CN116363135A - Infrared target detection method, device, medium and equipment based on Gaussian similarity - Google Patents

Infrared target detection method, device, medium and equipment based on Gaussian similarity Download PDF

Info

Publication number
CN116363135A
CN116363135A CN202310637222.3A CN202310637222A CN116363135A CN 116363135 A CN116363135 A CN 116363135A CN 202310637222 A CN202310637222 A CN 202310637222A CN 116363135 A CN116363135 A CN 116363135A
Authority
CN
China
Prior art keywords
gray
similarity
target pixel
vector
gaussian
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN202310637222.3A
Other languages
Chinese (zh)
Other versions
CN116363135B (en
Inventor
涂兵
汪文
郭龙源
何伟
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.)
Nanjing University of Information Science and Technology
Original Assignee
Nanjing University of Information Science and 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 Nanjing University of Information Science and Technology filed Critical Nanjing University of Information Science and Technology
Priority to CN202310637222.3A priority Critical patent/CN116363135B/en
Publication of CN116363135A publication Critical patent/CN116363135A/en
Application granted granted Critical
Publication of CN116363135B publication Critical patent/CN116363135B/en
Active 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
    • G06T7/0002—Inspection of images, e.g. flaw detection
    • 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
    • G06V10/28—Quantising the image, e.g. histogram thresholding for discrimination between background and foreground patterns
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/74—Image or video pattern matching; Proximity measures in feature spaces
    • G06V10/761—Proximity, similarity or dissimilarity measures
    • 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
    • G06T2207/10048—Infrared image
    • 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)
  • Theoretical Computer Science (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Multimedia (AREA)
  • Computing Systems (AREA)
  • Artificial Intelligence (AREA)
  • Health & Medical Sciences (AREA)
  • Databases & Information Systems (AREA)
  • Evolutionary Computation (AREA)
  • General Health & Medical Sciences (AREA)
  • Medical Informatics (AREA)
  • Software Systems (AREA)
  • Quality & Reliability (AREA)
  • Photometry And Measurement Of Optical Pulse Characteristics (AREA)

Abstract

The invention discloses an infrared target detection method, device, medium and equipment based on Gaussian similarity, wherein the detection method comprises the following steps: s1: acquiring a thermal infrared image to be detected, and obtaining a gray image matrix; s2: constructing a face kernel, and filtering by using a gray level image matrix of the face kernel to obtain a candidate target pixel point set; s3: mapping the candidate target pixel point set into a thermal infrared image, and constructing a longitudinal gray scale distribution vector and a transverse gray scale distribution vector for each candidate target pixel point; s4: constructing a Gaussian function vector, and carrying out similarity analysis on each candidate target pixel point based on the Gaussian function vector to obtain a target pixel point set; s5: and constructing and obtaining a detection image of the infrared target based on the target pixel point set.

