CN115511006A - Target detection method and device based on unitary transformation, electronic equipment and storage medium - Google Patents

Target detection method and device based on unitary transformation, electronic equipment and storage medium Download PDF

Info

Publication number
CN115511006A
CN115511006A CN202211339176.0A CN202211339176A CN115511006A CN 115511006 A CN115511006 A CN 115511006A CN 202211339176 A CN202211339176 A CN 202211339176A CN 115511006 A CN115511006 A CN 115511006A
Authority
CN
China
Prior art keywords
tensor
target
space
time
unitary transformation
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.)
Pending
Application number
CN202211339176.0A
Other languages
Chinese (zh)
Inventor
董晓刚
余杭
彭真明
朱勇
董纯柱
包醒东
毛宏霞
王龙
马晨
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.)
Beijing Institute of Environmental Features
Original Assignee
Beijing Institute of Environmental Features
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 Beijing Institute of Environmental Features filed Critical Beijing Institute of Environmental Features
Priority to CN202211339176.0A priority Critical patent/CN115511006A/en
Publication of CN115511006A publication Critical patent/CN115511006A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
    • G06F17/10—Complex mathematical operations
    • G06F17/14—Fourier, Walsh or analogous domain transformations, e.g. Laplace, Hilbert, Karhunen-Loeve, transforms
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
    • G06F17/10—Complex mathematical operations
    • G06F17/16—Matrix or vector computation, e.g. matrix-matrix or matrix-vector multiplication, matrix factorization

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Mathematical Physics (AREA)
  • Pure & Applied Mathematics (AREA)
  • Mathematical Analysis (AREA)
  • Mathematical Optimization (AREA)
  • Computational Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Theoretical Computer Science (AREA)
  • Algebra (AREA)
  • Databases & Information Systems (AREA)
  • Software Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Computing Systems (AREA)
  • Image Analysis (AREA)

Abstract

The present invention relates to the field of target detection technologies, and in particular, to a target detection method and apparatus based on unitary transformation, and a storage medium. The method comprises the following steps: constructing a space-time tensor of the multi-frame infrared image, wherein the space-time tensor comprises a background tensor and a target tensor; constructing an objective function of the space-time tensor, wherein the objective function is to utilize a tensor nuclear norm under a unitary transformation domain to constrain the background tensor and utilize a joint space-time total variation sum L 1 The norm is obtained by constraining the target tensor; solving the target function to obtain the target tensor; and reconstructing the solved target tensor into a plurality of single-frame target images, and outputting the detection result of each target image. The method can accurately detect the target in the complex background and has strong detection capability.

