Nothing Special   »   [go: up one dir, main page]

CN113658191A - Infrared dim target detection method based on local probability hypergraph dissimilarity measure - Google Patents

Infrared dim target detection method based on local probability hypergraph dissimilarity measure Download PDF

Info

Publication number
CN113658191A
CN113658191A CN202110756141.6A CN202110756141A CN113658191A CN 113658191 A CN113658191 A CN 113658191A CN 202110756141 A CN202110756141 A CN 202110756141A CN 113658191 A CN113658191 A CN 113658191A
Authority
CN
China
Prior art keywords
node
dissimilarity
nodes
edge
hypergraph
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
CN202110756141.6A
Other languages
Chinese (zh)
Other versions
CN113658191B (en
Inventor
杨小冈
卢瑞涛
张涛
夏克寒
范继伟
刘闯
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.)
Rocket Force University of Engineering of PLA
Original Assignee
Rocket Force University of Engineering of PLA
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 Rocket Force University of Engineering of PLA filed Critical Rocket Force University of Engineering of PLA
Priority to CN202110756141.6A priority Critical patent/CN113658191B/en
Publication of CN113658191A publication Critical patent/CN113658191A/en
Application granted granted Critical
Publication of CN113658191B publication Critical patent/CN113658191B/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
    • G06T7/10—Segmentation; Edge detection
    • G06T7/11—Region-based segmentation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/136—Segmentation; Edge detection involving thresholding
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/194—Segmentation; Edge detection involving foreground-background 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
    • G06T2207/10048—Infrared image
    • 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
    • G06T2207/20—Special algorithmic details
    • G06T2207/20076—Probabilistic image processing

Landscapes

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

Abstract

The invention discloses an infrared dim target detection method based on local probability hypergraph dissimilarity measurement, which introduces a probability hypergraph model, designs a brand-new measurement mode called probability hypergraph dissimilarity and improves the description robustness; then, multi-scale local probability hypergraph dissimilarity measurement is constructed, and a target area and a background area are effectively distinguished; finally, effective detection of the target is achieved through adaptive threshold segmentation.