Description

Infrared target detection method, device, medium and equipment based on Gaussian similarity
Technical Field
The invention relates to the field of computers, in particular to an infrared target detection method, device, medium and equipment based on Gaussian similarity.
Background
Infrared imaging detection has several advantages over visible imaging detection and active radar imaging detection: the visible light imaging system is easily affected by illumination intensity, cannot work in a closed space or in low light conditions such as night, and the infrared radiation energy of an object is only related to the temperature and the material characteristics of the object, so that the infrared imaging is not affected by the illumination intensity, and can work normally in the low light environment; furthermore, since the infrared radiation intensity of an object is closely related to temperature, the infrared imaging system is able to sense the temperature difference; the active radar imaging needs to actively emit electromagnetic waves to the outside, the system is easy to expose itself or suffer electromagnetic interference when in operation, the infrared imaging detection belongs to a passive detection technology, and the concealment is stronger; the visible light imaging detection has poor penetrating capability, is easily influenced by interference factors such as severe weather environment or object shielding, the active radar detection capability is limited by the rapid development of the current radar stealth technology, and the infrared imaging detection has strong penetrating capability and good anti-interference performance, and is not influenced by various camouflage technologies such as radar stealth. Thanks to the advantages mentioned above, infrared imaging systems have been successfully and widely used in civil fields such as medical diagnostic analysis, agricultural and industrial monitoring, facial recognition, etc., and in addition, their application value in military fields such as military reconnaissance, early warning and guidance is more remarkable, while infrared search and tracking (IRST) systems are one of the central components.
The infrared small target detection has the following difficulties: 1. the target size is small and the signal strength is weak. In general, the distance between the target and the infrared detector is far, so that the pixel size of the target in the infrared image is small, and the target has no abundant shape and texture characteristics, so that the traditional target detection and tracking algorithm is not applicable any more; 2. during long-distance imaging, the infrared radiation of a target received by an imaging detector is interfered by various radiation sources such as atmosphere, cloud and the like to generate serious radiation energy attenuation, so that the signal intensity of the target in an infrared image is weaker, the pixel brightness value is lower, and even the pixel brightness value is possibly lower than that of some background pixels; 3. the detection scene is complex, and the background clutter interference is serious. In most application scenes, such as sky, land, sea and air detection scenes, clutter backgrounds with larger signal intensities, such as cloud layers, houses, sea waves and the like, appear in the field of view of the detector, and objects in the backgrounds show contrast characteristics similar to infrared small targets, so that false detection of an algorithm is easy to form.
Aiming at the above-mentioned difficulties, especially the infrared small target image with complex background, the infrared image contains various clutter background infrared small target detection, and the existing method has the problems of high false detection and long detection time for infrared small target detection, so that it is needed to provide an effective infrared small target detection method to improve the infrared small target detection accuracy, reduce the false detection rate and shorten the detection time.
Disclosure of Invention
In order to overcome the defects in the prior art, the invention provides an infrared target detection method, device, medium and equipment based on Gaussian similarity, which are used for solving at least one of the technical problems.
Based on one aspect of the specification, the invention provides an infrared target detection method based on Gaussian similarity, which specifically comprises the following steps:
s1: acquiring a thermal infrared image to be detected, and correspondingly arranging gray values of pixel points of the thermal infrared image and coordinate positions of the pixel points to obtain a gray image matrix;
s2: constructing a face kernel, and filtering by using the face kernel gray level image matrix to obtain a candidate target pixel point set;
Facet Kernel(facial core) originFacetThe model, assuming that the gray-scale intensity surface of a pixel can be approximated by a binary cubic polynomial, but using the least squares method to approximate the intensity surface, is computationally complex. Therefore, if the intensity curved surface can be formed by linearly combining a stack of discrete orthogonal polynomials, the calculation is very convenient, and the linear combination of the discrete orthogonal polynomials is used for simplifying the calculation and improving the calculation efficiency;
s3: mapping the candidate target pixel point set into a thermal infrared image, and carrying out the following processing on each candidate target pixel point: constructing a candidate target pixel point in the thermal infrared imagen×nRectangular window and based onn×nRectangular window construction of vertical gray distribution vector and horizontal grayA distribution vector;
mapping the candidate target pixel point set to the infrared image means that the position coordinates of the candidate target pixel points in the gray image matrix are extracted, the extracted position coordinates are used as pixel point coordinates, and the corresponding pixel points in the infrared thermal image are extracted in the thermal infrared image according to the pixel point coordinates to be used as candidate target pixel points for subsequent processing;
s4: symmetrically intercepting Gao Sizi curves from standard Gaussian curves, constructing and obtaining Gaussian function vectors based on Gao Sizi curves, and carrying out similarity analysis on each candidate target pixel point based on the Gaussian function vectors to obtain a target pixel point set; the specific method for similarity analysis is as follows: calculating based on Gaussian function vectors to obtain the horizontal gray distribution similarity and the vertical gray distribution similarity of the candidate target pixel points, and if the horizontal gray distribution similarity of the candidate target pixel points is greater than or equal to a similarity threshold value and the vertical gray distribution similarity is greater than the similarity threshold value, identifying the candidate target pixel points as target pixel points;
s5: and constructing and obtaining a detection image of the infrared target based on the target pixel point set.
In the above technical solution, the image gradient map (i.e. saliency image) can be obtained by convolving the surface check gray image matrix, and if the saliency of the pixel is high, it indicates that the pixel may be the target pixel, i.e. considered as the candidate target pixel,
the surface kernel filtering has high processing speed, better background inhibition and target enhancement capability, and better candidate target detection capability and simultaneously ensures the real-time performance of the algorithm.
After obtaining candidate target pixel points, the principle of confirming the target pixel points by adopting Gaussian similarity is as follows: the gray scale of the thermal infrared target shows a characteristic of gradually attenuating from point to peripheral divergence, and shows a characteristic of approximate Gaussian distribution. Therefore, when the horizontal gray distribution and the vertical gray distribution of the thermal infrared target are observed, the gray distribution is similar to a Gaussian image, the gray distribution in the background is observed, only one direction of the gray distribution possibly shows the characteristic similar to the Gaussian image or the two directions do not meet the similar characteristic of the Gaussian image of the gray distribution, the non-target area does not meet the Gaussian image similarity of the gray distribution, and the candidate target pixel points are judged by utilizing the characteristic of the target area and the non-target area. Therefore, in the technical scheme provided by the invention, whether the candidate target pixel point is the target pixel point can be judged by adopting Gaussian similarity analysis.
