CN116893429B - Single-pixel imaging method and target identification method based on circular harmonic Fourier light field - Google Patents

Single-pixel imaging method and target identification method based on circular harmonic Fourier light field Download PDF

Info

Publication number
CN116893429B
CN116893429B CN202310860591.9A CN202310860591A CN116893429B CN 116893429 B CN116893429 B CN 116893429B CN 202310860591 A CN202310860591 A CN 202310860591A CN 116893429 B CN116893429 B CN 116893429B
Authority
CN
China
Prior art keywords
fourier
target object
distribution matrix
intensity
echo
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN202310860591.9A
Other languages
Chinese (zh)
Other versions
CN116893429A (en
Inventor
韩凯
王彦
孟琪
来文昌
雷国忠
崔文达
陈俊侣
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.)
National University of Defense Technology
Original Assignee
National University of Defense 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 National University of Defense Technology filed Critical National University of Defense Technology
Priority to CN202310860591.9A priority Critical patent/CN116893429B/en
Publication of CN116893429A publication Critical patent/CN116893429A/en
Application granted granted Critical
Publication of CN116893429B publication Critical patent/CN116893429B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • 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
    • G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
    • G01S17/00—Systems using the reflection or reradiation of electromagnetic waves other than radio waves, e.g. lidar systems
    • G01S17/88—Lidar systems specially adapted for specific applications
    • G01S17/89—Lidar systems specially adapted for specific applications for mapping or imaging
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
    • G01S7/00—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00
    • G01S7/48—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00 of systems according to group G01S17/00
    • G01S7/4802—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00 of systems according to group G01S17/00 using analysis of echo signal for target characterisation; Target signature; Target cross-section
    • 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
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/40—Extraction of image or video features

Abstract

The invention provides a single-pixel imaging method and a target identification method based on a circular harmonic Fourier light field, wherein the single-pixel imaging method utilizes circular harmonic Fourier image moment to construct an orthogonal Fourier illumination light field, adopts a pre-measured intensity density distortion factor to optimize rotation invariance characteristics, realizes image reconstruction under extreme undersampling condition, solves the real-time problem of a single-pixel imaging system in the prior art, can directly classify a target object by extracting invariance characteristics of the target object before image reconstruction, greatly shortens the time of target identification by imaging in the prior art, further improves the real-time of the target identification technology based on the single-pixel system, and expands the commercial application value of the single-pixel technology.