Description

Target detection method and device based on unitary transformation, electronic equipment and storage medium
Technical Field
The present invention relates to the field of target detection technologies, and in particular, to a target detection method and apparatus based on unitary transformation, and a storage medium.
Background
The Infrared detection technology has the characteristics of strong anti-interference capability, all-weather operation and the like, so an Infrared search and tracking system (IRST) is widely applied to the military and civil fields. The infrared target detection is used as a basic function in an IRST system and plays an important role in the aspects of space reconnaissance, disaster rescue and the like.
In the related art, when the imaging environment of a target is complex, the existing target detection method cannot accurately detect the target, and the detection capability is poor.
Therefore, a need exists for a unitary transform-based target detection method to solve the above-mentioned technical problems.
Disclosure of Invention
Based on the problems of low target detection accuracy and poor detection capability of the existing target detection method, the embodiment of the invention provides a unitary transformation-based target detection method, a unitary transformation-based target detection device and a unitary transformation-based storage medium, which can accurately detect a target in a complex background and have strong detection capability.
In a first aspect, an embodiment of the present invention provides a target detection method based on unitary transformation, including:
constructing a space-time tensor of the multi-frame infrared image, wherein the space-time tensor comprises a background tensor and a target tensor;
constructing an objective function of the space-time tensor, wherein the objective function is to utilize a tensor nuclear norm under a unitary transformation domain to constrain the background tensor and utilize a unionSpace time total variation sum L 1 The norm is obtained by constraining the target tensor;
solving the target function to obtain the target tensor;
and reconstructing the solved target tensor into a plurality of single-frame target images, and outputting the detection result of each target image.
In one possible design, the objective function is:
Figure BDA0003915784840000021
Figure BDA0003915784840000022
in the formula,
Figure BDA0003915784840000023
for the purpose of the said background tensor,
Figure BDA0003915784840000024
for the purpose of the said target tensor,
Figure BDA0003915784840000025
is the space-time tensor in question,
Figure BDA0003915784840000026
in the case of random noise, the noise level is,
Figure BDA0003915784840000027
is a tensor kernel norm in unitary transform domain, numerically equal to
Figure BDA0003915784840000028
The modulo three fiber of (a) is multiplied by the unitary transformation matrix a to obtain the sum of the nuclear norms of all the frontal slices of the new tensor,
Figure BDA0003915784840000029
is L of tensor 1 The number of the norm is calculated,
Figure BDA00039157848400000210
is the norm of Frobenius,
Figure BDA00039157848400000211
is a space-time total variation of lambda 1 、λ 2 And λ 3 In order to balance the coefficients of the process,
Figure BDA00039157848400000212
the value range of p is 1-10, lambda 2 The value range of (a) is 0.01-0.1, lambda 3 The value range of (1) is 100-200, m is the maximum value of the length and width of each frame of the infrared image, and l is the total frame number of the infrared image.
In one possible design, solving the objective function to solve the target tensor comprises:
constructing the unitary transformation matrix by using a zero-frequency component of the space-time tensor time domain;
and solving the objective function based on the unitary transformation matrix to obtain the target tensor.
In one possible design, the constructing the unitary transformation matrix with zero frequency components of the space-time tensor time domain includes:
respectively performing one-dimensional Fourier transform on m multiplied by n modulus three fibers of the space-time tensor to obtain a first space-time tensor, wherein m and n are respectively the length and the width of the infrared image;
reserving a zero-frequency component in the first space-time tensor to obtain a second space-time tensor;
respectively performing one-dimensional inverse Fourier transform on the m multiplied by n modulus three fibers of the second space-time tensor to obtain a third space-time tensor;
and performing singular value decomposition on the modulus three-fiber expansion matrix of the third space-time tensor, and using the conjugate transpose of the obtained left singular matrix as the unitary transformation matrix.
In one possible design, solving the objective function based on the unitary transformation matrix to solve the target tensor comprises:
introducing auxiliary variables
Figure BDA00039157848400000213
And
Figure BDA00039157848400000214
reducing the objective function to a first objective function represented by:
Figure BDA00039157848400000215
Figure BDA00039157848400000216
in the formula,
Figure BDA00039157848400000217
is a space-time total variation operator;
the augmented Lagrangian function of the first objective function is:
Figure BDA0003915784840000031
in the formula, y 1 ,y 2 ,y＝[y v ,y h ,y t ]Beta is Lagrange multiplier, and represents a penalty factor;
and solving the Lagrangian function to obtain the target tensor.
In one possible design, the solving the lagrangian function to solve the target tensor comprises:
for the background tensor, in k +1 iterations, let
Figure BDA0003915784840000032
For is to
Figure BDA0003915784840000033
Is multiplied by the unitary transformation matrix A to obtain each module three fibers
Figure BDA0003915784840000034
To pair
Figure BDA0003915784840000035
Performing singular value contraction processing on each front slice to obtain an updated front slice
Figure BDA0003915784840000036
Then
Figure BDA0003915784840000037
The calculation formula of (c) is:
Figure BDA0003915784840000038
in the formula, A H A conjugate transpose matrix of the unitary transformation matrix a, i, j being natural numbers greater than 0, respectively;
for the target tensor, in k +1 iterations,
Figure BDA0003915784840000039
the calculation formula of (a) is as follows:
Figure BDA00039157848400000310
wherein,
Figure BDA00039157848400000311
sign () is a sign function;
in response to reaching a preset iteration stop condition, stopping computing and outputting the target tensor.
In one possible design, the preset iteration stop condition is: the number of iterations reaches a preset maximum number of iterations or
Figure BDA00039157848400000312
In a second aspect, an embodiment of the present invention further provides a target detection apparatus based on unitary transformation, including:
the device comprises a first construction module, a second construction module and a third construction module, wherein the first construction module is used for constructing a space-time tensor of a multi-frame infrared image, and the space-time tensor comprises a background tensor and a target tensor;