After all the target pixel points are obtained, the detection image of the thermal infrared target can be obtained based on the combination of the target pixel points.
Further, in step S2, the method for obtaining the candidate target pixel point set includes:
and carrying out convolution enhancement on the gray image matrix and the surface kernel to obtain a mapping matrix, judging whether elements in the mapping matrix are larger than or equal to a threshold value, if so, taking pixel points corresponding to the elements as candidate target pixel points, and constructing a candidate target pixel point set based on all the candidate target pixel points.
The setting of the threshold value is calculated according to the formula (1):
Figure SMS_1
，
wherein:
Figure SMS_2
for threshold value->
Figure SMS_3
For mapping the average value of the elements in the matrix, +.>
Figure SMS_4
To map the standard deviation of the elements in the matrix,
Figure SMS_5
is a coefficient.
Further, in the step S4, the method for constructing the gaussian function vector based on the Gao Sizi curve is as follows: selecting rectangular window side lengths at equal intervals along transverse axis on Gao Sizi curvenSelecting a corresponding point, and acquiring an ordinate value of the selected pointAnd constructing the obtained ordinate value as a vector element to obtain a Gaussian function vector.
The Gao Sizi curve is a continuous curve and the Gao Sizi curve is symmetrical to the symmetry axis of the standard gaussian curve.
Further, the calculation of the lateral gray distribution similarity is as shown in formula (2):
Figure SMS_6
，
wherein:
Figure SMS_7
for the lateral gray distribution similarity +.>
Figure SMS_8
Is an element in the lateral gray distribution vector, +.>
Figure SMS_9
For the selected rectangular window side length, +.>
Figure SMS_10
Is the element mean value in the transverse gray vector, +.>
Figure SMS_11
Is an element in a Gaussian function vector, +.>
Figure SMS_12
Is the average value of all elements in the Gaussian function vector;
the calculation of the vertical gray distribution similarity is shown in a formula (3):
Figure SMS_13
，
wherein:
Figure SMS_14
for longitudinal gray level distribution similarity +.>
Figure SMS_15
Is an element in the longitudinal gray distribution vector, +.>
Figure SMS_16
Is the mean value of the elements in the vertical gray vector.
Further, in step S2, before filtering the gray-scale image matrix by using the surface check, a sequential statistical filtering method is further used to remove clutter salient points in the gray-scale image matrix.
And (3) carrying out sequential statistical filtering on the gray image matrix, and removing clutter salient points, so that the picture is smoother, and the accuracy of the surface kernel filtering is improved.
Further, in step S3, the construction process of the vertical gray distribution vector is as follows: acquisition ofn×nThe gray values of pixel points distributed along the longitudinal axis of the square window in the square window are used as vector elements to construct a longitudinal gray distribution vector; the construction process of the transverse gray level distribution vector comprises the following steps: acquisition ofn×nAnd constructing a transverse gray level distribution vector by taking gray level values distributed along the transverse axis of the square window in the square window as vector elements.
The gray values of the pixel points distributed along the vertical axis are sequentially arranged to obtain a vertical gray distribution vector, and a horizontal gray distribution vector can be obtained by the same method.
Based on another aspect of the present disclosure, there is provided an infrared target detection apparatus based on gaussian similarity, including:
and a data acquisition module: the method comprises the steps of acquiring a thermal infrared image to be detected and obtaining a gray image matrix;
and a screening module: the method comprises the steps of constructing a face kernel, and filtering a gray image matrix by using the face kernel to obtain a candidate target pixel point set;
vector construction module: the method is used for constructing and obtaining a horizontal gray distribution vector, a vertical gray distribution vector and a Gaussian function vector;
similarity analysis module: the method comprises the steps of performing similarity analysis on candidate target pixel points to obtain target pixel points;
and an image construction module: the method is used for constructing and obtaining a detection image of the infrared target based on the target pixel point set.
Further, the detection device further comprises a clutter removal module, and the clutter removal module is used for removing clutter salient points in the gray image matrix by adopting a sequential statistical filtering method.
According to the technical scheme, a data acquisition module is adopted to obtain a gray image matrix, a candidate target pixel point is obtained based on a screening module, a vector construction module is adopted to construct a horizontal gray distribution vector, a vertical gray distribution vector and a Gaussian function vector according to the coordinate position of the candidate target pixel point, a similarity analysis module is adopted to obtain a target pixel point based on the horizontal gray distribution vector, the vertical gray distribution vector and the Gaussian function vector, and finally an image construction module is adopted to construct a detection image of an infrared target based on the target pixel point.
Further, the detection device further comprises a clutter removal module, and the clutter removal module is used for removing clutter salient points in the gray image matrix by adopting a sequential statistical filtering method.
Based on a further aspect of the present description, a computer readable storage medium is provided, on which a computer program is stored, wherein the computer program, when executed by a processor, implements the steps of the method for detecting an infrared target based on gaussian similarity.
Based on a further aspect of the present description, a computer device is provided, the computer device comprising a processor, a memory, and a computer program stored on the memory and executable by the processor, wherein the computer program, when executed by the processor, implements the steps of the gaussian similarity based infrared target detection method.
Compared with the prior art, the invention has the beneficial effects that:
according to the infrared target detection method based on the Gaussian similarity, the image gradient map can be obtained by carrying out convolution on the surface check gray image matrix, if the pixel point has high significance, the pixel point can be the target pixel point, namely the candidate target pixel point is considered, then the Gaussian similarity analysis is carried out on the candidate target pixel point to determine the target pixel point, and finally the detection image of the thermal infrared target is constructed according to the target pixel point. The method provided by the invention fully utilizes the Gaussian similarity of the gray level distribution of the infrared target in the infrared image, effectively improves the target detection accuracy, has high surface kernel filtering processing speed, has better background suppression and target enhancement capability, and ensures the real-time performance of an algorithm while having better candidate target detection capability;
according to the infrared target detection device based on Gaussian similarity, a gray image matrix is obtained by adopting a data acquisition module, candidate target pixel points are obtained by adopting a screening module, then a horizontal gray distribution vector, a vertical gray distribution vector and a Gaussian function vector are constructed by adopting a vector construction module according to the coordinate positions of the candidate target pixel points, then a target pixel point is obtained by adopting a similarity analysis module based on the horizontal gray distribution vector, the vertical gray distribution vector and the Gaussian function vector, and finally a detection image of an infrared target is constructed by adopting an image construction module based on the target pixel points. The device provided by the invention has higher detection accuracy and better detection instantaneity.