Description

Single-pixel imaging method and target identification method based on circular harmonic Fourier light field
Technical Field
The invention relates to the technical field of laser detection imaging, in particular to a single-pixel imaging method and a target identification method based on a circular harmonic Fourier light field.
Background
Single pixel imaging technology has the ability to acquire high resolution images using one single pixel detector, and has been applied in a variety of fields such as nuclear magnetic resonance imaging, space remote sensing, terahertz imaging, and hyperspectral imaging.
The concept of single pixel imaging originated from quantum correlation imaging (or so-called quantum ghost imaging), was first proposed and experimentally verified by Pittman et al in 1995. Single pixel imaging techniques are mainly based on associative measurement principles, imaging a target object by collecting intensity information of light. The single-pixel imaging adopts structured light to actively illuminate a target object, and a single-pixel detector is adopted at a detection end to collect echo signals. When the structure of the illumination light is changed, the corresponding echo signal is changed, and the intensity change content of the echo signal comprises the correlation information between the structure of the illumination light and the space information of the target object. Single pixel imaging is the imaging of objects by constantly changing the illumination light structure and accumulating the associated information.
In terms of illumination light field modes and image reconstruction algorithms, single-pixel imaging technology and associated imaging technology can mutually reference and share intercommunication. The difference is that single pixel imaging techniques use single pixel detectors without spatial resolution capability to acquire intensity image information of a target object. The single-pixel detector has a large photosensitive area and extremely high sensitivity due to the recording of single-point light intensity. Moreover, the single-pixel detector has a very wide response wave band, can realize imaging from near ultraviolet to far infrared even terahertz wave bands, and has a response bandwidth of tens of GHz at maximum. Therefore, shan Xiangsu imaging techniques have advantages in the field of dim target detection.
The Shan Xiangsu imaging technology has many disadvantages in commercial applications. For example, since a large amount of illumination light structures are required for illuminating the target object in the conventional single-pixel imaging, so as to ensure that the different echo signals have enough associated information to reconstruct the image of the target object, the conventional single-pixel imaging technology is very time-consuming, and therefore the real-time requirement of the imaging cannot be met in practical application. The object recognition technology based on single-pixel reconstruction image for feature extraction has serious time delay, so that recognition is difficult or tracking is lost, and conventional object recognition tasks cannot be completed.
In addition, due to the influence of factors such as light source stability and environmental noise, the illumination light field and the echo signal have random fluctuation of intensity. The prior single-pixel imaging technology can not eliminate the random fluctuation of the intensity, and limits the improvement of the quality of the reconstructed image of the target object to a certain extent.
Disclosure of Invention
Based on the above, the invention aims to provide a single-pixel imaging method and a target identification method based on a circular harmonic Fourier light field, wherein the single-pixel imaging method utilizes the circular harmonic Fourier moment to construct a Fourier illumination light field with orthogonality, adopts a pre-measured intensity density distortion factor to optimize a rotation invariance factor, realizes image reconstruction under an extreme undersampling condition, and solves the real-time problem of a single-pixel imaging system in the prior art.
In order to achieve the above object, the present invention provides a single-pixel imaging method based on a circular harmonic fourier light field for imaging a target object using a single-pixel imaging system including a light source for generating a laser beam, a spatial light modulator, and a single-pixel detector, the method comprising: s1, generating a two-dimensional intensity distribution matrix according to a circular harmonic Fourier image moment, loading the two-dimensional intensity distribution matrix into the spatial light modulator, and performing spatial intensity modulation on the laser beam to generate a circular harmonic Fourier beam; s2, the circular harmonic Fourier beam irradiates the target object, and the single-pixel detector receives an echo signal of the target object to obtain Fourier echo intensity; s3, reconstructing an image of the target object according to the two-dimensional intensity distribution matrix and the Fourier echo intensity to obtain a reconstructed image:
wherein n is the order of the moment of the circular harmonic Fourier image, and m is the repetition degree of the circumferential pixels; the two-dimensional intensity distribution matrix comprises a first distribution matrixAnd a second distribution matrix->The fourier echo intensities comprise a first fourier echo intensity +.>And second Fourier echo intensity->The method comprises the steps of carrying out a first treatment on the surface of the Said first distribution matrix->And the second distribution matrixThe two-dimensional intensity distribution matrix generated by the real part and the imaginary part of the circular harmonic fourier image moment, respectively; the first Fourier echo intensity +.>Is the first distribution matrix->Corresponding fourier echo intensities; the second Fourier transformWave intensity->Is the second distribution matrix->Corresponding fourier echo intensities.
Preferably, the single pixel imaging method further comprises: loading a geometric moment into the spatial light modulator before the target object is irradiated by using the circular harmonic Fourier beam, modulating the laser beam to irradiate the target object, and calculating the geometric center of the surface of the target object; when the target object is irradiated with the circular harmonic fourier beam, a transverse mode center of the circular harmonic fourier beam is made to coincide with a surface geometric center of the target object.
Specifically, when the matrix size of the two-dimensional intensity distribution matrix is u×u, the first distribution matrixAnd said second distribution matrix->Expressed as:
wherein the radial function
Where l is the number of pixels on the circumference and k is the circumference sampling interval.
Preferably, in said step S3Before, the single-pixel imaging method further includes: loading a low-order Fourier Mei Linju into the spatial light modulator, modulating the laser beam to irradiate the target object, and receiving an echo signal of the target object by the single-pixel detector to obtain low-order Meilin echo intensity; calculating an intensity density distortion factor g of the target object according to the low-order mellin echo intensity nm