a second constructing module, configured to construct an objective function of the space-time tensor, where the objective function is to constrain the background tensor by using a tensor kernel norm in a unitary transformation domain and to use a joint space-time total variation sum L 1 The norm is obtained by constraining the target tensor;
the solving module is used for solving the objective function to obtain the target tensor;
and the reconstruction output module is used for reconstructing the solved target tensor into a plurality of single-frame target images and outputting the detection result of each target image.
In a third aspect, an embodiment of the present invention further provides a computing device, including a memory and a processor, where the memory stores a computer program, and the processor, when executing the computer program, implements the method described in any embodiment of this specification.
In a fourth aspect, the present invention further provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed in a computer, the computer program causes the computer to execute the method described in any embodiment of the present specification.
The embodiment of the invention provides a target detection method and device based on unitary transformation, electronic equipment and a storage medium. According to the method, the multi-frame infrared image is used for constructing the space-time tensor, data processing can be carried out in a high-dimensional space, the data structure of the original infrared image is reserved, time domain information of the infrared image sequence is fully utilized, more image information is obtained, and prior information such as the shape and the motion trail of the target is obtained. Then, the background tensor is constrained based on the tensor nuclear norm under the unitary transformation domain and the joint space-time full transformation is utilizedIs divided by L 1 The norm restrains the target tensor so as to construct a target function of the space-time tensor; then, solving the objective function to obtain the objective tensor. In the two steps, the low rank of the space-time tensor in the infrared image is described by using the tensor nuclear norm based on unitary transformation, and the lower tensor rank can be obtained compared with the tensor nuclear norm based on Fourier transformation. The space-time total variation is used for constraining the target tensor, the space and time continuity of the target tensor is fully described, the internal smoothness of the target tensor is enhanced, and the detection performance in a complex scene is improved. Therefore, the method can accurately detect the target in the complex background and has strong detection capability.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below, and it is obvious that the drawings in the following description are some embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to these drawings without creative efforts.
Fig. 1 is a flowchart of a target detection method based on unitary transformation according to an embodiment of the present invention;
FIG. 2 is a flow chart of target detection using the method of FIG. 1;
FIG. 3 (a) is an infrared image containing a small target;
FIG. 3 (b) is a three-dimensional distribution diagram of the infrared image shown in FIG. 3 (a);
FIG. 4 (a) is a target image obtained by performing the target detection of FIG. 3 (a) by using the method of the present invention;
FIG. 4 (b) is a three-dimensional distribution diagram of the target image shown in FIG. 4 (a);
fig. 5 (a) is a target image obtained by performing target detection on fig. 3 (a) by using a Local Contrast Measure (LCM) method;
FIG. 5 (b) is a three-dimensional distribution diagram of the target image shown in FIG. 5 (a);
fig. 6 is a target Image obtained by performing target detection on the Image (a) in fig. 3 by using an Infrared Patch-Image (IPI) method;
FIG. 6 (b) is a three-dimensional distribution diagram of the target image shown in FIG. 6 (a);
fig. 7 is a target image obtained by performing target detection on the image (a) in fig. 3 by using a weighted not left patch (RIPT) method;
FIG. 7 (b) is a three-dimensional distribution diagram of the target image shown in FIG. 7 (a);
FIG. 8 is a target image obtained by performing target detection on the image of FIG. 3 (a) by using Partial Sum of the sensor Nuclear Norm (PSTNN);
FIG. 8 (b) is a three-dimensional distribution diagram of the target image shown in FIG. 8 (a);
FIG. 9 is a diagram of a hardware architecture of a computing device provided by an embodiment of the invention;
fig. 10 is a block diagram of an object detection apparatus according to a unitary transform according to an embodiment of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the embodiments of the present invention clearer and more complete, the technical solutions in the embodiments of the present invention will be described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are some, but not all, embodiments of the present invention, and based on the embodiments of the present invention, all other embodiments obtained by a person of ordinary skill in the art without creative efforts belong to the scope of the present invention.
Specific implementations of the above concepts are described below.
Referring to fig. 1, an embodiment of the present invention provides a unitary transform-based target detection method, including:
step 100, constructing a space-time tensor of a multi-frame infrared image, wherein the space-time tensor comprises a background tensor and a target tensor;
step 102, constructing an objective function of the space-time tensor, wherein the objective function is to utilize a tensor nuclear norm under a unitary transformation domain to constrain a background tensor and utilize a combined space-time total variation sumL 1 The norm is obtained by constraining a target tensor;
step 104, solving an objective function to obtain a target tensor;
and 106, reconstructing the solved target tensor into a plurality of single-frame target images, and outputting the detection result of each target image.
The embodiment of the invention provides a target detection method based on unitary transformation. According to the method, the space-time tensor is constructed by using the multi-frame infrared images, data processing can be carried out in a high-dimensional space, the data structure of the original infrared image is reserved, and the time domain information of the infrared image sequence is fully utilized, so that more image information is obtained, and the prior information such as the shape, the motion trail and the like of the target is obtained. Then, the background tensor is constrained based on the tensor nuclear norm under the unitary transformation domain, and the joint space-time total variation sum L is utilized 1 The norm restrains the target tensor so as to construct a target function of the space-time tensor; then, solving the objective function to obtain the objective tensor. In the two steps, the low rank of the space-time tensor in the infrared image is described by using the tensor nuclear norm based on unitary transformation, and the lower tensor rank can be obtained compared with the tensor nuclear norm based on Fourier transformation. The space-time total variation is used for constraining the target tensor, the space and time continuity of the target tensor is fully described, the internal smoothness of the target tensor is enhanced, and the detection performance in a complex scene is improved. Therefore, the method can accurately detect the target in the complex background and has strong detection capability.