Drawings
FIG. 1 is a flow chart of a detection method according to an embodiment of the invention;
FIG. 2 is a longitudinal gray scale profile and a transverse gray scale profile of a thermal infrared target according to an embodiment of the present invention;
FIG. 3 is a longitudinal gray scale profile and a transverse gray scale profile of a background according to an embodiment of the present invention
FIG. 4 is a schematic diagram of a Gao Sizi curve according to an embodiment of the invention;
FIG. 5 is a schematic diagram of detection results under different detection methods according to an embodiment of the present invention;
FIG. 6 is a schematic diagram of still another detection result under a different detection method according to an embodiment of the present invention;
fig. 7 is a schematic structural diagram of a detection device according to an embodiment of the invention.
Detailed Description
The following description of the embodiments of the present invention will be made in detail and with reference to the accompanying drawings, wherein it is apparent that the embodiments described are only some, but not all embodiments of the present invention. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to fall within the scope of the invention.
As shown in fig. 1, the present embodiment provides an infrared target detection method based on gaussian similarity, which includes the following steps:
s1: acquiring a thermal infrared image Z to be detected, and correspondingly arranging gray values of pixel points of the thermal infrared image Z and coordinate positions of the pixel points to obtain a gray image matrix X with the size of A multiplied by B;
s2: removing clutter salient points in the gray image matrix by adopting a sequential statistical filtering method; the sequential statistical filtering has high processing speed, so that the real-time performance of detection is enhanced, and a filtering window of the sequential statistical filtering in the embodiment is as follows:
Figure SMS_17
，
the smooth statistical filtering is to arrange non-zero pixel values except the center pixel point in 9 pixel points in the image in ascending order to obtain the maximum gray value except the center pixel point so as to obtain a gray image matrix with clutter salient points removed
Figure SMS_18
。
And constructing a face check, and filtering the gray image matrix by using the face check.
Facet Kernel(facial core) originFacet The model assumes that the gray-scale intensity surface of a pixel can be approximated by a binary cubic polynomial, but the least square method is used to approximate the intensity surface, which is computationally complex. It is computationally much easier if the intensity surface can be linearly combined from a stack of discrete orthogonal polynomials. The use of linear combinations of discrete orthogonal polynomials is to simplify the computation and increase the computational efficiency.
The present embodiment employs 5×5Facet KernelI.e. filtering with a 5 x 5 window, first set
Figure SMS_19
And->
Figure SMS_20
，/>
Figure SMS_21
，/>
Figure SMS_22
The corresponding discrete orthogonal polynomials are then respectively:
Figure SMS_23
，
Figure SMS_24
，
in the method, in the process of the invention,
Figure SMS_25
is->
Figure SMS_26
Corresponding set of discrete orthopolynomials, +.>
Figure SMS_27
Is an integer argument one, ++>
Figure SMS_28
Is->
Figure SMS_29
Corresponding set of discrete orthopolynomials, +.>
Figure SMS_30
Is an integer argument two;
for the obtained
Figure SMS_31
Is of the region of (2)Domain, neglecting the two-dimensional discrete orthogonal basis +.>
Figure SMS_32
Expressed as:
Figure SMS_33
，
Figure SMS_34
in the region of (2) pixel surface function +.>
Figure SMS_35
The method comprises the following steps:
Figure SMS_36
，
Figure SMS_37
for the position coordinates of the selected window, wherein the base coefficients +.>
Figure SMS_38
The method comprises the following steps:
Figure SMS_39
，
in the method, in the process of the invention,
Figure SMS_40
representative is->
Figure SMS_41
Gray value of the position.
Matrix inner quantity
Figure SMS_42
Can be expressed as:
Figure SMS_43
，
thus, the coefficient of the front of the base
Figure SMS_44
In practice it is possible to consist of the matrix content +.>
Figure SMS_45
And->
Figure SMS_46
Convolution results, expressed as:
Figure SMS_47
，
obtainable according to formulae (7) and (8)
Figure SMS_48
Window center pixel +.>
Figure SMS_49
Is a gradient of (2):
Figure SMS_50
，
Figure SMS_51
，
Figure SMS_52
，
wherein, the liquid crystal display device comprises a liquid crystal display device,
Figure SMS_53
and->
Figure SMS_54
Respectively indicate->
Figure SMS_55
Window center pixel +.>
Figure SMS_56
Second partial derivatives in the row and column directions.
Figure SMS_57
And->
Figure SMS_58
The +.>
Figure SMS_59
Rear and->
Figure SMS_60
Convolution is obtained, so->
Figure SMS_61
And->
Figure SMS_62
Can be obtained by solving the equation (14) and the equation (15):
Figure SMS_63
，
Figure SMS_64
，
Figure SMS_65
，/>
Figure SMS_66
the structure is the same, only transposed, so use +.>
Figure SMS_67
Go and->
Figure SMS_68
Convolution is performed. Thus->
Figure SMS_69
It is Facet Kernel:
Figure SMS_70
，
obtaining a candidate target pixel point set; the method for obtaining the candidate target pixel point set comprises the following steps:
matrix gray scale image
Figure SMS_71
And (2) are combined with the dough core>
Figure SMS_72
Performing convolution enhancement to obtain a mapping matrix->
Figure SMS_73
：
Figure SMS_74
，
Judging mapping matrix
Figure SMS_75
If the element in the set is greater than or equal to a threshold value, the pixel point corresponding to the element is a candidate target pixel point, and a candidate target pixel point set is constructed based on all the candidate target pixel points;
the calculation formula of the threshold value is as follows:
Figure SMS_76
，
wherein, the liquid crystal display device comprises a liquid crystal display device,
Figure SMS_77
and->
Figure SMS_78
Respectively->
Figure SMS_79
Mean and standard deviation of (a). After multiple experiments in this example, the coefficients +.>
Figure SMS_80
Set to 4, the test gave the best test result.
S3: mapping the candidate target pixel point set into a thermal infrared image, and carrying out the following processing on each candidate target pixel point: candidate in thermal infrared imagesConstructing a target pixel point as a centern×n(since the imaging size of infrared small targets is usually not more than 80 pixels, n takes a value of 9-15) rectangular window and is based onn×nRectangular window construction longitudinal gray scale distribution vector
Figure SMS_81
And lateral gray distribution vector->
Figure SMS_82
；
The construction process of the longitudinal gray scale distribution vector comprises the following steps: acquisition ofn×nThe gray values of pixel points distributed along the longitudinal axis of the square window in the square window are used as vector elements to be sequentially arranged in sequence to obtain a longitudinal gray distribution vector
Figure SMS_83
The method comprises the steps of carrying out a first treatment on the surface of the The construction process of the transverse gray level distribution vector comprises the following steps: acquisition ofn×nGray values distributed along the horizontal axis of the square window in the square window are used as vector elements to be sequentially arranged in sequence to construct a horizontal gray distribution vector +.>
Figure SMS_84
；
S4: symmetrically intercepting a Gao Sizi curve from a standard Gaussian curve, and constructing a Gaussian function vector based on the Gao Sizi curve;
since the thermal infrared target is not of a constant size, it is necessary to intercept three different Gao Sizi curves from a standard Gaussian curve according to different sizes, as shown in FIG. 4, the Gaussian curve is divided into 6 curve segments along the horizontal axis, namely curve segment (1) (-2.58)σ ~ -1.96σ) Curve segment (2) (1.96σ ~ 2.58σ) Curve segment (3) (-1.96)σ ~ -σ) Curve section (4)σ ~ 1.96σ) Curve segment (5)σ ~0) And curve segment (6) (0)~ σ) The three different Gao Sizi curves taken are: a first Gao Sizi curve (comprising curve segment (5) + curve segment (6)), a second Gao Sizi curve (comprising curve segment (5) + curve segment (6) + curve segment (4) + curve segment (3)), and a third Gao Sizi curve #Comprising a curve section (5) + a curve section (6) + a curve section (4) + a curve section (3) + a curve section (1) + a curve section (2)).
