CN108830883B - Visual attention SAR image target detection method based on super-pixel structure - Google Patents

Visual attention SAR image target detection method based on super-pixel structure Download PDF

Info

Publication number
CN108830883B
CN108830883B CN201810567306.3A CN201810567306A CN108830883B CN 108830883 B CN108830883 B CN 108830883B CN 201810567306 A CN201810567306 A CN 201810567306A CN 108830883 B CN108830883 B CN 108830883B
Authority
CN
China
Prior art keywords
image
super
formula
detection
target
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.)
Expired - Fee Related
Application number
CN201810567306.3A
Other languages
Chinese (zh)
Other versions
CN108830883A (en
Inventor
刘说
杨玲
于文涛
张无瑕
王海江
杨智鹏
徐梓欣
潘帆
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Chengdu University of Information Technology
Original Assignee
Chengdu University of Information Technology
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Chengdu University of Information Technology filed Critical Chengdu University of Information Technology
Priority to CN201810567306.3A priority Critical patent/CN108830883B/en
Publication of CN108830883A publication Critical patent/CN108830883A/en
Application granted granted Critical
Publication of CN108830883B publication Critical patent/CN108830883B/en
Expired - Fee Related legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/20—Analysis of motion
    • G06T7/246—Analysis of motion using feature-based methods, e.g. the tracking of corners or segments
    • 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
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/10—Image acquisition modality
    • G06T2207/10032—Satellite or aerial image; Remote sensing

Landscapes

  • Engineering & Computer Science (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Multimedia (AREA)
  • Image Analysis (AREA)

Abstract

The invention discloses a super-pixel-based methodA structural visual attention SAR image target detection method belongs to radar remote sensing or image processing technology and mainly solves the problems of low detection rate, high false alarm rate and omission ratio and distortion of detected targets during SAR image target detection. The method comprises the following implementation steps: determining an SAR image to be input, and filtering the SAR image; then extracting gray level and direction primary visual features; carrying out normalization and significance processing; generating a saliency map; setting a threshold SthGenerating a binaryzation saliency map to select a candidate target area; multiplying the binaryzation saliency map and the filtered image point by point; segmenting the image into super pixel areas by using a SLIC super pixel generation algorithm; setting a threshold R for corner detectionthHarris corner detection is carried out on the image to highlight the difference of the structural features of the super pixels of the target and the background; counting the number of angular points in each super pixel area; and setting a threshold Th to perform outlier detection to eliminate false alarms contained in the candidate target area, so as to obtain a final SAR image target detection result. The SAR image target detection is realized by fully utilizing a method combining superpixel, visual attention and Harris corner detection, and the obtained detection result shows that the method has high detection rate, low false alarm rate and low omission factor, and the detection result is not distorted, namely the target form of the detected SAR image can be completely reserved.