Wherein the low-order mellin echo intensity comprises zero-order mellin echo intensityAnd first order mellin echo intensityThe method comprises the steps of carrying out a first treatment on the surface of the The zero-order mellin echo intensity +.>Is immediately following the first distribution matrix +.>Or the second distribution matrixBefore or after, the acquired echo intensities are mapped using a zero order Fourier Mei Linju with zero circumferential repetition, the first order Merlin echo intensity +.>Is immediately following the first distribution matrix +.>Or the second distribution matrix->Before or after, using a Fourier Mei Linju with a first order and a zero circumferential repetition to correspondingly acquire the echo intensity; />Is the target set +.>Is the minimum value of (a).
Still further, the single pixel imaging method further includes: distortion factor g according to intensity density of the target object nm The reconstructed image of the target object is modified,
wherein the first Fourier echo intensityIs the first distribution matrix->Corresponding fourier echo intensities; second Fourier echo intensity->Is the second distribution matrix->Corresponding fourier echo intensities.
The invention also provides a single-pixel target identification method based on the circular harmonic Fourier light field, which uses the single-pixel imaging system to classify and identify the target object, and comprises the following steps:
s21, acquiring a two-dimensional intensity distribution matrix and Fourier echo intensity of the target object by using the single-pixel imaging method;
s22, using the above-mentioned monoscopic imageA prime imaging method acquires an intensity density distortion factor g of the target object mn The method comprises the steps of carrying out a first treatment on the surface of the Acquiring a target setCalculating the scale distortion factor q of the target object mn ：
Wherein,is the target set +.>Is the minimum value of (a);
s23, according to the intensity density distortion factor g mn And the scale distortion factor q mn Calculating a rotation invariance factor eta of the target object nm And a scale invariance factor phi nm Extracting invariance characteristics of the target object, and classifying the target object;
s24, when the prior information of the target object is not available, reconstructing an image of the target object according to the two-dimensional intensity distribution matrix and the Fourier echo intensity obtained in the step S21, and identifying the target object according to the reconstructed image of the target object and the invariance characteristics of the target object; when the prior information of the target object exists, identifying the target object according to the prior information of the target object and the invariance characteristic of the target physical; wherein the invariance features include translational invariance features, rotational invariance features, and scale invariance features.
Specifically, the rotation invariance factor η nm Expressed as:
wherein,is the first distribution matrix->Corresponding fourier echo intensities; />Is the second distribution matrix->Corresponding fourier echo intensities; i nm Is the equivalent Fourier echo intensity with the circumference repetition degree of m and the order of n.
Further, the scale invariance factor phi is calculated nm The method of (1) comprises: using the scale distortion factor q mn Modifying the first distribution matrixAnd a second distribution matrix->Modified first modification distribution matrix +.>And a second correction distribution matrix->Expressed as:
matrix the first correction distributionAnd said second modified distribution matrix->Loading the space light modulator, modulating the laser beam to irradiate the target object, and obtaining modified Fourier echo intensity; calculating the scale invariance factor phi nm ：
Wherein,is the first modified distribution matrix->Corresponding modified fourier echo intensities,/>Is said second modified distribution matrix->Corresponding modified Fourier echo intensity, phi nm Is the equivalent modified fourier echo intensity with a circle repetition degree of m order n.
The beneficial effects achieved by the technical scheme are as follows:
1) The method has the advantages that the concept of Fourier image moment is applied to the structured light illumination of a single-pixel imaging technology, a target object is illuminated through a circular harmonic Fourier orthogonal light field, the rotation invariance of the target object is extracted through a low-order Fourier melin light field, the random fluctuation of the intensity of the illumination light field and the echo light field is resisted, the influence of environmental noise is eliminated, the quality of single-pixel imaging is improved, the sampling rate of the traditional single-pixel imaging technology is reduced, the real-time imaging of a long-distance dark and weak target is possible, and the commercial application of the single-pixel imaging technology under the extreme undersampling condition is promoted.
2) According to the technical scheme, before the target image is reconstructed, the translation, rotation and scale invariance characteristics of the target are extracted, so that the time consumption of target identification by utilizing the image reconstruction in a traditional single-pixel system is avoided, the efficiency of target identification is improved, and the accuracy of single-pixel target identification is improved by optimizing the target invariance characteristics.
Drawings
FIG. 1 is a flow chart of a single pixel imaging method according to an embodiment of the invention;
FIG. 2 is a schematic diagram of a single pixel imaging system according to an embodiment of the present invention;
FIG. 3 is a two-dimensional intensity distribution matrix generated by a circular harmonic Fourier image moment in accordance with an embodiment of the present invention;
FIG. 4 is a flow chart of a single-pixel target recognition method based on a circular harmonic Fourier light field according to an embodiment of the invention;
FIG. 5 is a graph of the target classification effect based on rotational invariance factors and scale invariance factors according to an embodiment of the present invention;
FIG. 6 is a comparison of a reconstructed image of a target object with an original image under undersampling conditions in accordance with an embodiment of the present invention;
reference numerals illustrate:
1. a light source; 2. a spatial light modulator; 3. a target object; 4. a single pixel detector.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the following detailed description of the present invention will be made with reference to examples. It should be understood that the examples described herein are for illustrative purposes only and are not intended to limit the scope of the present invention.
Referring to fig. 1, which is a schematic flow chart of a single-pixel imaging method according to an embodiment of the present invention, it can be seen that the embodiment of the present invention provides a single-pixel imaging method based on a circular harmonic fourier light field, and the method uses the single-pixel imaging system in fig. 2 to image a target object 3, and specifically includes the following steps:
s1, generating a two-dimensional intensity distribution matrix according to the circular harmonic Fourier image moment, loading the two-dimensional intensity distribution matrix into a spatial light modulator 2, and performing spatial intensity modulation on a laser beam to generate a circular harmonic Fourier beam.