Description

Infrared dim target detection method based on local probability hypergraph dissimilarity measure
Technical Field
The invention belongs to the field of image processing, and particularly relates to an infrared small target detection method based on local probability hypergraph dissimilarity measurement.
Background
The detection of the infrared small and weak moving target is the research characteristics and difficulty in the field of computer vision, and has wide application in a plurality of fields such as early warning and air defense, image guidance, disaster relief and exploration, security and protection inspection and the like. Since the aircraft is usually away from the target by several kilometers, the imaging pixels of the target are usually within dozens or even dozens of pixels, and there is no shape, texture or structural information in the infrared image, so that the target is difficult to describe effectively. In addition, due to different imaging mechanisms of the infrared images, more background interferences, such as noise interference, strong edge interference, clutter interference, highlight region interference, and the like, exist in the images, which causes an extremely low signal-to-noise ratio, and brings great difficulty and challenge to target detection.
The current weak and small target detection algorithms include an algorithm based on the detection of a single frame image and a detection algorithm of a plurality of frame images. The multi-frame detection algorithm is based on the result of single-frame detection for subsequent processing, and the single-frame detection algorithm has the advantages of simple structure, high calculation efficiency, easiness in implementation and the like, so that the method is paid more attention by researchers. The current detection algorithm based on a single frame mainly comprises three categories: algorithms based on traditional filtering, methods based on matrix decomposition and methods based on human vision. The algorithm based on the traditional filtering utilizes a filtering algorithm to perform filtering operation on an image in a space domain and a frequency domain, so as to enhance a target and inhibit a background, and the commonly used algorithms comprise median filtering, mean filtering, morphological filtering and the like. The algorithm can usually obtain a better structure under a simple background or a priori knowledge condition, but has a poor detection effect on a complex scene. The algorithm based on matrix decomposition usually adopts a block diagram representation mode to express the infrared image, background clutter interference has certain correlation and therefore has low rank, and a target has sparsity and therefore can eliminate the background clutter interference by a matrix decomposition method. Such algorithms suffer from a large false detection rate in response to strong edge and texture interference. The detection algorithm based on human vision is based on the central contrast characteristic of a target, the detection is carried out by utilizing the brightness contrast of a central region and a surrounding neighborhood, the classical algorithms comprise LCM, ILCM, WLDM, MPCM and the like, the calculation efficiency of the algorithms is high, but the algorithms mostly adopt Euclidean distance for measurement, and the measurement is unreliable in a complex scene.
Disclosure of Invention
The invention provides an infrared dim target detection method based on local probability hypergraph dissimilarity measure, and aims to solve the technical problem of improving the detection effect of infrared dim targets in a complex scene.
In order to achieve the purpose, the invention adopts the following technical scheme:
an infrared dim target detection method based on local probability hypergraph dissimilarity measurement comprises the following steps:
s1: constructing a probability hypergraph model as G ═ V, E and w, wherein V represents a node set, E represents a hyper-edge set, and w is a hyper-edge weight; and designing two nodes v in the probabilistic hypergraph modeliAnd vjDescription of dissimilarity between ds (v)i,vj)；
S2: obtaining the dissimilarity description between the central node and the neighboring node as ds (v) according to the dissimilarity description between the two nodes in step S10,vi) Constructing dissimilarity operator D (v) of central node and neighborhood node0,vi)＝ds(v0,vi)·ω(v0,vi) Where ω (v)0,vi) The super edge weight ratio between the central node and the neighborhood node; constructing local probability hypergraph model dissimilarity measure
Figure BDA0003147564370000021
Wherein, two nodes viAnd vi’Along a central node v0Symmetry, D (v)0,vi') As a central node and a node vi’Of a dissimilarity operator of (a) is two nodes viAnd vi’The number of nodes that differ;
s3: performing sliding window operation on the whole image by using the dissimilarity measurement of the local probability hypergraph model to obtain a saliency map; performing maximum pooling operation on the significance maps of multiple scales to obtain a final significance map;
s4: and (4) segmenting the final saliency map based on an adaptive threshold segmentation algorithm, and judging the region larger than the threshold as an infrared weak target region.
The invention also comprises the following technical characteristics:
specifically, the step S1 includes:
step S1.1, defining a probabilistic hypergraph model as G ═ V, E, w, where V denotes a node set, E denotes a hyperedge set, and w denotes a weight of the hyperedge; the membership of nodes and hyperedges is as follows:
Figure BDA0003147564370000022
wherein
Figure BDA0003147564370000023
Indicating a super edge
Figure BDA0003147564370000024
The average intensity of the light beam of (a),
Figure BDA0003147564370000025
vgindicating a super edge
Figure BDA0003147564370000026
Is connected to the network node in the network,
Figure BDA0003147564370000027
indicating a super edge
Figure BDA0003147564370000028
The number of contained nodes, f (-) represents the characteristics corresponding to the nodes, namely the corresponding gray value in the image;
Figure BDA00031475643700000214
representing a node viAnd the super edge
Figure BDA0003147564370000029
Similarity of (2), super edge
Figure BDA00031475643700000210
Is formed by node vlAnd its nearest k nodes together,
Figure BDA00031475643700000215
convertible into node viAnd the super edge
Figure BDA00031475643700000211
Node v oflThe distance between them, is expressed as | f (v)i)-f(vl) L and λ are weighting coefficients, usually taken as 2;
Figure BDA00031475643700000212
wherein v isiIt represents the (i) th node of the (i) th node,
Figure BDA00031475643700000213
is shown as
Figure BDA00031475643700000216
Strip overlap;
step S1.2, designing the dissimilarity description between two nodes in the probability hypergraph model as follows:
ds(vi,vj)＝|uxi|1·|xj|1-|xi|1·|uxj|1 (2-1)
wherein
Figure BDA0003147564370000031
Is the feature vector of the node-super edge,
Figure BDA0003147564370000032
for incremental feature vectors, vector xmAnd uxmThe dimension of (A) is N; in the above formulas (2-2) and (2-3),
Figure BDA00031475643700000311
degree of over-edge, xmThe ith element of (c) contains three parts:
Figure BDA00031475643700000312
Figure BDA0003147564370000033
and
Figure BDA0003147564370000034
specifically, the method comprises the following steps: first item
Figure BDA00031475643700000313
Representing a node vmAnd the super edge
Figure BDA0003147564370000035
The distance of (d); second item
Figure BDA0003147564370000036
Shows a super edge
Figure BDA0003147564370000037
Can be compacted by
Figure BDA0003147564370000038
Calculating the standard deviation of the reference point; item III
Figure BDA0003147564370000039
Is a super edge
Figure BDA00031475643700000310
Average intensity of (d).
Specifically, the step 3 includes: performing sliding window operation on the whole image by using the local probability dissimilarity measurement of a specific scale to obtain a saliency map under the specific scale; performing maximum pooling operation on the saliency maps of multiple scales to obtain a final saliency map LPHDMm(r,c)＝max(LPHDMs(r, c)), s ═ 1.., L, where s denotes a scale parameter and L is a set maximum scale; and (r, c) are the abscissa and ordinate in the image.
Specifically, the step S4 includes: final significance map LPHDM using adaptive threshold valuesmCarrying out segmentation, and detecting weak and small targets by the segmentation; the adaptive threshold T is represented as: t ═ k × μ + (1-k) × σ, where μ denotes the mean of the final saliency map and σ denotes the final saliency mapThe variance of the graph, k, is given as a set parameter and is set to 0.5-0.8; after segmentation, the remaining area is the infrared weak target area.
Compared with the prior art, the invention has the beneficial technical effects that:
the invention designs an infrared dim target detection method based on local probability hypergraph dissimilarity measurement, introduces a probability hypergraph model, designs a brand new measurement mode called probability hypergraph dissimilarity, and improves the description robustness; then, multi-scale local probability hypergraph dissimilarity measurement is constructed, a target area is effectively enhanced, a background area is inhibited, and the target area and the background area are effectively distinguished; finally, effective detection of the target is realized through self-adaptive threshold segmentation; the method solves the problem of how to improve the detection effect of the infrared weak and small moving target under the complex dynamic background of the open space.
Drawings
FIG. 1 is a flow chart of a method according to an embodiment of the present invention.
FIG. 2 is a diagram illustrating a comparison between a simple diagram and a hypergraph.
FIG. 3 is a diagram of a super edge and a node in a local window.
Fig. 4 is a comparison of the detection results of the five algorithms.
Detailed Description
The invention designs an infrared dim target detection method based on local probability hypergraph dissimilarity measurement, introduces a probability hypergraph model, designs a brand new measurement mode called probability hypergraph dissimilarity, and improves the description robustness; then, multi-scale local probability hypergraph dissimilarity measurement is constructed, and a target area and a background area are effectively distinguished; finally, effective detection of the target is achieved through adaptive threshold segmentation.