The construction method of the Gaussian function vector is as follows: in each Gaussian sub-curve case, rectangular window side lengths are selected from Gao Sizi curves at equal intervals along the transverse axisnSelecting a corresponding point, acquiring an ordinate value of the selected point, and constructing a Gaussian function vector by taking the acquired ordinate value as a vector element.
Carrying out similarity analysis on each candidate target pixel point based on the Gaussian function vector to obtain a target pixel point set; the gray scale of the thermal infrared target exhibits a characteristic of gradual attenuation from point to ambient dispersion and exhibits a characteristic of approximately gaussian distribution (as shown in fig. 2). Therefore, when the horizontal gray distribution and the vertical gray distribution of the thermal infrared target are observed, the gray distribution is similar to a Gaussian image, the gray distribution in the background is observed, only one direction of the gray distribution possibly shows similar characteristics with the Gaussian image or the two directions do not meet the similar characteristics of the Gaussian image of the gray distribution (as shown in fig. 3), the non-target area does not meet the Gaussian image similarity of the gray distribution, and then the candidate target pixel point is judged by utilizing the characteristics of the target area and the non-target area and selecting the Gaussian similarity.
The specific method for similarity analysis is as follows: calculating based on Gaussian function vectors to obtain the horizontal gray distribution similarity and the longitudinal gray distribution similarity of the candidate target pixel points, and if the horizontal gray distribution similarity of the candidate target pixel points is greater than or equal to a similarity threshold value and the longitudinal gray distribution similarity is greater than the similarity threshold value, identifying the candidate target pixel points as target pixel points;
the calculation formula of the lateral gray level distribution similarity is as follows:
Figure SMS_85
，
wherein:
Figure SMS_86
for the lateral gray distribution similarity +.>
Figure SMS_87
Is a transverse gray level distribution vector->
Figure SMS_88
Element of (a)>
Figure SMS_89
For the selected rectangular window side length, +.>
Figure SMS_90
Is the element mean value in the transverse gray vector, +.>
Figure SMS_91
As an element in the vector of the gaussian function,
Figure SMS_92
is the average value of all elements in the Gaussian function vector;
the calculation formula of the longitudinal gray distribution similarity is shown as formula (3):
Figure SMS_93
，
wherein:
Figure SMS_94
for longitudinal gray level distribution similarity +.>
Figure SMS_95
Is a longitudinal gray level distribution vector->
Figure SMS_96
Element of (a)>
Figure SMS_97
Is the mean value of the elements in the vertical gray vector.
In the present embodiment calculate respectively
Figure SMS_99
Three gaussian directions corresponding to three Gao Sizi curvesThe similarity of the quantities yields three corresponding lateral gray-scale distribution similarities +.>
Figure SMS_101
(corresponding to the first Gao Sizi curve), ->
Figure SMS_102
(corresponding to the second Gao Sizi curve) and +.>
Figure SMS_100
(corresponding to the third Gao Sizi curve) calculating +.>
Figure SMS_103
The similarity of the three gaussian function vectors corresponding to the three Gao Sizi curves yields three corresponding lateral gray distribution similarities +.>
Figure SMS_104
(corresponding to the first Gao Sizi curve), ->
Figure SMS_105
(corresponding to the second Gao Sizi curve) and +.>
Figure SMS_98
(corresponding to the third Gao Sizi curve).
And when the similarity meets the similarity threshold requirement, the candidate target pixel point is considered as the target pixel point. And (3) judging the target pixel point by adopting a formula (18):
Figure SMS_106
，
wherein:
Figure SMS_107
for similarity threshold, ++>
Figure SMS_108
Is a label drawing of infrared small target detection, < >>
Figure SMS_109
Taking 1 to represent candidate target pixel points (x, y) as target pixel points, and adding +.>
Figure SMS_110
Taking 0 as candidate target pixel points (x, y) as non-target pixel points.
S5: and constructing and obtaining a detection image of the infrared target based on the target pixel point set.
Will be
Figure SMS_111
Performing dot multiplication on the target detection result graph obtained by the thermal infrared image Z, and if the target detection result graph is the target, indicating that the target is detected, wherein the gray value of the target detection result graph is the original gray value of Z; otherwise, the gray value is 0, which represents the background, and the infrared small target detection result image is obtained. As shown in formula (19):
Figure SMS_112
，
wherein:
Figure SMS_113
is the corresponding pixel location.
In this embodiment, 5 sets of images of commonly used small infrared target datasets (date 1, date2, date3, MDFA, and sir) are used for identification, where targets are aircraft, automobiles, boats, etc., in different complex contexts. data1 consists of an image with a 599 Zhang Fenbian rate of 256×256 pixels, the background is a simpler sky background, and the target is two planes. data2 was 256×256 pixels at 399 Zhang Fenbian rate with heavy cloudy background clutter, targeting two aircraft. data3 consists of an image with a 499 Zhang Fenbian rate of 256×256 pixels, contains a complex farmland forest background, and targets an aircraft. MDFA consists of 100 images of small infrared targets, including various natural or synthetic small infrared targets. sirst consists of 427 Zhang Gongwai small target images, including 480 instances, which contain mostly only one target, and a minority of multiple targets.
Using FKRW (based on surface kernel filtering and random walk, respectivelyIPI (infrared small target detection algorithm based on infrared patch image model), NRAM (infrared small target detection algorithm based on non-convex rank approximation), NOLC (infrared small target detection algorithm based on non-convex optimization with Lp norm constraint), top-hat (Top hat filter algorithm) and the method (protected) provided by the invention test 5 groups of different infrared small target data sets, the data set detection time and the detection rate of the data sets are [ ]pd) The FAR is shown in table 1,F1 Scoreas shown in table 2, wherein:
Figure SMS_114
，
Figure SMS_115
，
Figure SMS_116
。
the shorter the infrared small target detection time is, the higher the real-time performance and efficiency of the algorithm are, and the target detection task can be completed in a shorter time.pdRepresenting the detection rate, a higher detection rate generally means that the algorithm performs better in detecting small targets. Lower FAR generally means that the algorithm performs better in detecting small targets. The F1 score ranges from 0 to 1, where 1 indicates perfect detection and 0 indicates that the model cannot detect the target.
TABLE 1 comparison of infrared small target detection indicators
Figure SMS_117
TABLE 2 comparison Table for infrared small target detection F1 Score
Figure SMS_118
From the data of tables 1 and 2, it can be seen that the object of the present invention is to ensure higherpdAt the same time have a relatively highThe low FAR and the F1 score are relatively high, and the detection time of the method has obvious advantages compared with other methods with similar indexes, and is superior to other methods.
Fig. 5 and fig. 6 are diagrams of small target images detected by the detection method provided in this embodiment and five other existing detection methods, where Image is an original thermal infrared Image and groudtruth is a real result diagram. The small target in fig. 5 is an airplane, the background is simple, and the air and the ground are provided. The small object of fig. 6 is an airplane, the background is complex, and the building effect is achieved.
The method has the advantages that the small target detection in the complex background in the infrared image is verified, clutter salient points can be removed by utilizing sequential statistical filtering, target pixel candidate points are extracted through Facet Kernel, and finally candidate target points are further screened through multi-situation Gaussian similarity, so that the Gaussian similarity of gray distribution of the target in the infrared image is fully utilized, the target detection accuracy is effectively improved, and FAR is reduced, and the method is feasible in infrared small target detection.