Description

Visual attention SAR image target detection method based on super-pixel structure
Technical Field
The invention belongs to a radar remote sensing or image processing technology, namely, an image processing technology is used for analyzing radar observation information, and particularly relates to application of a method combining visual attention, superpixels and Harris angular point detection in SAR image target detection.
Background
Synthetic Aperture Radar (SAR) imaging is not limited by conditions such as weather and illumination, and can perform all-weather and all-time reconnaissance on an interested target. The research on SAR image target detection is carried out, and the method is very important for obtaining military information.
There are many methods for detecting the SAR image target, including detection methods based on segmentation method, fractal theory, wavelet decomposition, template, likelihood ratio, multi-polarization data, Constant False Alarm Rate (CFAR), and the like. Among them, the detection algorithm based on CFAR is most widely used. The learner also uses a double-parameter CFAR detection algorithm to carry out SAR image target detection, and a better effect is achieved.
CFAR based on a background clutter statistical model and SAR image target detection methods based on double-parameter CFAR are based on pixel-level detection, and have certain defects. After the super-pixel concept is proposed, the SAR image target detection based on the super-pixel develops rapidly. Some researchers provide a target detection method combining constant false alarm rate CFAR based on superpixels with morphological processing, but the method has the defects that the superpixels in the SAR image need to be traversed by using a sliding window, the calculation amount is large, and the target morphology of a final detection result is distorted. In addition, researchers have proposed a target screening method based on pixel classification, which presets a certain number of different types of target combinations, but in the context of complex SAR images, it is difficult for background and target pixels to be represented by all of the number of different types.
Although there are many researchers who use the visual attention model to realize the target detection of the SAR image, the simple visual attention detection method has many problems. In a single-scene and simple SAR image environment, the simple visual attention SAR image target detection method can achieve a good effect, but for a complex SAR image background, the simple visual attention SAR image target detection method generally has the problems of low detection rate, high omission factor and more false alarms. In addition, the pure visual attention SAR image target detection method is difficult to simultaneously ensure higher detection rate, lower omission factor and lower false alarm rate.
Although many optical image detections use Harris corner detection, the research on the corner detection of the SAR image is very little, and the research on false alarm filtering of the SAR image target is almost none. However, the number of the corner points of the target and the background of the SAR image has a significant difference, and the false alarm filtering can be realized by detecting the number of the corner points of the target and the background of the SAR image. In addition, the corner detection is combined with visual attention and superpixels, so that the SAR image target detection in a simple scene can be finished, and the SAR image target detection in a complex scene can be finished.
Disclosure of Invention
The invention aims to overcome the defects of the existing SAR image target detection method, improve the detection rate of SAR image target detection, reduce the false alarm rate and the omission factor, retain the original form of a target and position the target, and particularly provides a visual attention SAR image target detection method based on a super-pixel structure. The method comprises the following implementation steps: determining an SAR image to be input, and filtering the SAR image; then extracting gray level and direction primary visual features; carrying out normalization and significance processing; generating a saliency map; setting a threshold SthGenerating a binaryzation saliency map to select a candidate target area; multiplying the binaryzation saliency map and the filtered image point by point; segmenting the image into super pixel areas by using a SLIC super pixel generation algorithm; setting a threshold R for corner detectionthHarris corner detection is carried out on the image to highlight the difference of the structural features of the super pixels of the target and the background; counting the number of angular points in each super pixel area; and setting a threshold Th to perform outlier detection to eliminate false alarms contained in the candidate target area, so as to obtain a final SAR image target detection result.
The detailed technical scheme of the invention is as follows:
a visual attention SAR image target detection method based on a super-pixel structure is disclosed, as shown in FIG. 1, and comprises the following steps:
step 1: filtering an original SAR image, wherein a matrix A (m, n) represents the original SAR image, a matrix B represents a convolution template of a filter, and a filtering formula is a formula (1):
Figure DEST_PATH_IMAGE001
（1）
wherein the content of the first and second substances,
Figure 155010DEST_PATH_IMAGE002
，
Figure 665626DEST_PATH_IMAGE003
，Mris the number of rows, M, of the matrix ACIs the column number, N, of the matrix ArIs the number of rows of matrix B, NCThe number of columns of matrix B.
Step 2: extracting gray primary visual feature I of the image C (s, t) after the operation of the step 1int(x, y) in the formula (2):