In particular, the circular harmonic fourier image moment is a special form of jacobian-fourier moment, with extremely strong noise immunity and image description capabilities. The circular harmonic fourier image moments are orthogonal within a unit circle, so are also called circular harmonic fourier orthogonal moments, whose kernel function in a polar coordinate system is defined as
P nm ＝T n (r)exp(jmθ) (1)
T n (r) is a radial function, exp (jmθ) is an angular function, n is the order of the circular harmonic Fourier function, i.e., the order of the radial function, T n The mathematical expression of (r) is
Thus, the circular harmonic Fourier image moment in a unit circle in the polar coordinate system is defined as,
from the theory of orthogonal functions, the image moment taking the orthogonal polynomial as the kernel function can reconstruct the original image according to the finite moment set, so that the image reconstruction by adopting the circular harmonic Fourier image moment can be expressed as,
in equation (4), the image needs to be normalized into a unit circle. According to equation (3), the circular harmonic Fourier image moment has a complex form, the real part isAnd imaginary part->Respectively denoted as,
s2, irradiating a target object 3 by using a circular harmonic Fourier beam, and receiving an echo signal of the target object 3 by using a single-pixel detector 4 to obtain Fourier echo intensity;
and S3, carrying out image reconstruction on the target object 3 according to the two-dimensional intensity distribution matrix and the Fourier echo intensity, and obtaining a reconstructed image.
In particular, the process of laser illumination and reconstruction of an image may be equivalent to a projection model, according to the single pixel imaging principle. In the embodiment of the invention, the real part of the intensity distribution matrix of the z-th circular harmonic Fourier image moment is assumed to beThe echo intensity received by the corresponding single-pixel detector 4 is +.>The imaginary part of the intensity distribution matrix of the z-th circular harmonic Fourier image moment isThe echo intensity received by the corresponding single-pixel detector 4 is +.>Then the calculation is performed according to the different circle harmonic fourier image moments, and the reconstructed image of the target object 3 is obtained as follows:
in the above equation, Z is the total number of circular harmonic fourier fields in this embodiment.
In an embodiment of the present invention, the reconstructed image is represented as:
wherein n is the order of the moment of the circular harmonic Fourier image, and m is the repetition degree of the circumferential pixels; the two-dimensional intensity distribution matrix comprises a first distribution matrixAnd a second distribution matrix->The fourier echo intensities comprise a first fourier echo intensity +.>And second Fourier echo intensity->The method comprises the steps of carrying out a first treatment on the surface of the First distribution matrix->And a second distribution matrix->A two-dimensional intensity distribution matrix generated by the real part and the imaginary part of the moment of the circular harmonic Fourier image respectively; first Fourier echo intensity->Is a first distribution matrix->Corresponding fourier echo intensities; second FourierShe Huibo Strength->Is a second distribution matrix->Corresponding fourier echo intensities.
In image processing techniques, all radial image moments have a translational invariance feature as long as a polar coordinate system is built with the centroid of the image. This requires that the illumination structure light field center be aligned with the geometric center of the target object 3 surface in single pixel imaging techniques. The embodiment of the invention is sufficient to illuminate and measure through a 3-frame geometrical moment light field. According to the measurement result, the illumination structure beam orientation device is adjusted to enable the illumination beam to always point to the geometric center of the surface of the target object 3, and translational invariance of the target object 3 can be obtained. Thus, translational invariance of the target object 3 is obtained by stable tracking of the target object 3. Therefore, the single pixel imaging method of the embodiment of the invention further comprises the following steps: before the target object 3 is irradiated by using the circular harmonic Fourier beam, loading a geometric moment into the spatial light modulator 2, modulating the laser beam to irradiate the target object 3, and calculating the geometric center of the surface of the target object 3; when the target object 3 is irradiated with the circular harmonic fourier beam, the center of the transverse mode of the circular harmonic fourier beam is made coincident with the geometric center of the surface of the target object 3.
In the embodiment of the invention, when the matrix size of the two-dimensional intensity distribution matrix is u×u, the specific calculation process of the single-pixel imaging method based on the circular harmonic fourier light field is as follows:
first distribution matrixAnd a second distribution matrix->Expressed as:
where m is the repetition of the circumferential pixels, n is the order of the moment of the circular harmonic fourier image, l is the number of pixels on the circumference, and k is the circumference sampling interval.
Referring to fig. 2, a single-pixel imaging system according to an embodiment of the present invention includes a light source 1 for generating a laser beam, a spatial light modulator 2 and a single-pixel detector 4. The laser beam of the embodiment of the invention propagates to the spatial light modulator 2 through beam expansion, the spatial light modulator 2 generates a structural illumination light field, the structural illumination light field is focused and irradiated to the target object 3 through optical path elements such as lenses, the target object 3 reflects echoes, and finally the single-pixel detector 4 detects the echoes to obtain the echo intensity. In the embodiment of the present invention, the spatial light modulator 2 is a DMD. In fig. 2, the discretization size of the target object 3 is a matrix of u×u pixels, the actual size of the image of the target object 3 is a rectangle of Udx × Udy, where U is a positive integer and dx and dy are the geometric sizes of one image pixel in the x and y directions, respectively. Generating a first distribution matrix of a series of two-dimensional intensity distribution matrices of the circular harmonic Fourier orthogonal light field according to equation (9) and equation (10)And a second distribution matrixThe first distribution matrix->And a second distribution matrix->Sequentially loaded on the spatial light modulator 2, and the laser beams are modulated, so that different illumination structure light fields can be generated to irradiate the target object 3. At the bookIn embodiments, the Shan Xiangsu imaging system employs transmission-type imaging, and in other embodiments, reflective-type imaging may also be employed. In the embodiment of the present invention, the laser beam is modulated by the spatial light modulator 2 and then irradiates the target object 3, and in other embodiments, the laser beam may irradiate the object first and then be modulated by the spatial light modulator 2. After the data acquisition of the single-pixel detector 4 is finished, the data are processed and calculated by a computer to generate a reconstructed image, so that single-pixel imaging is also called calculation imaging.