As shown in fig. 7, this embodiment further provides an infrared target detection device based on gaussian similarity, including:
and a data acquisition module: the method comprises the steps of acquiring a thermal infrared image to be detected and obtaining a gray image matrix;
clutter module: the method is used for removing clutter salient points in the gray image matrix by adopting a sequential statistical filtering method;
and a screening module: the method comprises the steps of constructing a face kernel, and filtering a gray image matrix by using the face kernel to obtain candidate target pixel points;
vector construction module: the method is used for constructing and obtaining a horizontal gray distribution vector, a vertical gray distribution vector and a Gaussian function vector; the method is specifically used for completing the following steps: mapping the candidate target pixel point set into a thermal infrared image, and carrying out the following processing on each candidate target pixel point: constructing a candidate target pixel point in the thermal infrared imagen×nRectangular window and based onn×nThe rectangular window constructs a longitudinal gray scale distribution vector and a transverse gray scale distribution vector;
similarity analysis module: the method comprises the steps of performing similarity analysis on candidate target pixel points to obtain target pixel points; the method is specifically used for completing the following steps: symmetrically intercepting Gao Sizi curves from standard Gaussian curves, constructing and obtaining Gaussian function vectors based on Gao Sizi curves, and carrying out similarity analysis on each candidate target pixel point based on the Gaussian function vectors to obtain a target pixel point set; the specific method for similarity analysis is as follows: calculating based on Gaussian function vectors to obtain the horizontal gray distribution similarity and the longitudinal gray distribution similarity of the candidate target pixel points, and if the horizontal gray distribution similarity of the candidate target pixel points is greater than or equal to a similarity threshold value and the longitudinal gray distribution similarity is greater than the similarity threshold value, identifying the candidate target pixel points as target pixel points;
and an image construction module: the method is used for constructing and obtaining a detection image of the infrared target based on the target pixel point set.
Further, the screening module is further configured to complete the following steps: and carrying out convolution enhancement on the gray image matrix and the surface kernel to obtain a mapping matrix, judging whether elements in the mapping matrix are larger than or equal to a threshold value, if so, taking pixel points corresponding to the elements as candidate target pixel points, and constructing a candidate target pixel point set based on all the candidate target pixel points.
Further, the similarity analysis module is further configured to complete the following steps: selecting rectangular window side lengths at equal intervals along transverse axis on Gao Sizi curvenSelecting a corresponding point, acquiring an ordinate value of the selected point, and constructing a Gaussian function vector by taking the acquired ordinate value as a vector element.
Further, the vector construction module is further configured to complete the following steps: acquisition ofn×nThe gray values of pixel points distributed along the longitudinal axis of the square window in the square window are used as vector elements to construct a longitudinal gray distribution vector; acquisition ofn×nAnd constructing a transverse gray level distribution vector by taking gray level values distributed along the transverse axis of the square window in the square window as vector elements.
The present embodiment also provides a computer readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the gaussian similarity based infrared target detection method.
The computer readable storage medium may be an internal storage unit of the computer device according to the foregoing embodiment, for example, a hard disk or a memory of the computer device. The computer readable storage medium may also be an external storage device of the computer device, such as a plug-in hard disk, a Smart Media Card (SMC), a Secure Digital (SD) Card, a Flash memory Card (Flash Card) or the like, which are provided on the computer device.
The embodiment also provides a computer device, which may be an industrial personal computer, a server or a computer terminal.
The computer device comprises a processor, a memory, and a computer program stored on the memory and executable by the processor, wherein the computer program, when executed by the processor, implements the steps of the gaussian similarity based infrared target detection method.
The computer device includes a processor, a memory, and a network interface connected by a system bus, where the memory may include a non-volatile storage medium and an internal memory.
The non-volatile storage medium may store an operating system and a computer program. The computer program comprises program instructions that, when executed, cause the processor to perform any one of a number of infrared target detection methods based on gaussian similarity.
The processor is used to provide computing and control capabilities to support the operation of the entire computer device.
The internal memory provides an environment for the execution of a computer program in a non-volatile storage medium that, when executed by a processor, causes the processor to perform any one of a number of infrared target detection methods based on gaussian similarity.
The network interface is used for network communication such as transmitting assigned tasks and the like.
It should be appreciated that the processor may be a central processing unit (Central Processing Unit, CPU), but may also be other general purpose processors, digital signal processors (Digital Signal Processor, DSP), application specific integrated circuits (Application Specific Integrated Circuit, ASIC), field-programmable gate arrays (Field-Programmable Gate Array, FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, or the like. Wherein the general purpose processor may be a microprocessor or the processor may be any conventional processor or the like.
Wherein in one embodiment the processor is configured to run a computer program stored in the memory to implement the steps of:
s1: acquiring a thermal infrared image to be detected, and obtaining a gray image matrix;
s2: constructing a face kernel, and filtering by using the face kernel gray level image matrix to obtain a candidate target pixel point set;
s3: mapping the candidate target pixel point set into a thermal infrared image, and carrying out the following processing on each candidate target pixel point: constructing a candidate target pixel point in the thermal infrared imagen×nRectangular window and based onn×nThe rectangular window constructs a longitudinal gray scale distribution vector and a transverse gray scale distribution vector;
s4: symmetrically intercepting Gao Sizi curves from standard Gaussian curves, constructing and obtaining Gaussian function vectors based on Gao Sizi curves, and carrying out similarity analysis on each candidate target pixel point based on the Gaussian function vectors to obtain a target pixel point set; the specific method for similarity analysis is as follows: calculating based on Gaussian function vectors to obtain the horizontal gray distribution similarity and the longitudinal gray distribution similarity of the candidate target pixel points, and if the horizontal gray distribution similarity of the candidate target pixel points is greater than or equal to a similarity threshold value and the longitudinal gray distribution similarity is greater than the similarity threshold value, identifying the candidate target pixel points as target pixel points;
s5: and constructing and obtaining a detection image of the infrared target based on the target pixel point set.
Finally, it should be noted that: the above embodiments are only for illustrating the technical solution of the present invention, and are not limiting; although the invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical scheme described in the foregoing embodiments can be modified or some or all of the technical features thereof can be replaced with equivalents; these modifications or substitutions do not depart from the essence of the corresponding technical solutions from the technical solutions of the embodiments of the present invention.