Figure 914205DEST_PATH_IMAGE004
（2）
wherein, the symbol
Figure 309414DEST_PATH_IMAGE005
Representing the gray values of the computed image.
And step 3: extracting the directional primary visual feature I of the image C (s, t) operated in the step 1 by adopting a Gabor filterori(x, y), the formula of the direction primary visual feature extraction is formula (3), formula (4), formula (5), formula (6) and formula (7):
Figure 717262DEST_PATH_IMAGE006
（3）
Figure 8566DEST_PATH_IMAGE007
（4）
Figure 275599DEST_PATH_IMAGE008
（5）
Figure 67975DEST_PATH_IMAGE009
（6）
Figure 205695DEST_PATH_IMAGE010
（7）
wherein the content of the first and second substances,
Figure 199059DEST_PATH_IMAGE011
is the standard deviation of the gaussian function on the x-axis,
Figure 555652DEST_PATH_IMAGE012
is the standard deviation of the gaussian function on the y-axis,
Figure 761506DEST_PATH_IMAGE013
is the direction of the light beam emitted by the light source,
Figure 284891DEST_PATH_IMAGE014
is the wavelength, sign, of a sine wave
Figure 308211DEST_PATH_IMAGE005
Representing gray values of the computed image;
and 4, step 4: normalization and significance processing, namely, processing the gray level image I obtained after the operation of the step 2 and the step 3int(x, y) and Iori(x, y) the following operations were performed, respectively: the gray values of all pixels in the image are first expanded or compressed to a uniform gray scale range [0, N]Then multiplying the gray values of all pixels in the image by a coefficient
Figure 18678DEST_PATH_IMAGE015
Obtaining N (I)int(x, y)) and N (I)ori(x,y))；
Wherein N represents the setting parameter of the maximum value of the gray scale in the normalized image,
Figure 293801DEST_PATH_IMAGE016
indicates the average of the gradation values of all the pixels excluding the maximum gradation value N.
And 5: after the operation of step 4, use formula (8)) Obtain a saliency map Smap(x,y)；
Figure 530748DEST_PATH_IMAGE017
（8）
Step 6: generating a binary saliency map, and setting a threshold value SthThe gray scale image S after the operation of step 5 is expressed by the formula (9)map(x, y) binarizing;
Figure 334756DEST_PATH_IMAGE018
（9）
and 7: the image S after the operation of the step 6 is processedmapPerforming dot multiplication on the (x, y) and the image obtained after the operation in the step 1 to obtain an image
Figure 63677DEST_PATH_IMAGE019
；
And 8: for the image after the operation of the step 7
Figure 735967DEST_PATH_IMAGE019
Performing superpixel operation by using a simple linear iterative clustering algorithm to obtain an image I with k superpixel regions; in this step, the edges of the superpixel regions fit the edges of the target in the SAR image.
And step 9: performing Harris corner detection, firstly calculating the gradient I of the image I in the x direction after the operation of the step 8 according to a formula (10)xCalculating the gradient I of the image I in the two directions y after the operation of the step 8 according to the formula (11)y：
Figure 233944DEST_PATH_IMAGE020
（10）
Figure 474433DEST_PATH_IMAGE021
（11）
Then, the product of the gradients in the two directions of the image is calculated according to the formula (12), the formula (13) and the formula (14):
Figure 18547DEST_PATH_IMAGE022
（12）
Figure 901052DEST_PATH_IMAGE023
（13）
Figure 722377DEST_PATH_IMAGE024
（14）
then, setting a matrix
Figure 258401DEST_PATH_IMAGE025
E, F, C of the matrix M is obtained by using a Gaussian function as the window function w and calculating according to the formula (15), the formula (16) and the formula (17):
Figure 696336DEST_PATH_IMAGE026
（15）
Figure 585794DEST_PATH_IMAGE027
（16）
Figure 58364DEST_PATH_IMAGE028
（17）
next, the Harris corner response function R for each pixel is calculated according to equation (18):
Figure 765289DEST_PATH_IMAGE029
（18）
wherein the content of the first and second substances,
Figure 159361DEST_PATH_IMAGE030
is a constant;
next, a threshold value R is setthAccording to formula (19)The operation is as follows:
Figure 977144DEST_PATH_IMAGE031
（19）
and then, next, in
Figure 38641DEST_PATH_IMAGE032
Non-maximum suppression is performed in the neighborhood of (c),
Figure 791834DEST_PATH_IMAGE032
the local maximum point in the neighborhood is the corner point in the image.
Step 10: counting the number of angular points in each super-pixel region in the image I, calculating the percentage of the number of the angular points in the total number of pixels in each super-pixel region, and drawing a distribution histogram of percentage data.
Step 11: and (3) filtering false alarms contained in the candidate target area by outlier detection, firstly calculating a mean value and a variance corresponding to percentage data, and then carrying out outlier detection on the percentage of the angular points in each super-pixel area. The outlier detection formula for the percentage of corners within a superpixel region is shown in (20):
Figure 800765DEST_PATH_IMAGE033
（20）
wherein, variable
Figure 828764DEST_PATH_IMAGE034
Is the percentage of corner points within the super-pixel area,
Figure 72664DEST_PATH_IMAGE035
is the nth super pixel region, k is the number of super pixel regions in the image I, and variable
Figure 731178DEST_PATH_IMAGE036
And variables
Figure 224476DEST_PATH_IMAGE037
A variable Td is a threshold for outlier detection for the mean and variance of the data obtained in step 10; and obtaining a final SAR image target detection result and positioning the target.
Compared with the prior art, the invention has the following advantages:
1. compared with the traditional detection method using pixels as target detection units, the method provided by the invention can completely retain the original form of the target in the SAR image, has no distortion of the target detection result, and is suitable for the target detection of the SAR image which is more complex, has a larger scene and contains more targets.