Referring to fig. 3, a two-dimensional intensity distribution matrix generated by a circular harmonic fourier image moment according to an embodiment of the present invention is shown. In fig. 3, the first row of images corresponds to a first distribution matrix diagram of different circumferential repetition degrees m and orders n, and the second row of images corresponds to a second distribution matrix diagram.
According to equation (3), the angle θ of the circular harmonic fourier image moment is in the phase term, indicating that the amplitude of the image moment is independent of angle. Thus, a circular harmonic fourier image moment of arbitrary order and circumferential repetition is employed, whose corresponding fourier echoes have rotational invariance. Its rotation invariance factor eta nm It can be expressed as that,
wherein,is a first distribution matrix->Corresponding fourier echo intensities; />Is a second distribution matrixCorresponding fourier echo intensities.
It should be noted that equation (11) is only an ideal calculation. The intensity of the illumination structure light may be randomly fluctuated due to the stability of the light source 1 and the environmental noise, and the intensity random noise is generated, thereby generating detection errors. Meanwhile, as the resolution of the speckle in the structural light field is fixed, the discrete error also causes the received echo light intensity to change along with the change of the scale or angle of the target object 3, and intensity noise is generated. The intensity noise is eliminated by the reconstruction of the image in single pixel imaging technology.
Specifically, in the embodiment of the present invention, before the image reconstruction step, the intensity density distortion factor of the target object 3 may be extracted in advance, which specifically includes the following steps: loading the low-order Fourier Mei Linju into the spatial light modulator 2, modulating the laser beam to irradiate the target object 3, and receiving an echo signal of the target object 3 by the single-pixel detector 4 to obtain the low-order Meilin echo intensity; calculating intensity density distortion factor g of target object 3 according to low-order mellin echo intensity nm
Wherein the low-order mellin echo intensity comprises zero-order mellin echo intensityAnd first order mellin echo intensityZero order mellin echo intensity->Is followed by a first distribution matrix->Or a second distribution matrix->Before or after using zero-order Fourier with zero circumferential repetitionMei Linju corresponds to the acquired echo intensity, first-order Meilin echo intensity>Is followed by a first distribution matrix->Or a second distribution matrix->Before or after, using a Fourier Mei Linju with a first order and a zero circumferential repetition to correspondingly acquire the echo intensity; />Is the target set +.>Is the minimum value of (a).
Distortion factor g according to intensity density of target object 3 nm The reconstructed image of the target object 3 is corrected,
wherein the first Fourier echo intensityIs a first distribution matrix->Corresponding fourier echo intensities; second Fourier echo intensity->Is a second distribution matrix->Corresponding fourier echo intensities.
Example two
Invariance features are inherent features of the target object 3, and do not change along with state change of the target object 3, so that the invariance features are often used as the basis for classifying and identifying the target object 3. According to the embodiment of the invention, the target object 3 is classified and identified by extracting the translational invariance feature, the rotational invariance feature and the scale invariance feature of the target object 3. In addition, since there is often not only translational movement but also rotational movement of the actual target object 3, rotational invariance features are important features for the recognition of the target object 3. In addition, in single-pixel target recognition, the distance of the target object 3 has a great influence on recognition. This is because the resolution of a single pixel system is determined by the speckle resolution of the structured light field spot. Since the actual laser beam has a certain divergence angle, the closer the target distance, the smaller the speckle size, the larger the recognition size of the target object 3 with respect to the speckle size, and vice versa. The motion process of the target object 3 from far to near corresponds to the process of identifying the target with a small to large size, so that the scale invariance feature has important significance for classifying and identifying the truly moving target object 3.
Referring to fig. 4, a flow chart of a single-pixel target recognition method based on a circular harmonic fourier light field according to an embodiment of the invention is shown, and the method specifically includes the following steps:
s21, loading the geometric moment into the spatial light modulator 2, modulating the laser beam to generate a geometric moment beam to irradiate the target object 3, and calculating the geometric center of the surface of the target object 3; generating a two-dimensional intensity distribution matrix according to the circular harmonic Fourier image moment, loading the two-dimensional intensity distribution matrix into a spatial light modulator 2, and performing spatial intensity modulation on a laser beam to generate a circular harmonic Fourier beam; the center of the transverse mode of the circular harmonic Fourier beam is overlapped with the geometric center of the target object 3, the target object 3 is irradiated, the single-pixel detector 4 receives the echo signal of the target object 3, and Fourier echo intensity is obtained;
s22, loading the low-order Fourier Mei Linju into the spatial light modulator 2, modulating the laser beam to irradiate the target object 3, receiving the echo signal of the target object 3 by the single-pixel detector 4, and obtaining the low-order Meilin echo intensity M 00 And M 10 The method comprises the steps of carrying out a first treatment on the surface of the Target set based on low-order mellin echo intensitiesCalculating intensity density distortion factor g of target object 3 nm And scale distortion factor q mn 。
Wherein the intensity density distortion factor g nm The definition and calculation of (1) are the same as those of the first embodiment of the present invention, and are not repeated here;
scale distortion factor q mn The calculation is as follows:
from this, the intensity density distortion factor g mn Representing the target object 3 relative to the target setThe intensity density of the minimum of (a) varies by a multiple. Scale distortion factor q mn Representing the target object 3 relative to the target set->A multiple of the dimensional change of the minimum of (a).
S23, distortion factor g according to intensity density mn And scale distortion factor q mn Calculating a rotation invariance factor eta of the target object 3 nm And a scale invariance factor phi nm Extracting invariance characteristics of the target object 3, and classifying the target object 3;
S25:
when no prior information of the target object 3 exists, carrying out image reconstruction on the target object 3 according to the two-dimensional intensity distribution matrix and the Fourier echo intensity, and identifying the target object 3 according to the reconstructed image of the target object 3 and the invariance characteristics of the target object 3;