The following describes in detail specific embodiments of the present invention. It should be understood that the detailed description and specific examples, while indicating the present invention, are given by way of illustration and explanation only, not limitation.
A method for detecting infrared dim targets based on local probability hypergraph dissimilarity measure is shown in figure 1 and comprises the following steps:
s1: constructing a probability hypergraph model as G ═ V, E and w, wherein V represents a node set, E represents a hyper-edge set, and w is a hyper-edge weight; and designing two nodes v in the probabilistic hypergraph modeliAnd vjDescription of dissimilarity between ds (v)i,vj) (ii) a The robustness of difference description between nodes is effectively improved; step S1 specifically includes:
step S1.1, constructing a probability hypergraph model:
defining a probability hypergraph model as G ═ (V, E, w), wherein V represents a node set, E represents a hyper-edge set, and w is the weight of the hyper-edge; generally speaking, the description of the hypergraph is more robust and efficient than the simple graph, and fig. 2 gives a relational description of the simple graph and the hypergraph.
The membership of nodes and hyperedges is as follows:
Figure BDA0003147564370000041
wherein
Figure BDA0003147564370000042
Indicating a super edge
Figure BDA0003147564370000043
The average intensity of the light beam of (a),
Figure BDA0003147564370000044
vgindicating a super edge
Figure BDA0003147564370000045
Is connected to the network node in the network,
Figure BDA0003147564370000046
indicating a super edge
Figure BDA0003147564370000047
The number of contained nodes, f (-) represents the characteristics corresponding to the nodes, namely the corresponding gray value in the image;
Figure BDA00031475643700000411
representing a node viAnd the super edge
Figure BDA0003147564370000048
Similarity of (2), super edge
Figure BDA0003147564370000049
Is formed by node vlAnd its nearest k nodes together,
Figure BDA00031475643700000412
convertible into node viAnd the super edge
Figure BDA00031475643700000410
Node v oflThe distance between them, is expressed as | f (v)i)-f(vl) L and λ are weighting coefficients, usually taken as 2;
Figure BDA0003147564370000051
wherein v isiIt represents the (i) th node of the (i) th node,
Figure BDA0003147564370000052
is shown as
Figure BDA00031475643700000511
Strip overlap;
in the embodiment, an effective probability hypergraph model is constructed to capture the local characteristics of the target, and the hyperedges are defined as a central node and K nodes which are nearest to the central node (in the invention, K is generally 5-15)
Therefore, the membership matrix of all nodes and hyperedges in the probabilistic hypergraph model can be obtained:
Figure BDA0003147564370000053
Hkthe membership matrix of all nodes and super edges is shown, the | V | represents the number of nodes, and the | E | represents the number of super edges;
step S1.2, designing dissimilarity description among nodes:
the dissimilarity between two nodes in the probabilistic hypergraph is described as:
Figure BDA0003147564370000054
wherein
Figure BDA0003147564370000055
Represents the distance between two super edges, defined as
Figure BDA0003147564370000056
Based on
Figure BDA0003147564370000057
The description of dissimilarity between the two nodes of the above equation can be continued by:
Figure BDA0003147564370000058
wherein
Figure BDA0003147564370000059
Is the feature vector of the node-super edge,
Figure BDA00031475643700000510
for incremental feature vectors, vector xmAnd uxmThe dimension of (A) is N; in the above formulas (2-2) and (2-3),
Figure BDA00031475643700000512
for the degree of excess edge, the degree of excess edge is defined
Figure BDA00031475643700000513
Standard deviation distributed for each node within the super-edge:
Figure BDA0003147564370000061
wherein l is p or q; x is the number ofmThe first element ofThe element comprises three parts:
Figure BDA0003147564370000062
specifically, the method comprises the following steps: first item
Figure BDA00031475643700000610
Representing a node vmAnd the super edge
Figure BDA0003147564370000063
The distance of (d); second item
Figure BDA0003147564370000064
Shows a super edge
Figure BDA0003147564370000065
Can be compacted by
Figure BDA0003147564370000066
Calculating the standard deviation of the reference point; item III
Figure BDA0003147564370000067
Is a super edge
Figure BDA0003147564370000068
Average intensity of (d).
S2: performing spatial division on a local region of the image, obtaining the dissimilarity between a central pixel and surrounding neighborhood pixels of the image after the spatial division according to a dissimilarity calculation method between two nodes in the probability hypergraph model in the step S1, and introducing a weight coefficient to construct a dissimilarity operator of the central pixel and the surrounding neighborhood pixels; overlapping dissimilarity operators in the same direction, performing product operation on dissimilarity operators in different directions, and constructing local probability dissimilarity measurement;
step S2 specifically includes:
s21: obtaining dissimilarity description of the central node and the neighborhood nodes according to the dissimilarity description between the two nodes in the step S1, and constructing dissimilarity operators of the central node and the neighborhood nodes;