2. First, a threshold S is setthGenerating a binaryzation saliency map to select a candidate target area; then, a threshold value R for Harris corner detection is setthHighlighting differences in structural features of superpixels of the target and the background; and finally, setting a threshold Td of outlier detection, filtering false alarms contained in the candidate target area, and finally obtaining an SAR image target detection result. The three thresholds are arranged in a loop-to-loop mode, when the first threshold is set, more false alarms are generated rather, and the false alarms cannot be missed, the false alarms are filtered step by step through setting of the second threshold and the third threshold, so that the SAR image target detection is guaranteed to reach higher detection rate and lower false alarm rate, and the SAR image target detection result is finally obtained and the target is positioned.
Drawings
FIG. 1 is a flow chart of a target detection method of a visual attention SAR image based on a super-pixel structure according to the present invention;
FIG. 2 is a raw SAR image tested;
FIG. 3 a filtered SAR image;
FIG. 4 is a graph of extracted gray-level primary visual features;
FIG. 5 is a diagram of extracted directional primary visual features;
FIG. 6 is a graph of normalized and saliency-processed directional primary visual features;
FIG. 7 is a graph of normalized and saliency processed gray scale primary visual features;
FIG. 8 is a saliency map;
FIG. 9 is a saliency map after binarization;
FIG. 10 is a result of dot multiplication of the binarized saliency map and the filtered SAR image;
FIG. 11 is an image after SLIC superpixel;
fig. 12 is an image after Harris corner detection;
FIG. 13 is a distribution histogram of the obtained percentage data, in which the abscissa is the super pixel patch number and the ordinate is the percentage;
fig. 14 is a result after outlier detection, that is, a final target detection result of the SAR image;
fig. 15 is a comparative detection result graph of the CFAR detection algorithm based on the Gamma distribution with the false alarm rate set to 0.001.
Detailed Description
The invention is further described with reference to the following figures and examples.
(I) Experimental conditions
An experiment platform: matlab R2015a, the same processor is a sixth generation core i5-6200U, and the memory is an 8G associative E41 notebook computer.
(II) measured data
The experimental data figure 2 is an original SAR image containing 6 targets with a size of
Figure 259429DEST_PATH_IMAGE038
。
(III) simulation experiment content
Simulation experiments first filter fig. 2, resulting in fig. 3.
The grayscale primary visual feature of fig. 3 is extracted, resulting in fig. 4.
The directional primary visual features of fig. 3 are extracted, resulting in fig. 5.
The normalization operation and significance processing are performed on fig. 4, resulting in fig. 6. In this operation step, the parameter N = 0.8.
The normalization operation and significance processing are performed on fig. 5, resulting in fig. 7.
Operating fig. 6 and 7 according to step 5 results in a saliency map, as in fig. 8.
And (5) binarizing the saliency map to obtain a map 9. Wherein in the operation step, the threshold value S is binaryth=0.45。
Fig. 10 was calculated according to step 7.
Fig. 11 is obtained by performing superpixel on fig. 10, and a total of 2501 superpixel regions are generated, and the edges of the superpixel regions are attached to the edge of the target.
Harris corner detection was performed on FIG. 11 to obtain FIG. 12. In this operation step, the threshold value Rth=0.06。
Fig. 13 is a distribution histogram of the calculated percentage data. In the figure it can be seen that there are 6 maxima, corresponding to 6 targets in the SAR image.
Outlier detection was performed to obtain FIG. 14. In this operation step, the threshold Th = 0.06. When the percentage is larger than a set threshold value, the super pixel area is judged as a target; otherwise, the super pixel area is judged as the background.
In order to evaluate the target detection performance, the detection is carried out in the same Matlab operating environment, and the target is detected by using a CFAR detection algorithm based on Gamma distribution. The detection results are shown in fig. 15.
The two detection methods are shown in table 1 for the detection performance ratio.
TABLE 1 comparison of the test Performance of the two test methods
Figure 561097DEST_PATH_IMAGE039
As can be seen from comparison experiments, the detection performance of the method is superior to that of a CFAR detection method based on Gamma distribution, the target object is completely detected, no false alarm target exists, and the original form of the target is completely reserved. And the detection result of the CFAR detection algorithm based on Gamma distribution has more false alarms, and the form of the target in the detection result is seriously distorted.
In conclusion, the SAR image detection method has the advantages of high detection rate, low false alarm rate and low omission factor of the target in the SAR image, can completely reserve the original form of the detected target and effectively positions the target. In addition, the super-pixels are used as processing units, and the algorithm operation efficiency is high.
The simulation experiment verifies the correctness, effectiveness and reliability of the invention.
Various modifications and alterations of this invention may be made by those skilled in the art without departing from the scope of this invention.