In this embodiment, the targets in the infrared image may be targets of various sizes, and a good detection effect is also obtained for small targets.
The manner in which the various steps shown in fig. 1 are performed is described below.
First, for step 100, a space-time tensor of a multi-frame infrared image is constructed, where the space-time tensor includes a background tensor and a target tensor.
In the step, the number of the plurality of frames of infrared images is at least 3, and the sampling time interval between two adjacent frames of infrared images is smaller than the preset time interval. Therefore, time domain information of the infrared image sequence can be fully utilized, continuity of the target in space and time domain is fully utilized, and a more accurate detection result is obtained. Of course, it is preferable that the plurality of infrared images are consecutive infrared images.
In this step, the target tensor and the background tensor are unknown tensors, and an objective function of the space-time tensor needs to be constructed to solve the two tensors.
Then, for step 102, an objective function of the space-time tensor is constructed, wherein the objective function is to utilize a tensor nuclear norm number under a unitary transformation domain to constrain the background tensor and utilize a joint space-time total variation sum L 1 The norm is obtained by constraining the target tensor.
In the step, tensor nuclear norm based on unitary transformation is used for describing low rank of a space-time tensor in an infrared image, a unitary transformation matrix used is related to a zero-frequency component of a time dimension in the space-time tensor, and a lower tensor rank is obtained compared with the tensor nuclear norm based on Fourier transformation. The space-time total variation is used for constraining the target tensor, the space and time continuity of the target tensor is fully described, the internal smoothness of the target tensor is enhanced, and the detection performance in a complex scene is improved. Compared with the infrared small target detection method which uses the total variation and restrains the background tensor, the method can effectively reduce the running time.
In some embodiments, the objective function is:
Figure BDA0003915784840000071
Figure BDA0003915784840000072
in the formula,
Figure BDA0003915784840000073
for the purpose of the background tensor,
Figure BDA0003915784840000074
is the tensor of interest and,
Figure BDA0003915784840000075
is a space-time tensor that is,
Figure BDA0003915784840000076
in the form of a random noise, the noise is,
Figure BDA0003915784840000077
is a tensor kernel norm in unitary transform domain, numerically equal to
Figure BDA0003915784840000078
The modulo three fiber of (a) is multiplied by the unitary transformation matrix a to obtain the sum of the nuclear norms of all the front slices of the new tensor,
Figure BDA0003915784840000079
is L of tensor 1 The norm of the number of the first-order-of-arrival,
Figure BDA00039157848400000710
is the Frobenius norm,
Figure BDA00039157848400000711
is a space-time total variation of lambda 1 、λ 2 And λ 3 In order to balance the coefficients of the process,
Figure BDA00039157848400000712
p ranges from 1 to 10, lambda 2 The value range of (A) is 0.01-0.1, lambda 3 The value range of (1) is 100-200, m is the maximum value of the length and width of each frame of infrared image, and l is the total frame number of the infrared image.
Then, in step 104, the objective function is solved to obtain the objective tensor.
Of course, by solving the objective function, the background tensor can also be solved.
In some embodiments, solving the objective function to solve the target tensor comprises:
constructing a unitary transformation matrix by using a zero-frequency component of a space-time tensor time domain;
and solving an objective function based on the unitary transformation matrix to obtain an objective tensor.
In some embodiments, constructing a unitary transformation matrix using zero frequency components of a space-time tensor time domain includes:
respectively carrying out one-dimensional Fourier transform on m multiplied by n modular triples of the space-time tensor to obtain a first space-time tensor, wherein m and n are respectively the length and the width of the infrared image;
reserving a zero-frequency component in the first space-time tensor to obtain a second space-time tensor;
respectively performing one-dimensional inverse Fourier transform on the m multiplied by n modular three fibers of the second space-time tensor to obtain a third space-time tensor; the third space-time tensor can be an estimate of a background tensor;
and performing singular value decomposition on a mode three-fiber expansion matrix of the third space-time tensor, and taking the conjugate transpose of the obtained left singular matrix as a unitary transformation matrix.
In some embodiments, solving the objective function based on the unitary transformation matrix to solve the target tensor comprises:
introducing auxiliary variables
Figure BDA0003915784840000081
And
Figure BDA0003915784840000082
reducing the objective function to a first objective function represented by:
Figure BDA0003915784840000083
Figure BDA0003915784840000084
in the formula,
Figure BDA0003915784840000085
is a space-time total variation operator, which comprises three components, respectively D v 、D h And D t Three points ofThe quantities represent the difference operators in the vertical, horizontal and temporal directions, respectively.
The augmented Lagrangian function of the first objective function is:
Figure BDA0003915784840000086
in the formula, y 1 ,y 2 ,y＝[y v ,y h ,y t ]Beta represents a penalty factor for lagrange multipliers;
and solving a Lagrange function to obtain a target tensor.
In some embodiments, solving the lagrangian function to solve the target tensor comprises:
for the background tensor, in k +1 iterations, let
Figure BDA0003915784840000087
To pair
Figure BDA0003915784840000088
Each module of the three-fiber is multiplied by a unitary transformation matrix A to obtain
Figure BDA0003915784840000089
For is to
Figure BDA00039157848400000810
Performing singular value contraction processing on each front slice to obtain an updated front slice
Figure BDA00039157848400000811
Taking the singular value contraction processing on the lth front slice as an example:
to pair
Figure BDA00039157848400000812
The L-th front slice is subjected to singular value decomposition to obtain a left singular matrix U L Right singular matrix V L And diagonal matrix sigma L And performing singular value contraction on the front section to obtain
Figure BDA00039157848400000813
Figure BDA00039157848400000814
Go through
Figure BDA00039157848400000815
All the front slices are subjected to singular value contraction processing to obtain updated front slices
Figure BDA00039157848400000816
Thereby obtaining
Figure BDA00039157848400000817
The calculation formula of (2):
Figure BDA00039157848400000818