subject to local metric detectionEnlightening an algorithm, namely capturing the structural relationship between a central pixel and surrounding neighborhood pixels by adopting a sliding window of 3P multiplied by 3P; in the constructed probability hypergraph model, each node corresponds to each pixel in a window, and each hyperedge comprises a central node and K nodes which are nearest on the gray level of the central node; to sufficiently enhance the local characteristics of the target while satisfying the computation load, the center node v is selected0And 8 nodes v around it1～v8As shown in fig. 3; the dissimilarity operator of the central node and the peripheral nodes is defined as:
D(v0,vi)＝ds(v0,vi)·ω(v0,vi)
wherein ds (v)0,vi) Representing a central node v0And neighborhood node viLocal probability hypergraph dissimilarity between, ω (v)0,vi) Representing a weight coefficient between the central node and the neighborhood node; omega (v)0,vi) Expressed as all containing nodes v0And the contained node viThe ratio of the gray level mean of the super edge:
Figure BDA0003147564370000069
ω(v0,vi) The high-order constraint relation of the hypergraph is fully utilized, the center contrast characteristic of the target is obviously enhanced, and clutter interference of the background is effectively inhibited.
S22: generally, the gray level of a target node is higher than that of surrounding neighborhood nodes, so that for a specific direction, dissimilarity operators of a central node and the neighborhood nodes can be superposed to enhance the directivity of an infrared dim target, and multiplication operation is performed on all directions to suppress background interference, so that the constructed local probability hypergraph dissimilarity metric is obtained
Figure BDA0003147564370000071
Wherein, two nodes viAnd vi’Along a central node v0Symmetry, D (v)0,vi') As a central node and a node vi’The dissimilarity operator of (1), Δ being two nodesviAnd vi’The number of nodes that differ; more specifically, in the present embodiment, the center node v is selected0And 8 nodes v around it1～v8To calculate, therefore:
Figure BDA0003147564370000072
wherein, two nodes viAnd vi+4Along a central node v0Symmetry, D (v)0,vi+4) As a central node and a node vi+4The dissimilarity operator of (1);
s3: performing sliding window operation on the whole image by using the local probability dissimilarity measurement of a specific scale to obtain a saliency map under the specific scale; performing maximum pooling operation on the significance maps of multiple scales to obtain a final significance map;
step S3 specifically includes:
in order to guarantee the validity of the proposed local metric when dealing with target scale variations, the size of the window should coincide as much as possible with the target scale. Therefore, for a local probability hypergraph dissimilarity measure of a certain size, a saliency map at that size can be obtained by means of a sliding window. Then for the significance map obtained at different scales, the final significance map can be obtained by maximum pooling (max boosting) operation:
LPHDMm(r,c)＝max(LPDMs(r,c)),s＝1,...,L
wherein s represents a scale parameter, and s is 0.5 (P-1), L is a set maximum scale, and in the invention, L is 2, so that the maximum value of the scale parameter s corresponding to L is 5; r, c are the abscissa and ordinate in the image.
S4: and (4) segmenting the final saliency map based on an adaptive threshold segmentation algorithm, and determining the region larger than the threshold as a target region. Step S4 specifically includes:
final significance map LPHDM using adaptive threshold pairsmAnd (3) carrying out segmentation to segment and detect the weak and small targets, wherein the adaptive threshold T can be expressed as: t ═ k × μ + (1-k) × σ; where μ represents the mean of the final saliency map and σ represents the final saliencyThe variance, k, of the graph is given as a set parameter, typically set to 0.5-0.8. After segmentation, the remaining regions are considered to be true infrared weak targets.
Further illustrated by the following simulations:
1. simulation conditions
To test the effectiveness of the present invention, simulation verification and comparative evaluation were performed on 4 sets of truly acquired infrared image sequences. The present invention contrasts four popular weak and small target detection algorithms, including RLCM, MLHM, MPCM, and WLDM. All simulation environments are Matlab R2016b, and the hardware operating platform is a notebook computer with 1.9-GHz i7 processor and 16GB memory.
2. Simulation experiment
Fig. 4 shows typical test results of all five test algorithms in 4 sets of image sequences. It can be seen from the figure that the test image contains serious background interference, such as background highlight regions, prominent edges, and strong corner points. Compared with other detection algorithms, the method has the advantage that a better detection result is obtained. The weak and small targets are effectively enhanced, and meanwhile, the background interference is well suppressed. RLCM gave satisfactory results in both Seq1 and Seq 4. Although MLHM and MPCM also achieve unusual results in Seq1, Seq3 and Seq4, there are significant background noise points on the three-dimensional gray scale distributions of these two algorithms.