Claims (3)

1. A visual attention SAR image target detection method based on a super-pixel structure comprises the following steps:
step 1: filtering and matrix processing are carried out on the original SAR image
Figure 473721DEST_PATH_IMAGE001
Representing the original SAR image, the matrix B representing the convolution template of the filter, the filter formula being formula (1):
Figure 390861DEST_PATH_IMAGE002
（1）
wherein the content of the first and second substances,
Figure 170598DEST_PATH_IMAGE003
，
Figure 628124DEST_PATH_IMAGE004
，
Figure 442497DEST_PATH_IMAGE005
the number of rows in the matrix a is,
Figure 264959DEST_PATH_IMAGE006
is the number of columns of the matrix a,
Figure 531992DEST_PATH_IMAGE007
the number of rows of the matrix B,
Figure 730893DEST_PATH_IMAGE008
is the column number of the matrix B;
step 2: extracting the image operated in the step 1
Figure 462088DEST_PATH_IMAGE009
Gray scale primary visual feature of
Figure 455452DEST_PATH_IMAGE010
The calculation formula is formula (2):
Figure 209781DEST_PATH_IMAGE011
（2）
wherein, the symbol
Figure 946793DEST_PATH_IMAGE012
Representing gray values of the computed image;
and step 3: extracting the image operated in the step 1 by adopting a Gabor filter
Figure 470179DEST_PATH_IMAGE009
Directional primary visual feature of
Figure 962340DEST_PATH_IMAGE013
The formula for extracting the direction primary visual feature is formula (3), formula (4), formula (5), formula (6) and formula (7):
Figure 203965DEST_PATH_IMAGE014
（3）
Figure 479089DEST_PATH_IMAGE015
（4）
Figure 122560DEST_PATH_IMAGE016
（5）
Figure 21508DEST_PATH_IMAGE017
（6）
Figure 750429DEST_PATH_IMAGE018
（7）
wherein the content of the first and second substances,
Figure 829244DEST_PATH_IMAGE019
is a Gaussian function in
Figure 327221DEST_PATH_IMAGE020
The standard deviation on the axis of the shaft,
Figure 98868DEST_PATH_IMAGE021
is a Gaussian function in
Figure 49507DEST_PATH_IMAGE022
The standard deviation on the axis of the shaft,
Figure 994329DEST_PATH_IMAGE023
is the direction of the light beam emitted by the light source,
Figure 346813DEST_PATH_IMAGE024
is the wavelength, sign, of a sine wave
Figure 289361DEST_PATH_IMAGE012
Representing gray values of the computed image;
and 4, step 4: normalization and significance processing, namely processing the gray level images obtained after the operation of the step 2 and the step 3
Figure 727296DEST_PATH_IMAGE025
And
Figure 210230DEST_PATH_IMAGE026
the following operations were performed: first, the drawings are drawnThe gray scale values of all pixels in the image are expanded or compressed to a uniform gray scale range
Figure 682799DEST_PATH_IMAGE027
Then multiplying the gray values of all pixels in the image by a coefficient
Figure 796249DEST_PATH_IMAGE028
To obtain
Figure 721479DEST_PATH_IMAGE029
And
Figure 945787DEST_PATH_IMAGE030
(ii) a Wherein the normalization and saliency processing is to grayscale images
Figure 69601DEST_PATH_IMAGE031
And
Figure 353952DEST_PATH_IMAGE026
the following operations were performed:
4.1: the maximum value of the gray scale in the normalized image was set to 0.8,
Figure 766479DEST_PATH_IMAGE032
represents the average of the gradation values of all pixels except the maximum gradation value of 0.8;
4.2: expanding or compressing the gray values of all pixels in the image to a uniform gray scale range