when the prior information of the target object 3 exists, identifying the target object 3 according to the prior information of the target object 3 and the invariance characteristics of the target object 3;
wherein the invariance features include translational invariance features, rotational invariance features, and scale invariance features.
In particular, the intensity density distortion factor g based on different angles of the target mn The rotation invariance factor eta of the target object 3 nm Can be expressed as:
wherein,is a first distribution matrix->Corresponding fourier echo intensities; />Is a second distribution matrixCorresponding fourier echo intensities; i nm Is the equivalent Fourier echo intensity with the circumference repetition degree of m and the order of n.
In particular, the scale invariance factor Φ nm The calculation process of (2) is as follows: using the scale distortion factor q mn Correcting the first distribution matrixAnd a second distribution matrix->Modified first modification distribution matrix +.>And a second correction distribution matrixExpressed as:
matrix a first correction distributionAnd a second correction distribution matrix->Loading the laser beam into a spatial light modulator 2, modulating the laser beam to irradiate a target object 3, and obtaining corrected Fourier echo intensity; then the scale invariance factor phi nm Can be expressed as:
wherein,is a first modified distribution matrix->Corresponding fourier echo intensities,/">Is a second modified distribution matrix->Corresponding Fourier echo intensity, phi nm The equivalent Fourier echo intensity corresponding to the corrected two-dimensional intensity distribution matrix.
According to the formulas (13) to (18), not only the rotational invariance is extractedFactor eta nm And a scale invariance factor phi nm Compared with the prior art, the method can directly extract the characteristics of the target object 3 without reconstructing an image, so that the real-time performance of target identification is greatly enhanced.
In actual operation, the DMD is modulated by a light field intensity distribution matrix, the real-time monitoring requirement on the state of the target object 3 is considered in the arrangement sequence of the matrix sequence, and the low-order fourier Mei Linju is inserted into the two-dimensional intensity distribution matrix sequence of the circular harmonic fourier moment, so that the intensity density distortion factor of each circular harmonic fourier beam when illuminating the target object 3 is conveniently calculated. In the embodiment of the present invention, the arrangement order in which the spatial light modulator 2 loads the modulation matrix may be expressed as,
wherein L is 00 A zero order fourier Mei Linju representing zero circumferential repetition; l (L) 10 A fourier Mei Linju of order one, representing zero circumferential repetition.
Referring to FIG. 5, FIG. 5 is a graph showing the rotational invariance factor η based on an embodiment of the present invention nm And a scale invariance factor phi nm Wherein the target is a A, B, C, D four letter image. FIG. 5 (a) shows a graph based on a rotation invariance factor η nm Is a target classification effect of (1). FIG. 5 (b) shows a scale invariance factor Φ nm Is a target classification effect of (1). In the target classification based on the rotation invariance factor, the four letters are rotated by 15 °, 30 °, 45 °, 60 °, 75 °, 90 °, 105 °, 120 °, 135 °, 150 °, 175 °, respectively, and the intensity density change factors are respectively: 1. 1.0106, 0.9943,1.0055,1.0123. This means that the intensity density will also be distorted when the target is rotated. This may be caused by instability of the illumination intensity and speckle dispersion errors. The rotation invariance factors of four letters under different angles can be calculated according to the received light intensity of the single pixel detector 4 and the intensity density distortion factors. Clustering of FIG. 5 (a)The graph has the amplitudes of the circular harmonic fourier orthogonal light fields D32 and D22 as the ordinate and abscissa, respectively. As can be seen from fig. 5 (a), images with the same letter (or class) and different rotation angles can be obviously classified into one class. In classification based on multiple distortion invariance factors, the four letters are respectively reduced by 1 time, 0.81 time, 0.61 time, 0.41 time and 0.21 time, and corresponding scale change factors are calculated as follows: 1. 0.8177, 0.621,0.4,179,0.2264, the corresponding density change factors are calculated as: 1. 1.0021, 1.0102,1.0199,1.0342. This illustrates that the intensity density is distorted as the target undergoes dimensional changes. In fig. 5 (b), the cluster map has the amplitudes of the circular harmonic fourier image moments D22 and D11 as the ordinate and abscissa, respectively. As can be seen from fig. 5 (b), images of the same letter and different scale can be clearly classified into one type. In addition, compared with fig. 5 (a) and 5 (b), the accuracy of classification according to the scale invariance factor is better than the accuracy of classification based on rotation invariance.
Referring to fig. 6, fig. 6 is an image of a target object reconstructed under undersampling conditions in accordance with an embodiment of the present invention. In fig. 6, the resolution of the image is 128×128, and the number of spots corresponding to the sampling rates of 0.012 and 0.024 are 200 and 400 frames, respectively. The Hadamard code (Hadamard) of fig. 6 uses a walsh-Hadamard light field based on Z-sampling order, which is the best-performing Hadamard single-pixel imaging method for undersampling imaging at present. According to fig. 6, under the undersampling condition, the circular harmonic fourier light field still has strong target image reconstruction capability, and the reconstructed image is clear and distinguishable. The Walsh-Hard code light field has very low imaging resolution under the condition of extreme undersampling, and the object can not be identified almost. The results of the structural similarity (ssim) evaluation of the two light field reconstructed images are also given in fig. 6. To ensure the objectivity of the evaluation, ssim evaluation of both light field reconstructed images was performed in the imaging circle domain. According to fig. 6, the ssim evaluation result of the circular harmonic fourier light field imaging is higher than that of the Z-order walsh-hadamard code light field imaging, which is consistent with the subjective judgment result of human eyes. In practice, 0.012 and 0.024 are extremely undersampled in single pixel imaging, and existing single pixel imaging light field modulation and detection devices can already do real-time computational imaging at this sampling rate.
Finally, it should be noted that the above-mentioned embodiments are only for illustrating the technical solution of the present invention, and not for limiting the same, and although the present invention has been described in detail with reference to the preferred embodiments, it should be understood by those skilled in the art that modifications and equivalents may be made to the technical solution of the present invention without departing from the spirit and scope of the technical solution of the present invention, and all such modifications and equivalents are intended to be encompassed in the scope of the claims of the present invention.