Claims (4)

1. A method for detecting infrared dim targets based on local probability hypergraph dissimilarity measurement is characterized by comprising the following steps:
s1: constructing a probability hypergraph model as G ═ V, E and w, wherein V represents a node set, E represents a hyper-edge set, and w is a hyper-edge weight; and designing two nodes v in the probabilistic hypergraph modeliAnd vjDescription of dissimilarity between ds (v)i,vj)；
S2: obtaining the dissimilarity description between the central node and the neighboring node as ds (v) according to the dissimilarity description between the two nodes in step S10,vi) Constructing dissimilarity operator D (v) of central node and neighborhood node0,vi)＝ds(v0,vi)·ω(v0,vi) Where ω (v)0,vi) The super edge weight ratio between the central node and the neighborhood node; constructing local probability hypergraph model dissimilarity measure
Figure FDA0003147564360000011
Wherein, two nodes viAnd vi’Along a central node v0Symmetry, D (v)0,vi') As a central node and a node vi’Of a dissimilarity operator of (a) is two nodes viAnd vi’The number of nodes that differ;
s3: performing sliding window operation on the whole image by using the dissimilarity measurement of the local probability hypergraph model to obtain a saliency map; performing maximum pooling operation on the significance maps of multiple scales to obtain a final significance map;
s4: and (4) segmenting the final saliency map based on an adaptive threshold segmentation algorithm, and judging the region larger than the threshold as an infrared weak target region.
2. The method for detecting infrared dim target based on the difference measure of local probability hypergraph as claimed in claim 1, wherein said step S1 comprises:
step S1.1, defining a probabilistic hypergraph model as G ═ V, E, w, where V denotes a node set, E denotes a hyperedge set, and w denotes a weight of the hyperedge; the membership of nodes and hyperedges is as follows:
Figure FDA0003147564360000012
wherein
Figure FDA0003147564360000013
Indicating a super edge
Figure FDA0003147564360000014
The average intensity of the light beam of (a),
Figure FDA0003147564360000015
vgindicating a super edge
Figure FDA0003147564360000016
Is connected to the network node in the network,
Figure FDA0003147564360000017
indicating a super edge
Figure FDA0003147564360000018
The number of contained nodes, f (-) represents the characteristics corresponding to the nodes, namely the corresponding gray value in the image;
Figure FDA00031475643600000112
representing a node viAnd the super edge
Figure FDA0003147564360000019
Similarity of (2), super edge
Figure FDA00031475643600000110
Is formed by node vlAnd its nearest k nodes together,
Figure FDA00031475643600000113
convertible into node viAnd the super edge
Figure FDA00031475643600000111
Node v oflThe distance between them, is expressed as | f (v)i)-f(vl) L and λ are weighting coefficients, usually taken as 2;
Figure FDA0003147564360000021
wherein v isiIt represents the (i) th node of the (i) th node,
Figure FDA0003147564360000022
is shown as
Figure FDA00031475643600000215
Strip overlap;
step S1.2, designing the dissimilarity description between two nodes in the probability hypergraph model as follows:
ds(vi,vj)＝|uxi|1·|xj|1-|xi|1·|uxj|1 (2-1)
wherein
Figure FDA0003147564360000023
Is the feature vector of the node-super edge,
Figure FDA0003147564360000024
for incremental feature vectors, vector xmAnd uxmThe dimension of (A) is N; in the above formulas (2-2) and (2-3),
Figure FDA00031475643600000216
degree of over-edge, xmThe ith element of (c) contains three parts:
Figure FDA0003147564360000025
Figure FDA0003147564360000026
and
Figure FDA0003147564360000027
specifically, the method comprises the following steps: first item
Figure FDA0003147564360000028
Representing a node vmAnd the super edge
Figure FDA0003147564360000029
The distance of (d); second item
Figure FDA00031475643600000210
Shows a super edge
Figure FDA00031475643600000211
Can be compacted by
Figure FDA00031475643600000212
Calculating the standard deviation of the reference point; item III
Figure FDA00031475643600000213
Is a super edge
Figure FDA00031475643600000214
Average intensity of (d).
3. The method for detecting infrared dim targets based on the dissimilarity measure of local probabilistic hypergraph as claimed in claim 1, wherein said step 3 comprises: performing sliding window operation on the whole image by using the local probability dissimilarity measurement of a specific scale to obtain a saliency map under the specific scale; performing maximum pooling operation on the saliency maps of multiple scales to obtain a final saliency map LPHDMm(r,c)＝max(LPHDMs(r, c)), s ═ 1.., L, where s denotes a scale parameter and L is a set maximum scale; and (r, c) are the abscissa and ordinate in the image.
4. The method for detecting infrared dim target based on the difference measure of local probability hypergraph as claimed in claim 1, wherein said step S4 comprises: final significance map LPHDM using adaptive threshold valuesmCarrying out segmentation, and detecting weak and small targets by the segmentation; the adaptive threshold T is represented as: t ═ k × μ + (1-k) × σ, where μ denotes the mean of the final saliency map, σ denotes the variance of the final saliency map, and k is a set parameter and is set to 0.5-0.8; after segmentation, the remaining area is the infrared weak target area.
CN202110756141.6A 2021-07-05 2021-07-05 Infrared dim target detection method based on local probability hypergraph dissimilarity measure Active CN113658191B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110756141.6A CN113658191B (en) 2021-07-05 2021-07-05 Infrared dim target detection method based on local probability hypergraph dissimilarity measure