Figure 794478DEST_PATH_IMAGE033
；
4.3: multiplying the gray values of all pixels in the image by a coefficient
Figure 976060DEST_PATH_IMAGE034
To obtain
Figure 165733DEST_PATH_IMAGE035
And
Figure 629338DEST_PATH_IMAGE036
；
wherein the content of the first and second substances,
Figure 195448DEST_PATH_IMAGE037
a setting parameter representing a maximum value of the gray level in the normalized image,
Figure 497117DEST_PATH_IMAGE038
indicating removal of maximum gray value
Figure 857691DEST_PATH_IMAGE037
The average of the gradation values of all the pixels except for the above;
and 5: after the operation of step 4, the saliency map is obtained using equation (8)
Figure 307127DEST_PATH_IMAGE039
；
Figure 676928DEST_PATH_IMAGE040
（8）
Step 6: generating a binary saliency map, and setting a threshold value
Figure 567524DEST_PATH_IMAGE041
The gray level image after the operation of step 5 is expressed by the formula (9)
Figure 364579DEST_PATH_IMAGE039
Binaryzation;
Figure 973414DEST_PATH_IMAGE042
（9）
and 7: the image after the operation of the step 6 is processed
Figure 209224DEST_PATH_IMAGE039
Performing dot multiplication with the image operated in the step 1 to obtain an image
Figure 219905DEST_PATH_IMAGE043
；
And 8: for the image after the operation of the step 7
Figure 187861DEST_PATH_IMAGE043
Performing superpixel operation by using simple linear iterative clustering algorithm to obtain the super pixel
Figure 283993DEST_PATH_IMAGE044
Image of a super pixel area
Figure 995597DEST_PATH_IMAGE045
；
And step 9: harris angular point detection is carried out, and the steps are as follows:
9.1: calculating the image after the operation of the step 8 according to the formula (10)
Figure 923102DEST_PATH_IMAGE045
In that
Figure 61959DEST_PATH_IMAGE020
Gradient of direction
Figure 645387DEST_PATH_IMAGE046
Calculating the image after the operation of step 8 according to the formula (11)
Figure 895103DEST_PATH_IMAGE045
In that
Figure 880377DEST_PATH_IMAGE022
Gradient in two directions
Figure 190135DEST_PATH_IMAGE047
：
Figure 824641DEST_PATH_IMAGE048
（10）
Figure 940365DEST_PATH_IMAGE049
（11）
9.2: calculating the product of the gradients of the two directions of the image according to the formula (12), the formula (13) and the formula (14):
Figure 780145DEST_PATH_IMAGE050
（12）
Figure 260805DEST_PATH_IMAGE051
（13）
Figure 553246DEST_PATH_IMAGE052
（14）
9.3: setting matrix
Figure 410343DEST_PATH_IMAGE053
Using a Gaussian function as the window function
Figure 166947DEST_PATH_IMAGE054
Then calculating according to formula (15), formula (16) and formula (17) to obtain matrix
Figure 818508DEST_PATH_IMAGE055
Is/are as follows
Figure 863824DEST_PATH_IMAGE056
：
Figure 259033DEST_PATH_IMAGE057
（15）
Figure 807826DEST_PATH_IMAGE058
（16）
Figure 958185DEST_PATH_IMAGE059
（17）
9.4: the Harris corner response function for each pixel is calculated according to equation (18)
Figure 225218DEST_PATH_IMAGE060
：
Figure 424119DEST_PATH_IMAGE061
（18）
Wherein the content of the first and second substances,
Figure 827418DEST_PATH_IMAGE062
is a constant;
9.5: setting a threshold value
Figure 86361DEST_PATH_IMAGE063
Operating according to equation (19):
Figure 404472DEST_PATH_IMAGE064
（19）
9.6: in that
Figure 141484DEST_PATH_IMAGE065
Non-maximum suppression is performed in the neighborhood of (c),