in the formula, A H A conjugate transpose matrix of the unitary transformation matrix A, wherein i and j are natural numbers larger than 0 respectively;
for the target tensor, in k +1 iterations,
Figure BDA00039157848400000819
the calculation formula of (c) is as follows:
Figure BDA00039157848400000820
wherein,
Figure BDA00039157848400000821
sign () is a sign function;
and stopping calculating and outputting the target tensor in response to reaching the preset iteration stop condition. Of course, after the iterative computation is completed, the user may also solve the background tensor as needed.
In some embodiments, the preset iteration stop condition is: the number of iterations reaches a preset maximum number of iterations or
Figure BDA0003915784840000091
Finally, in step 106, the solved target tensor is reconstructed into a plurality of single-frame target images, and the detection result of each target image is output.
The advantageous effects of the process according to the invention are illustrated below in a specific example.
Referring to fig. 2, which is a flow chart of the method of the present invention, the number of ir images is 3, and each ir image has a size of 256 × 256, wherein one ir image containing a small target and its three-dimensional distribution map are shown in fig. 3 (a) and 3 (b). In the infrared image, the target center is located at (145, 115). The imaging background is complex, and the background comprises forests, grasses, farmlands and the like. Due to the existence of a lot of strong radiation and noise from the ground, besides the target, a lot of high-brightness regions and high-brightness noise points exist in the image, the high-brightness regions can interfere the detection of the target, and the target is difficult to detect without the help of interframe information in the infrared image sequence.
The inventors processed the infrared image shown in fig. 3 (a) by the method of the present invention, the LCM method, the IPI method, the RIPT method, and the PSTNN method, respectively, and then detected the results as shown in fig. 4 (a), 4 (b) to 8 (a), and 8 (b), respectively. As can be seen from fig. 4 (a) and 4 (b), the background is effectively suppressed and only the target remains in the image by using the method of the present invention. As can be seen from fig. 5 (a) and 5 (b), the images processed by the LCM method have a significant "blocky" effect and are very sensitive to noise. As can be seen from fig. 6 (a), 6 (b) to 8 (a) and 8 (b), in the detection results of the IPI method, the RIPT method and the PSTNN method, except for the small target, other pixel points also have higher gray values, which easily causes a false alarm, and these three methods are difficult to completely solve the influence of the highlight noise point and the clutter interference on the target.
Therefore, the method and the device can well reduce the influence of noise and highlight areas by constructing the space-time tensor and restraining the target tensor in time by utilizing the space-time total variation, effectively utilize time domain information and further detect the target.
As shown in fig. 9 and 10, an embodiment of the present invention provides an object detection apparatus based on unitary transformation. The device embodiments may be implemented by software, or by hardware, or by a combination of hardware and software. From a hardware level, as shown in fig. 9, a hardware architecture diagram of a computing device where an object detection apparatus based on unitary transformation according to an embodiment of the present invention is located is provided, where the computing device where the apparatus is located in the embodiment may generally include other hardware, such as a forwarding chip responsible for processing a packet, in addition to the processor, the memory, the network interface, and the nonvolatile memory shown in fig. 9. Taking a software implementation as an example, as shown in fig. 10, as a logical means, the device is formed by reading a corresponding computer program in a non-volatile memory into a memory by a CPU of a computing device where the device is located and running the computer program. The present embodiment provides a target detection apparatus based on unitary transformation, including:
the first constructing module 1000 is configured to construct a space-time tensor of a multi-frame infrared image, where the space-time tensor includes a background tensor and a target tensor;
a second constructing module 1002, configured to construct an objective function of the space-time tensor, where the objective function is to constrain a background tensor by using a tensor kernel norm in a unitary transformation domain and to use a joint space-time total variation sum L 1 The norm is obtained by constraining the target tensor;
a solving module 1004, configured to solve the objective function to obtain an objective tensor;
and a reconstruction output module 1006, configured to reconstruct the solved target tensor into a plurality of single-frame target images, and output a detection result of each target image.
In an embodiment of the present invention, the first building module 1000 may be configured to perform step 100 in the above-described method embodiment, the second building module 1002 may be configured to perform step 102 in the above-described method embodiment, the solving module 1004 may be configured to perform step 104 in the above-described method embodiment, and the reconstruction output module 1006 may be configured to perform step 106 in the above-described method embodiment.
In one embodiment of the invention, the objective function is:
Figure BDA0003915784840000101
Figure BDA0003915784840000102
in the formula,
Figure BDA0003915784840000103
for the purpose of the background tensor is,
Figure BDA0003915784840000104
to be the tensor of interest,
Figure BDA0003915784840000105
is a space-time tensor that is,
Figure BDA0003915784840000106
in the case of random noise, the noise level is,
Figure BDA0003915784840000107
is a tensor kernel norm in unitary transformation domain and is equal to
Figure BDA0003915784840000108
The modulo three fiber of (a) is multiplied by the unitary transformation matrix a to obtain the sum of the nuclear norms of all the front slices of the new tensor,
Figure BDA0003915784840000109
is L of tensor 1 The norm of the number of the first-order-of-arrival,
Figure BDA00039157848400001010
is FroThe norm of the benius is given by the number of benius,
Figure BDA00039157848400001011
is a space-time total variation of lambda 1 、λ 2 And λ 3 In order to balance the coefficients of the process,
Figure BDA00039157848400001012
p ranges from 1 to 10, lambda 2 The value range of (a) is 0.01-0.1, lambda 3 The value range of (1) is 100-200, m is the maximum value of the length and width of each frame of infrared image, and l is the total frame number of the infrared image.
In an embodiment of the present invention, the solving module 1004 is configured to perform:
constructing a unitary transformation matrix by using a zero-frequency component of a space-time tensor time domain;
and solving the target function based on the unitary transformation matrix to obtain a target tensor.
In the embodiment of the present invention, constructing a unitary transformation matrix by using a zero-frequency component of a time domain of a space-time tensor includes:
respectively carrying out one-dimensional Fourier transform on m multiplied by n modular triples of the space-time tensor to obtain a first space-time tensor, wherein m and n are respectively the length and the width of the infrared image;
reserving a zero-frequency component in the first space-time tensor to obtain a second space-time tensor;
respectively performing one-dimensional Fourier inversion on the m multiplied by n modular triples of the second space-time tensor to obtain a third space-time tensor;
and performing singular value decomposition on the modulus three-fiber expansion matrix of the third space-time tensor, and using the conjugate transpose of the obtained left singular matrix as a unitary transformation matrix.
In the embodiment of the present invention, solving an objective function based on a unitary transformation matrix to obtain an objective tensor comprises:
introducing auxiliary variables
Figure BDA0003915784840000111
And
Figure BDA0003915784840000112
the objective function is simplified to a first objective function shown in the following equation:
Figure BDA0003915784840000113
Figure BDA0003915784840000114
in the formula,
Figure BDA0003915784840000115
is a space-time total variation operator;
the augmented Lagrangian function of the first objective function is:
Figure BDA0003915784840000116
in the formula, y 1 ,y 2 ,y＝[y v ,y h ,y t ]Beta represents a penalty factor for lagrange multipliers;
and solving a Lagrange function to obtain a target tensor.
In the embodiment of the present invention, solving the lagrangian function to solve the target tensor comprises:
for the background tensor, in k +1 iterations, let
Figure BDA0003915784840000117
For is to
Figure BDA0003915784840000118
Each modulo three fiber of (A) is multiplied by a unitary transformation matrix (A) to obtain
Figure BDA0003915784840000119
For is to
Figure BDA00039157848400001110
Performing singular value contraction processing on each front slice to obtain an updated front slice
Figure BDA00039157848400001111
Then the
Figure BDA00039157848400001112
The calculation formula of (c):
Figure BDA00039157848400001113
in the formula, A H A conjugate transpose matrix of the unitary transformation matrix a, i, j being natural numbers greater than 0, respectively;
for the target tensor, in k +1 iterations,
Figure BDA00039157848400001114
the calculation formula of (a) is as follows:
Figure BDA0003915784840000121
wherein,
Figure BDA0003915784840000122
sign () is a sign function;
and stopping calculating and outputting the target tensor in response to reaching the preset iteration stop condition.
In the embodiment of the present invention, the preset iteration stop condition is: the number of iterations reaches a preset maximum number of iterations or
Figure BDA0003915784840000123
It is to be understood that the illustrated structure of the embodiment of the present invention does not constitute a specific limitation to an object detection apparatus based on unitary transformation. In other embodiments of the present invention, an apparatus for unitary transform based object detection may include more or fewer components than shown, or some components may be combined, some components may be split, or a different arrangement of components. The illustrated components may be implemented in hardware, software, or a combination of software and hardware.
Because the content of information interaction, execution process, and the like among the modules in the device is based on the same concept as the method embodiment of the present invention, specific content can be referred to the description in the method embodiment of the present invention, and is not described herein again.
The embodiment of the invention also provides a computing device, which comprises a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to realize the target detection method based on unitary transformation in any embodiment of the invention.
Embodiments of the present invention further provide a computer-readable storage medium, on which a computer program is stored, where the computer program, when executed by a processor, causes the processor to execute a unitary transform-based target detection method in any embodiment of the present invention.
Specifically, a system or an apparatus equipped with a storage medium on which software program codes that realize the functions of any of the embodiments described above are stored may be provided, and a computer (or a CPU or MPU) of the system or the apparatus is caused to read out and execute the program codes stored in the storage medium.
In this case, the program code itself read from the storage medium can realize the functions of any of the above-described embodiments, and thus the program code and the storage medium storing the program code constitute a part of the present invention.
Examples of the storage medium for supplying the program code include a floppy disk, a hard disk, a magneto-optical disk, an optical disk (e.g., CD-ROM, CD-R, CD-RW, DVD-ROM, DVD-RAM, DVD-RW, DVD + RW), a magnetic tape, a nonvolatile memory card, and a ROM. Alternatively, the program code may be downloaded from a server computer via a communications network.
Further, it should be clear that the functions of any one of the above-described embodiments may be implemented not only by executing the program code read out by the computer, but also by causing an operating system or the like operating on the computer to perform a part or all of the actual operations based on instructions of the program code.
Further, it is to be understood that the program code read out from the storage medium is written to a memory provided in an expansion board inserted into the computer or to a memory provided in an expansion module connected to the computer, and then causes a CPU or the like mounted on the expansion board or the expansion module to perform part or all of the actual operations based on instructions of the program code, thereby realizing the functions of any of the above-described embodiments.
It is noted that, herein, relational terms such as first and second, and the like may be used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Also, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one of 8230" does not exclude the presence of additional like elements in a process, method, article, or apparatus comprising the element.
Those of ordinary skill in the art will understand that: all or part of the steps for realizing the method embodiments can be completed by hardware related to program instructions, the program can be stored in a computer readable storage medium, and the program executes the steps comprising the method embodiments when executed; and the aforementioned storage medium includes: ROM, RAM, magnetic or optical disks, etc. that can store program codes.
Finally, it should be noted that: the above examples are only intended to illustrate the technical solution of the present invention, but not to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, it should be understood by those of ordinary skill in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some technical features may be equivalently replaced; and such modifications or substitutions do not depart from the spirit and scope of the corresponding technical solutions of the embodiments of the present invention.