Claims (8)

1. A single pixel imaging method based on a circular harmonic fourier light field, the method imaging a target object using a single pixel imaging system comprising a light source for generating a laser beam, a spatial light modulator and a single pixel detector, the method comprising:
s1, generating a two-dimensional intensity distribution matrix according to a circular harmonic Fourier image moment, loading the two-dimensional intensity distribution matrix into the spatial light modulator, and performing spatial intensity modulation on the laser beam to generate a circular harmonic Fourier beam;
s2, the circular harmonic Fourier beam irradiates the target object, and the single-pixel detector receives an echo signal of the target object to obtain Fourier echo intensity;
s3, reconstructing an image of the target object according to the two-dimensional intensity distribution matrix and the Fourier echo intensity to obtain a reconstructed image:
wherein n is the order of the moment of the circular harmonic Fourier image, and m is the repetition degree of the circumferential pixels;
the two-dimensional intensity distribution matrix comprises a first distribution matrixAnd a second distribution matrix->The fourier echo intensities comprise a first fourier echo intensity +.>And second Fourier echo intensity->
The first distribution matrixAnd said second distribution matrix->The two-dimensional intensity distribution matrix generated by the real part and the imaginary part of the circular harmonic fourier image moment, respectively;
the first Fourier echo intensityIs the first distribution matrix->Corresponding fourier echo intensities; the second Fourier echo intensity +.>Is the second distribution matrix->Corresponding fourier echo intensities.
2. The single pixel imaging method of claim 1, further comprising:
loading a geometric moment into the spatial light modulator, modulating the laser beam to irradiate the target object, and calculating the geometric center of the surface of the target object;
when the target object is irradiated with the circular harmonic fourier beam, a transverse mode center of the circular harmonic fourier beam is made to coincide with a surface geometric center of the target object.
3. A single pixel imaging method according to claim 1 or 2, wherein,
when the matrix size of the two-dimensional intensity distribution matrix is U×U, the first distribution matrixAnd said second distribution matrix->Expressed as:
wherein the radial function
Where l is the number of pixels on the circumference and k is the circumference sampling interval.
4. The single pixel imaging method according to claim 1, characterized in that before said step S3, said single pixel imaging method further comprises:
loading a low-order Fourier Mei Linju into the spatial light modulator, modulating the laser beam to irradiate the target object, and receiving an echo signal of the target object by the single-pixel detector to obtain low-order Meilin echo intensity;
calculating an intensity density distortion factor g of the target object according to the low-order mellin echo intensity nm
Wherein the low-order mellin echo intensity comprises zero-order mellin echo intensityAnd first order mellin echo intensityThe zero-order mellin echo intensity +.>Is immediately following the first distribution matrix +.>Or the second distribution matrixBefore or after, the acquired echo intensities are mapped using a zero order Fourier Mei Linju with zero circumferential repetition, the first order Merlin echo intensity +.>Is immediately following the first distribution matrix +.>Or the second distribution matrix->Before or after using a circumference repetition degree ofA fourier Mei Linju with zero order of one corresponds to the acquired echo intensity; />Is the target set +.>Is the minimum value of (a).
5. The single pixel imaging method of claim 4, further comprising:
distortion factor g according to intensity density of the target object nm The reconstructed image of the target object is modified,
wherein the first Fourier echo intensityIs the first distribution matrix->Corresponding fourier echo intensities; second Fourier echo intensity->Is the second distribution matrix->Corresponding fourier echo intensities.
6. A single-pixel target recognition method based on a circular harmonic fourier light field, which classifies and recognizes the target object using the single-pixel imaging system of claim 1, characterized in that the single-pixel target recognition method comprises:
s21, acquiring a two-dimensional intensity distribution matrix and Fourier echo intensities of the target object by using the single-pixel imaging method of claim 3;
s22, obtaining the intensity density distortion factor g of the target object by using the single-pixel imaging method as claimed in claim 4 mn ；
The single pixel imaging method of claim 4, obtaining the target setCalculating a scale distortion factor of the target object:
wherein,is the target set +.>Is the minimum value of (a);
s23, according to the intensity density distortion factor g mn And the scale distortion factor q mn Calculating a rotation invariance factor eta of the target object nm And a scale invariance factor phi nm Extracting invariance characteristics of the target object, and classifying the target object;
s24, when the prior information of the target object is not available, reconstructing an image of the target object according to the two-dimensional intensity distribution matrix and the Fourier echo intensity obtained in the step S21, and identifying the target object according to the reconstructed image of the target object and the invariance characteristics of the target object;
when the prior information of the target object exists, identifying the target object according to the prior information of the target object and the invariance characteristic of the target physical;
wherein the invariance features include translational invariance features, rotational invariance features, and scale invariance features.
7. The method of claim 6, wherein,
the rotation invariance factor eta nm Expressed as:
wherein,is the first distribution matrix->Corresponding fourier echo intensities; />Is the second distribution matrix->Corresponding fourier echo intensities; i nm Is the equivalent Fourier echo intensity with the circumference repetition degree of m and the order of n.
8. The method of claim 6, wherein the calculating scale invariance factor Φ nm The method of (1) comprises:
using the scale distortion factor q mn Modifying the first distribution matrixAnd a second distribution matrix->Modified first modification distribution matrix->And a second correction distribution matrix->Expressed as:
matrix the first correction distributionAnd said second modified distribution matrix->Loading the space light modulator, modulating the laser beam to irradiate the target object, and obtaining modified Fourier echo intensity;
calculating the scale invariance factor phi nm ：
Wherein,is the first modified distribution matrix->Corresponding modified fourier echo intensities,/>Is said second modified distribution matrix->Corresponding modified Fourier echo intensity, phi nm Is the equivalent modified fourier echo intensity for a circle repetition of order m of order n.
CN202310860591.9A 2023-07-13 2023-07-13 Single-pixel imaging method and target identification method based on circular harmonic Fourier light field Active CN116893429B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202310860591.9A CN116893429B (en) 2023-07-13 2023-07-13 Single-pixel imaging method and target identification method based on circular harmonic Fourier light field