Figure 664869DEST_PATH_IMAGE066
local maximum points in the neighborhood are angular points in the image;
step 10: statistical images
Figure 94714DEST_PATH_IMAGE045
Calculating the percentage of the number of the angular points in each super pixel region to the total number of pixels in each super pixel region, and drawing a distribution histogram of percentage data;
step 11: outlier detection filters false alarms contained in the candidate target region, first, a mean value and a variance corresponding to percentage data are calculated, then, outlier detection is performed on the percentage of the corners in each super-pixel region, and an outlier detection formula of the percentage of the corners in the super-pixel region is shown as (20):
Figure 70760DEST_PATH_IMAGE067
（20）
wherein, variable
Figure 673779DEST_PATH_IMAGE068
Is the percentage of corner points within the super-pixel area,
Figure 317250DEST_PATH_IMAGE069
is the n-th super-pixel region,
Figure 652417DEST_PATH_IMAGE044
as an image
Figure 381338DEST_PATH_IMAGE045
Number of Mega superpixel regions, variables
Figure 460153DEST_PATH_IMAGE070
And variables
Figure 20447DEST_PATH_IMAGE071
Mean and variance, respectively, of the data obtained in step 10
Figure 152614DEST_PATH_IMAGE072
A threshold for outlier detection; and obtaining a final SAR image target detection result and positioning the target.
2. The method for detecting the target of the visual attention SAR image based on the super-pixel structure as claimed in claim 1, wherein when Harris corner detection is used and the number of target corners in each super-pixel region is counted to realize outlier detection, to filter out a false alarm and detect the super-pixel region where the target is located, the structural features of the super-pixels are used to realize the false alarm filtering and detect the target in the SAR image;
wherein the corner response function
Figure 368831DEST_PATH_IMAGE073
Expressed as formula (21):
Figure 251337DEST_PATH_IMAGE074
（21）。
3. the method for detecting the target of the visual attention SAR image based on the super-pixel structure as claimed in claim 1, wherein for the step 11 of filtering out the outlier detection false alarm, the following steps are used:
firstly, establishing a distribution histogram of corner point number data in a super-pixel region, and then calculating a corresponding mean value and a corresponding variance;
then, outlier detection is performed on the number of corner points in each superpixel region according to equation (22):
Figure 603821DEST_PATH_IMAGE075
（22）
wherein, variable
Figure 608686DEST_PATH_IMAGE076
The number of corner points in the superpixel region,
Figure 46620DEST_PATH_IMAGE077
is the n-th super-pixel region,
Figure 467237DEST_PATH_IMAGE044
as an image
Figure 939807DEST_PATH_IMAGE045
Number of Mega superpixel regions, variables
Figure 849994DEST_PATH_IMAGE078
And variables
Figure 775225DEST_PATH_IMAGE079
Of variable quantity
Figure 999533DEST_PATH_IMAGE080
。
CN201810567306.3A 2018-06-05 2018-06-05 Visual attention SAR image target detection method based on super-pixel structure Expired - Fee Related CN108830883B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201810567306.3A CN108830883B (en) 2018-06-05 2018-06-05 Visual attention SAR image target detection method based on super-pixel structure