Claims (10)

1. The infrared target detection method based on Gaussian similarity is characterized by comprising the following steps of:
s1: acquiring a thermal infrared image to be detected, and obtaining a gray image matrix;
s2: constructing a face kernel, and filtering by using the face kernel gray level image matrix to obtain a candidate target pixel point set;
s3: mapping the candidate target pixel point set into a thermal infrared image, and carrying out the following processing on each candidate target pixel point: constructing a candidate target pixel point in the thermal infrared imagen×nRectangular window and based onn×nThe rectangular window constructs a longitudinal gray scale distribution vector and a transverse gray scale distribution vector;
s4: symmetrically intercepting Gao Sizi curves from standard Gaussian curves, constructing and obtaining Gaussian function vectors based on Gao Sizi curves, and carrying out similarity analysis on each candidate target pixel point based on the Gaussian function vectors to obtain a target pixel point set; the specific method for similarity analysis is as follows: calculating based on Gaussian function vectors to obtain the horizontal gray distribution similarity and the vertical gray distribution similarity of the candidate target pixel points, and if the horizontal gray distribution similarity of the candidate target pixel points is greater than or equal to a similarity threshold value and the vertical gray distribution similarity is greater than the similarity threshold value, identifying the candidate target pixel points as target pixel points;
s5: and constructing and obtaining a detection image of the infrared target based on the target pixel point set.
2. The method for detecting an infrared target based on gaussian similarity according to claim 1, wherein in step S2, the method for obtaining the candidate target pixel set is as follows:
and carrying out convolution enhancement on the gray image matrix and the surface kernel to obtain a mapping matrix, judging whether elements in the mapping matrix are larger than or equal to a threshold value, if so, taking pixel points corresponding to the elements as candidate target pixel points, and constructing a candidate target pixel point set based on all the candidate target pixel points.
3. The method for detecting an infrared target based on gaussian similarity according to claim 1, wherein in step S4, the method for constructing a gaussian function vector based on a Gao Sizi curve is as follows: selecting rectangular window side lengths at equal intervals along a transverse axis on a Gao Sizi curve, selecting corresponding points, acquiring longitudinal coordinate values of the selected points, and constructing the acquired longitudinal coordinate values as vector elements to obtain Gaussian function vectors.
4. The method for detecting an infrared target based on gaussian similarity according to claim 3, wherein the calculation of the lateral gray level distribution similarity is as shown in formula (2):
Figure QLYQS_1
，
wherein:
Figure QLYQS_2
for the lateral gray distribution similarity +.>
Figure QLYQS_3
Is an element in the lateral gray distribution vector, +.>
Figure QLYQS_4
For the selected rectangular window side length, +.>
Figure QLYQS_5
Is the element mean value in the transverse gray vector, +.>
Figure QLYQS_6
Is an element in a Gaussian function vector, +.>
Figure QLYQS_7
Is the average value of all elements in the Gaussian function vector;
the calculation of the vertical gray distribution similarity is shown in a formula (3):
Figure QLYQS_8
，
wherein:
Figure QLYQS_9
for longitudinal gray level distribution similarity +.>
Figure QLYQS_10
Is an element in the longitudinal gray distribution vector, +.>
Figure QLYQS_11
Is the mean value of the elements in the vertical gray vector.
5. The method according to claim 1, wherein in step S2, clutter salient points in the gray image matrix are removed by a sequential statistical filtering method before filtering by using the face check gray image matrix.
6. The method for detecting an infrared target based on gaussian similarity according to claim 1, wherein in step S3, the construction process of the vertical gray scale distribution vector is as follows: acquisition ofn×nThe gray values of pixel points distributed along the longitudinal axis of the square window in the square window are used as vector elements to construct a longitudinal gray distribution vector; the construction process of the transverse gray level distribution vector comprises the following steps: acquisition ofn×nAnd constructing a transverse gray level distribution vector by taking gray level values distributed along the transverse axis of the square window in the square window as vector elements.
7. An infrared target detection device based on gaussian similarity, for implementing the steps of the infrared target detection method based on gaussian similarity according to any of claims 1 to 6, comprising:
and a data acquisition module: the method comprises the steps of acquiring a thermal infrared image to be detected and obtaining a gray image matrix;
and a screening module: the method comprises the steps of constructing a face kernel, and filtering a gray image matrix by using the face kernel to obtain a candidate target pixel point set;
vector construction module: the method is used for constructing and obtaining a horizontal gray distribution vector, a vertical gray distribution vector and a Gaussian function vector;
similarity analysis module: the method comprises the steps of performing similarity analysis on candidate target pixel points to obtain target pixel points;
and an image construction module: the method is used for constructing and obtaining a detection image of the infrared target based on the target pixel point set.
8. The infrared target detection device based on Gaussian similarity according to claim 7, further comprising a clutter removal module, wherein the clutter removal module is configured to remove clutter salient points in the gray image matrix by using a sequential statistical filtering method.
9. A computer readable storage medium, characterized in that the storage medium has stored thereon a computer program, wherein the computer program, when executed by a processor, implements the steps of the gaussian similarity based infrared target detection method according to any of claims 1 to 6.
10. A computer device comprising a processor, a memory, and a computer program stored on the memory and executable by the processor, wherein the computer program when executed by the processor implements the steps of the gaussian similarity based infrared target detection method according to any of claims 1 to 6.
CN202310637222.3A 2023-06-01 2023-06-01 Infrared target detection method, device, medium and equipment based on Gaussian similarity Active CN116363135B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202310637222.3A CN116363135B (en) 2023-06-01 2023-06-01 Infrared target detection method, device, medium and equipment based on Gaussian similarity