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202310860591.9A CN116893429B (en) 2023-07-13 2023-07-13 Single-pixel imaging method and target identification method based on circular harmonic Fourier light field

Publications (2)

Publication Number Publication Date
CN116893429A CN116893429A (en) 2023-10-17
CN116893429B true CN116893429B (en) 2024-01-30

Family

ID=88313165

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202310860591.9A Active CN116893429B (en) 2023-07-13 2023-07-13 Single-pixel imaging method and target identification method based on circular harmonic Fourier light field

Country Status (1)

Country Link
CN (1) CN116893429B (en)

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117471489B (en) * 2023-12-25 2024-03-22 中国人民解放军国防科技大学 Target detection method and device based on single-pixel imaging technology and computer equipment

Citations (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN1530854A (en) * 2003-03-11 2004-09-22 平子良 Use of circular convient-Fourier matrix in pattern recognition
CN1553373A (en) * 2003-06-03 2004-12-08 平子良 Jacobi-Fourier matrice series describing image
CN104504645A (en) * 2014-12-25 2015-04-08 辽宁师范大学 Method for embedding and detecting robust image watermarks on basis of circular-harmonic-Fourier moments
WO2016015516A1 (en) * 2014-07-29 2016-02-04 暨南大学 Optical imaging method using single pixel detector
CN106371201A (en) * 2016-11-03 2017-02-01 清华大学 Fourier overlapping correlation imaging system and method based on computational ghost imaging
CN110118755A (en) * 2019-04-30 2019-08-13 上海理工大学 Optical imaging method based on Fourier's light field and relevance imaging
CN113114882A (en) * 2021-03-26 2021-07-13 暨南大学 Fourier single-pixel imaging method with high sampling efficiency
CN114721006A (en) * 2022-06-08 2022-07-08 中国人民解放军国防科技大学 High-precision single-pixel imaging method and system

Patent Citations (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN1530854A (en) * 2003-03-11 2004-09-22 平子良 Use of circular convient-Fourier matrix in pattern recognition
CN1553373A (en) * 2003-06-03 2004-12-08 平子良 Jacobi-Fourier matrice series describing image
WO2016015516A1 (en) * 2014-07-29 2016-02-04 暨南大学 Optical imaging method using single pixel detector
CN104504645A (en) * 2014-12-25 2015-04-08 辽宁师范大学 Method for embedding and detecting robust image watermarks on basis of circular-harmonic-Fourier moments
CN106371201A (en) * 2016-11-03 2017-02-01 清华大学 Fourier overlapping correlation imaging system and method based on computational ghost imaging
CN110118755A (en) * 2019-04-30 2019-08-13 上海理工大学 Optical imaging method based on Fourier's light field and relevance imaging
CN113114882A (en) * 2021-03-26 2021-07-13 暨南大学 Fourier single-pixel imaging method with high sampling efficiency
CN114721006A (en) * 2022-06-08 2022-07-08 中国人民解放军国防科技大学 High-precision single-pixel imaging method and system

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
Spatial frequency domain imaging technology based on Fourier single-pixel imaging;Hui M. Ren 等;Journal of Biomedical Optics;第27卷(第1期);全文 *
采用圆谐·傅里叶矩的图像区域 复制粘贴篡改检测;秦娟;中国图象图形学报;第18卷(第8期);全文 *

Also Published As

Publication number Publication date
CN116893429A (en) 2023-10-17

Similar Documents

Publication Publication Date Title
Su et al. Deep end-to-end time-of-flight imaging
Patel et al. Sparse representations and compressive sensing for imaging and vision
US10302424B2 (en) Motion contrast depth scanning
CN116893429B (en) Single-pixel imaging method and target identification method based on circular harmonic Fourier light field
CN105589210B (en) A kind of digitlization synthetic aperture imaging method based on pupil modulation
CN111340183B (en) Deep learning-based scatter imaging device and method
CN113114882B (en) Fourier single-pixel imaging method with high sampling efficiency
CN106405546A (en) Quick correlated imaging system and method based on compression speckle
Zhai et al. Foveated ghost imaging based on deep learning
CN114721006B (en) High-precision single-pixel imaging method and system
Yang et al. Tracking compensation in computational ghost imaging of moving objects
US11860034B2 (en) Systems, devices, and methods for hyperspectral imaging
CN113504202B (en) Coherent modulation imaging method based on axial translation binary amplitude mask
Bacon et al. Denoising gravitational-wave signals from binary black holes with a dilated convolutional autoencoder
Sun et al. Optical scanning holography based on compressive sensing using a digital micro-mirror device
Zhang et al. Photon-starved snapshot holography
Wang et al. Local conditional neural fields for versatile and generalizable large-scale reconstructions in computational imaging
Tong et al. Quantitative pure-phase object reconstruction under single-shot Fourier measurement via deep learning
CN111596310B (en) Moving target ghost imaging system and method based on point detection
CN110703276A (en) Fourier imaging device and method under strong scattering condition
CN114859377B (en) Method and equipment for capturing single-pixel imaging of moving target in real time
Liu et al. Spatial anomaly detection in hyperspectral imaging using optical neural networks
Xiang et al. Phase Retrieval for Fourier THz Imaging with Physics-Informed Deep Learning
Zhang et al. U 2 Net-based Single-pixel Imaging Salient Object Detection
White et al. Super-resolving beyond satellite hardware using realistically degraded images

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