Claims (10)

1. A target detection method based on unitary transformation is characterized by comprising the following steps:
constructing a space-time tensor of the multi-frame infrared image, wherein the space-time tensor comprises a background tensor and a target tensor;
constructing an objective function of the space-time tensor, wherein the objective function is to utilize a tensor nuclear norm under a unitary transformation domain to constrain the background tensor and utilize a joint space-time total variation sum L 1 The norm is obtained by constraining the target tensor;
solving the target function to obtain the target tensor;
and reconstructing the solved target tensor into a plurality of single-frame target images, and outputting the detection result of each target image.
2. The method of claim 1, wherein the objective function is:
Figure FDA0003915784830000011
Figure FDA0003915784830000012
in the formula,
Figure FDA0003915784830000013
for the purpose of the said background tensor,
Figure FDA0003915784830000014
for the purpose of the said target tensor,
Figure FDA0003915784830000015
is the space-time tensor in question,
Figure FDA0003915784830000016
in the case of random noise, the noise level is,
Figure FDA0003915784830000017
is a tensor kernel norm in unitary transform domain, numerically equal to
Figure FDA0003915784830000018
The modulo three fiber of (a) is multiplied by the unitary transformation matrix a to obtain the sum of the nuclear norms of all the frontal slices of the new tensor,
Figure FDA0003915784830000019
is L of tensor 1 The number of the norm is calculated,
Figure FDA00039157848300000110
is the Frobenius norm,
Figure FDA00039157848300000111
is a space-time total variation of lambda 1 、λ 2 And λ 3 In order to balance the coefficients of the process,
Figure FDA00039157848300000112
p ranges from 1 to 10, lambda 2 The value range of (A) is 0.01-0.1, lambda 3 The value range of (1) is 100-200, m is the maximum value of the length and width of each frame of the infrared image, and l is the total frame number of the infrared image.
3. The method of claim 2, wherein solving the objective function to solve the target tensor comprises:
constructing the unitary transformation matrix by using a zero-frequency component of the space-time tensor time domain;
and solving the objective function based on the unitary transformation matrix to obtain the target tensor.
4. The method of claim 3, wherein said constructing the unitary transformation matrix using the zero frequency component of the space-time tensor time domain comprises:
respectively performing one-dimensional Fourier transform on m multiplied by n modular triples of the space-time tensor to obtain a first space-time tensor, wherein m and n are the length and the width of the infrared image respectively;
reserving a zero-frequency component in the first space-time tensor to obtain a second space-time tensor;
respectively performing one-dimensional Fourier inversion on the m multiplied by n modular triples of the second space-time tensor to obtain a third space-time tensor;
and performing singular value decomposition on the modulus three-fiber expansion matrix of the third space-time tensor, and using the conjugate transpose of the obtained left singular matrix as the unitary transformation matrix.
5. The method of claim 3, wherein solving the target function based on the unitary transformation matrix to solve the target tensor comprises:
introducing auxiliary variables
Figure FDA0003915784830000021
And
Figure FDA0003915784830000022
reducing the objective function to a first objective function represented by:
Figure FDA0003915784830000023
Figure FDA0003915784830000024
in the formula,
Figure FDA0003915784830000025
is a space-time total variation operator;
the augmented Lagrangian function of the first objective function is:
Figure FDA0003915784830000026
in the formula, y 1 ,y 2 ,y＝[y v ,y h ,y t ]Beta represents a penalty factor for lagrange multipliers;
and solving the Lagrangian function to obtain the target tensor.
6. The method as claimed in claim 5 wherein said solving said Lagrangian function to solve said target tensor comprises:
for the background tensor, in k +1 iterations, let
Figure FDA0003915784830000027
To pair
Figure FDA0003915784830000028
Is multiplied by the unitary transformation matrix A to obtain each module three fibers
Figure FDA0003915784830000029
To pair
Figure FDA00039157848300000210
Performing singular value contraction processing on each front slice to obtain an updated front slice
Figure FDA00039157848300000211
Then the
Figure FDA00039157848300000212
The calculation formula of (2) is as follows:
Figure FDA00039157848300000213
in the formula, A H A conjugate transpose matrix of the unitary transformation matrix a, i, j being natural numbers greater than 0, respectively;
for the target tensor, in k +1 iterations,
Figure FDA00039157848300000214
the calculation formula of (c) is as follows:
Figure FDA00039157848300000215
wherein,
Figure FDA00039157848300000216
sign () is a sign function;
in response to reaching a preset iteration stop condition, stopping computing and outputting the target tensor.
7. The method according to claim 6, wherein the preset iteration stop condition is: the number of iterations reaches a preset maximum number of iterations or
Figure FDA0003915784830000031
8. An object detection apparatus based on unitary transformation, comprising:
the device comprises a first construction module, a second construction module and a third construction module, wherein the first construction module is used for constructing a space-time tensor of a multi-frame infrared image, and the space-time tensor comprises a background tensor and a target tensor;
a second building block for building theAn objective function of the space-time tensor, wherein the objective function is to utilize a tensor kernel norm under a unitary transformation domain to constrain the background tensor and utilize a joint space-time total variation sum L 1 The norm is obtained by constraining the target tensor;
the solving module is used for solving the objective function to obtain the target tensor;
and the reconstruction output module is used for reconstructing the solved target tensor into a plurality of single-frame target images and outputting the detection result of each target image.
9. A computing device comprising a memory having stored therein a computer program and a processor that, when executing the computer program, implements the method of any of claims 1-7.
10. A computer-readable storage medium, on which a computer program is stored which, when executed in a computer, causes the computer to carry out the method of any one of claims 1-7.
CN202211339176.0A 2022-10-28 2022-10-28 Target detection method and device based on unitary transformation, electronic equipment and storage medium Pending CN115511006A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202211339176.0A CN115511006A (en) 2022-10-28 2022-10-28 Target detection method and device based on unitary transformation, electronic equipment and storage medium

