;第34卷(第11期);53-58 * |
Superpixel-Level CFAR Detectors for Ship Detection in SAR Imagery;Odysseas Pappas 等;《IEEE Geoscience and Remote Sensing Letters》;20160405;730-734 * |
Also Published As
Publication number | Publication date |
---|---|
CN108171193A (en) | 2018-06-15 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108171193B (en) | Polarized SAR (synthetic aperture radar) ship target detection method based on super-pixel local information measurement | |
Xia et al. | Infrared small target detection based on multiscale local contrast measure using local energy factor | |
Bujor et al. | Application of log-cumulants to the detection of spatiotemporal discontinuities in multitemporal SAR images | |
He et al. | A novel automatic PolSAR ship detection method based on superpixel-level local information measurement | |
Leng et al. | A bilateral CFAR algorithm for ship detection in SAR images | |
Parizzi et al. | Adaptive InSAR stack multilooking exploiting amplitude statistics: A comparison between different techniques and practical results | |
CN110349160B (en) | SAR image segmentation method based on super-pixel and fuzzy C-means clustering | |
WO2016101279A1 (en) | Quick detecting method for synthetic aperture radar image of ship target | |
Fan et al. | Oil spill monitoring based on SAR remote sensing imagery | |
Xia et al. | A novel sea-land segmentation algorithm based on local binary patterns for ship detection | |
CN107742113B (en) | One kind being based on the posterior SAR image complex target detection method of destination number | |
CN106646469B (en) | SAR ship detection optimization method based on VC Method | |
CN101482969B (en) | SAR image speckle filtering method based on identical particle computation | |
CN108133468B (en) | Constant false alarm rate ship detection method based on adaptive parameter enhancement and trail auxiliary detection | |
CN110889843A (en) | SAR image ship target detection method based on maximum stable extremal region | |
CN106291550B (en) | The polarization SAR Ship Detection of core is returned based on local scattering mechanism difference | |
CN114373135A (en) | Ship target detection method based on local significance characteristic measurement | |
CN112883877B (en) | Method and system for extracting polarized SAR mangrove based on double unsupervised classification | |
Huang et al. | Superpixel-based change detection in high resolution sar images using region covariance features | |
Jiang et al. | Ship detection based on feature confidence for high resolution SAR images | |
CN112101250B (en) | Method for detecting offshore ship target based on context semantic perception | |
CN105046707B (en) | SAR image ship detection method based on n-order polynomial Function Fitting sea clutter | |
CN104732190B (en) | A kind of synthetic aperture sonar object detection method based on orthogonal texture correlation analysis | |
CN108710816B (en) | Polarized SAR image building detection method based on probability fusion | |
CN113822361B (en) | SAR image similarity measurement method and system based on Hamming distance |
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 |