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201810567306.3A CN108830883B (en) 2018-06-05 2018-06-05 Visual attention SAR image target detection method based on super-pixel structure

Publications (2)

Publication Number Publication Date
CN108830883A CN108830883A (en) 2018-11-16
CN108830883B true CN108830883B (en) 2022-04-01

Family

ID=64143762

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201810567306.3A Expired - Fee Related CN108830883B (en) 2018-06-05 2018-06-05 Visual attention SAR image target detection method based on super-pixel structure

Country Status (1)

Country Link
CN (1) CN108830883B (en)

Families Citing this family (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110110618B (en) * 2019-04-22 2022-10-14 电子科技大学 SAR target detection method based on PCA and global contrast
CN110853058B (en) * 2019-11-12 2023-01-03 电子科技大学 High-resolution remote sensing image road extraction method based on visual saliency detection
CN112001239B (en) * 2020-07-15 2023-11-28 成都信息工程大学 SAR image target detection optimization method based on multiple SVD saliency fusion
CN111832502A (en) * 2020-07-20 2020-10-27 中国人民解放军战略支援部队航天工程大学 Remote sensing image visual salient region intelligent search method for satellite in-orbit application

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US8295647B2 (en) * 2010-03-01 2012-10-23 National Tsing Hua University Compressibility-aware media retargeting with structure preserving
CN106485740A (en) * 2016-10-12 2017-03-08 武汉大学 A kind of combination point of safes and the multidate SAR image registration method of characteristic point
CN107122728A (en) * 2017-04-21 2017-09-01 西安电子科技大学 Building method is detected in SAR image based on evidence theory algorithm

Family Cites Families (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
BRPI0400192A (en) * 2004-02-11 2005-10-04 Biolab Sanus Farmaceutica Ltda Pharmaceutical compositions of peptides secreted by snake venom glands, particularly bothrops jararaca, evasins, their analogues, derivatives and associated products for use as acetylcholine receptor modulating agents
US9466009B2 (en) * 2013-12-09 2016-10-11 Nant Holdings Ip. Llc Feature density object classification, systems and methods
CN105550685B (en) * 2015-12-11 2019-01-08 哈尔滨工业大学 The large format remote sensing image area-of-interest exacting method of view-based access control model attention mechanism
CN106651937B (en) * 2016-10-19 2019-10-18 成都电科智达科技有限公司 A kind of small drone object detection method based on super-pixel and scene prediction
CN107729903A (en) * 2017-10-17 2018-02-23 内蒙古科技大学 SAR image object detection method based on area probability statistics and significance analysis

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US8295647B2 (en) * 2010-03-01 2012-10-23 National Tsing Hua University Compressibility-aware media retargeting with structure preserving
CN106485740A (en) * 2016-10-12 2017-03-08 武汉大学 A kind of combination point of safes and the multidate SAR image registration method of characteristic point
CN107122728A (en) * 2017-04-21 2017-09-01 西安电子科技大学 Building method is detected in SAR image based on evidence theory algorithm

Also Published As

Publication number Publication date
CN108830883A (en) 2018-11-16

Similar Documents

Publication Publication Date Title
CN108830883B (en) Visual attention SAR image target detection method based on super-pixel structure
CN108510467B (en) SAR image target identification method based on depth deformable convolution neural network
Zhang et al. A fusion algorithm for infrared and visible images based on saliency analysis and non-subsampled Shearlet transform
CN110415208A (en) A kind of adaptive targets detection method and its device, equipment, storage medium
CN109726649B (en) Remote sensing image cloud detection method and system and electronic equipment
CN110135312B (en) Rapid small target detection method based on hierarchical LCM
CN109977899B (en) Training, reasoning and new variety adding method and system for article identification
Liu et al. Automatic lung segmentation based on image decomposition and wavelet transform
CN111339948A (en) Automatic identification method for newly-added buildings of high-resolution remote sensing images
CN110310263B (en) SAR image residential area detection method based on significance analysis and background prior
CN110070545B (en) Method for automatically extracting urban built-up area by urban texture feature density
CN107133964B (en) Image matting method based on Kinect
Kumar An efficient text extraction algorithm in complex images
CN104268845A (en) Self-adaptive double local reinforcement method of extreme-value temperature difference short wave infrared image
Azam et al. Segmentation of breast microcalcification using hybrid method of Canny algorithm with Otsu thresholding and 2D Wavelet transform
CN112101283A (en) Intelligent identification method and system for traffic signs
CN106778822B (en) Image straight line detection method based on funnel transformation
CN111767856A (en) Infrared small target detection algorithm based on gray value statistical distribution model
CN110276260B (en) Commodity detection method based on depth camera
CN115147613A (en) Infrared small target detection method based on multidirectional fusion
Shao et al. An adaptive image contrast enhancement algorithm based on retinex
CN110472472B (en) Airport detection method and device based on SAR remote sensing image
PL A study on various image processing techniques
Lu et al. Towards physical distortion identification and removal in document images
CN111382632A (en) Target detection method, terminal device and computer-readable storage medium

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
CF01 Termination of patent right due to non-payment of annual fee

Granted publication date: 20220401