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202211339176.0A CN115511006A (en) 2022-10-28 2022-10-28 Target detection method and device based on unitary transformation, electronic equipment and storage medium

Publications (1)

Publication Number Publication Date
CN115511006A true CN115511006A (en) 2022-12-23

Family

ID=84512876

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202211339176.0A Pending CN115511006A (en) 2022-10-28 2022-10-28 Target detection method and device based on unitary transformation, electronic equipment and storage medium

Country Status (1)

Country Link
CN (1) CN115511006A (en)

Similar Documents

Publication Publication Date Title
Gao et al. Using low-rank representation of abundance maps and nonnegative tensor factorization for hyperspectral nonlinear unmixing
Erichson et al. Randomized low-rank dynamic mode decomposition for motion detection
Guo et al. NERNet: Noise estimation and removal network for image denoising
Di Martino et al. Scattering-based SARBM3D
Wang et al. The iterative reweighted alternating direction method of multipliers for separating structural layovers in SAR tomography
Tan et al. Pixelwise Estimation of Signal‐Dependent Image Noise Using Deep Residual Learning
Qian et al. DTAE: Deep tensor autoencoder for 3-D seismic data interpolation
Wu et al. Superpixel-based image noise variance estimation with local statistical assessment
CN116485834A (en) Method, device, equipment and medium for detecting infrared dim target
Bianco et al. Deep learning approach for identification of h ii regions during reionization in 21-cm observations–II. Foreground contamination
Wang et al. Denoising with weak signal preservation by group-sparsity transform learning
US20160327671A1 (en) Processing hexagonally sampled seismic data
Liu et al. SACTNet: Spatial attention context transformation network for cloud removal
Jain et al. Iterative solvers for image denoising with diffusion models: A comparative study
Wang et al. A novel SAR image despeckling method based on local filter with nonlocal preprocessing
CN115511006A (en) Target detection method and device based on unitary transformation, electronic equipment and storage medium
Sadiq et al. Recovering defective Landsat 7 Enhanced Thematic Mapper Plus images via multiple linear regression model
Ozturk et al. Application of compressive sensing to refractivity retrieval using networked weather radars
Zhang et al. A fast algorithm for solving linear inverse problems with uniform noise removal
Di Simone et al. Sensitivity analysis of a scattering-based nonlocal means Despeckling Algorithm
Li et al. Simple framework for the contrastive learning of visual representations-based data-driven tight frame for seismic denoising and interpolation
Rogass et al. Performance of correlation approaches for the evaluation of spatial distortion reductions
Shao et al. Structural similarity-optimal total variation algorithm for image denoising
Grillenzoni Sequential mean shift algorithms for space–time point data
Wang et al. A geostatistical approach to the change-of-support problem and variable-support data fusion in spatial analysis

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