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202310637222.3A CN116363135B (en) 2023-06-01 2023-06-01 Infrared target detection method, device, medium and equipment based on Gaussian similarity

Publications (2)

Publication Number Publication Date
CN116363135A true CN116363135A (en) 2023-06-30
CN116363135B CN116363135B (en) 2023-09-12

Family

ID=86934835

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202310637222.3A Active CN116363135B (en) 2023-06-01 2023-06-01 Infrared target detection method, device, medium and equipment based on Gaussian similarity

Country Status (1)

Country Link
CN (1) CN116363135B (en)

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102567994A (en) * 2011-12-31 2012-07-11 南京理工大学 Infrared small target detection method based on angular point gaussian characteristic analysis
CN110288368A (en) * 2019-05-08 2019-09-27 江苏聚垚科技有限公司 A kind of method that compartmentalization passes through vehicles advertisement
CN113822352A (en) * 2021-09-15 2021-12-21 中北大学 Infrared dim target detection method based on multi-feature fusion
CN115409778A (en) * 2022-07-29 2022-11-29 南京航空航天大学 Threshold segmentation method for image after infrared small target background suppression

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102567994A (en) * 2011-12-31 2012-07-11 南京理工大学 Infrared small target detection method based on angular point gaussian characteristic analysis
CN110288368A (en) * 2019-05-08 2019-09-27 江苏聚垚科技有限公司 A kind of method that compartmentalization passes through vehicles advertisement
CN113822352A (en) * 2021-09-15 2021-12-21 中北大学 Infrared dim target detection method based on multi-feature fusion
CN115409778A (en) * 2022-07-29 2022-11-29 南京航空航天大学 Threshold segmentation method for image after infrared small target background suppression

Also Published As

Publication number Publication date
CN116363135B (en) 2023-09-12

Similar Documents

Publication Publication Date Title
CN103413151B (en) Hyperspectral image classification method based on figure canonical low-rank representation Dimensionality Reduction
CN109919870B (en) SAR image speckle suppression method based on BM3D
CN107560592B (en) Precise distance measurement method for photoelectric tracker linkage target
WO2014149115A2 (en) Reduction of cfar false alarms via classification and segmentation of sar image clutter
Ince et al. Evolutionary RBF classifier for polarimetric SAR images
Xiang et al. Adaptive superpixel generation for SAR images with linear feature clustering and edge constraint
Shaoqing et al. The comparative study of three methods of remote sensing image change detection
CN109544494B (en) Fusion method of passive millimeter wave image and visible light image in human body security check
Zhang et al. Infrared small target detection via interpatch correlation enhancement and joint local visual saliency prior
Peng et al. Infrared small-target detection based on multi-directional multi-scale high-boost response
CN109712149A (en) A kind of image partition method based on wavelet energy and fuzzy C-mean algorithm
CN106874932A (en) SAR target model recognition methods based on rapid sparse description
CN111861905A (en) SAR image speckle noise suppression method based on Gamma-Lee filtering
CN110490848B (en) Infrared target detection method, device and computer storage medium
Wang et al. A deep deformable residual learning network for SAR image segmentation
CN114373135A (en) Ship target detection method based on local significance characteristic measurement
Munawar et al. Isotropic surround suppression based linear target detection using hough transform
Peng et al. CourtNet: Dynamically balance the precision and recall rates in infrared small target detection
CN112598032B (en) Multi-task defense model construction method for anti-attack of infrared image
Ye et al. Optical and SAR image fusion based on complementary feature decomposition and visual saliency features
CN116363135B (en) Infrared target detection method, device, medium and equipment based on Gaussian similarity
Zhang et al. Learning nonlocal quadrature contrast for detection and recognition of infrared rotary-wing UAV targets in complex background
JP4230383B2 (en) Target detection device
CN111859266A (en) Spatial target structure inversion method and device based on multi-source information fusion
Xiao et al. Object detection algorithm combined with dynamic and static for air target intrusion

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