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110756141.6A CN113658191B (en) 2021-07-05 2021-07-05 Infrared dim target detection method based on local probability hypergraph dissimilarity measure

Publications (2)

Publication Number Publication Date
CN113658191A true CN113658191A (en) 2021-11-16
CN113658191B CN113658191B (en) 2022-09-30

Family

ID=78477936

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110756141.6A Active CN113658191B (en) 2021-07-05 2021-07-05 Infrared dim target detection method based on local probability hypergraph dissimilarity measure

Country Status (1)

Country Link
CN (1) CN113658191B (en)

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108022244A (en) * 2017-11-30 2018-05-11 东南大学 A kind of hypergraph optimization method for being used for well-marked target detection based on foreground and background seed
CN109272489A (en) * 2018-08-21 2019-01-25 西安电子科技大学 Inhibit the method for detecting infrared puniness target with multiple dimensioned local entropy based on background
CN109767439A (en) * 2019-01-10 2019-05-17 中国科学院上海技术物理研究所 A kind of multiple dimensioned difference of self-adapting window and the object detection method of bilateral filtering

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108022244A (en) * 2017-11-30 2018-05-11 东南大学 A kind of hypergraph optimization method for being used for well-marked target detection based on foreground and background seed
CN109272489A (en) * 2018-08-21 2019-01-25 西安电子科技大学 Inhibit the method for detecting infrared puniness target with multiple dimensioned local entropy based on background
CN109767439A (en) * 2019-01-10 2019-05-17 中国科学院上海技术物理研究所 A kind of multiple dimensioned difference of self-adapting window and the object detection method of bilateral filtering

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
王晓阳等: "局部对比度结合区域显著性红外弱小目标检测", 《强激光与粒子束》 *
谷雨等: "基于改进多尺度分形特征的红外图像弱小目标检测", 《光学精密工程》 *

Also Published As

Publication number Publication date
CN113658191B (en) 2022-09-30

Similar Documents

Publication Publication Date Title
CN109934815B (en) Tensor recovery infrared small target detection method combined with ATV constraint
CN109447073B (en) Infrared weak and small target detection method based on tensor robust principal component analysis
CN103942557B (en) A kind of underground coal mine image pre-processing method
CN109934178A (en) A kind of method for detecting infrared puniness target based on Kronecker base rarefaction representation
CN107403134B (en) Local gradient trilateral-based image domain multi-scale infrared dim target detection method
CN109215025B (en) Infrared weak and small target detection method based on non-convex rank approach minimization
CN108038856B (en) Infrared small target detection method based on improved multi-scale fractal enhancement
CN111768350B (en) Infrared image enhancement method and system
Li et al. A small target detection algorithm in infrared image by combining multi-response fusion and local contrast enhancement
CN113449658A (en) Night video sequence significance detection method based on spatial domain, frequency domain and time domain
CN112329764A (en) Infrared dim target detection method based on TV-L1 model
Juliu et al. Dim and small target detection based on improved spatio-temporal filtering
CN112395944A (en) Multi-scale ratio difference combined contrast infrared small target detection method based on weighting
CN110033006B (en) Vehicle detection tracking method based on color feature nonlinear dimension reduction
CN112163606B (en) Infrared small target detection method based on block contrast weighting
Li et al. Robust small infrared target detection using weighted adaptive ring top-hat transformation
CN113658191B (en) Infrared dim target detection method based on local probability hypergraph dissimilarity measure
CN110363105B (en) Method for inhibiting speckle of fully-polarized SAR image
CN111461999A (en) SAR image speckle suppression method based on super-pixel similarity measurement
CN111340741A (en) Particle swarm optimization gray level image enhancement method based on quaternion and L1 norm
Xiao et al. Underwater image classification based on image enhancement and information quality evaluation
CN115205216A (en) Infrared small target detection method based on significance and weighted guide filtering
CN116051444A (en) Effective infrared and visible light image self-adaptive fusion method
CN115731135A (en) Hyperspectral image denoising method and system based on low-rank tensor decomposition and adaptive graph total variation
CN110033477B (en) Road vehicle LBP feature-dependent filtering tracking method suitable for occlusion scene

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