Target image classification methods, systems, media, and devices based on complex backgrounds

Abstract

This invention discloses a target image classification method based on complex backgrounds, comprising: acquiring continuous and stable monochromatic p-polarized light; acquiring a target image and loading the target image onto an amplitude modulation array, using the amplitude modulation array after loading the target image to modulate the monochromatic p-polarized light to obtain a spatial-temporal optical signal of the target image containing a complex background; converting the spatial-temporal optical signal of the target image containing a complex background into a spatial frequency domain signal, wherein the spatial frequency domain signal is a spatial frequency distribution map containing spatial information of the target image; removing the background information from the spatial frequency domain signal to obtain a background-removed spatial frequency domain signal; converting the background-removed spatial frequency domain signal into a background-removed spatial-temporal optical signal, wherein the background-removed spatial-temporal optical signal is a spatial intensity distribution map containing spatial information of the target image; and classifying the target image corresponding to the background-removed spatial-temporal optical signal.

Description

Technical Field

[0001] This invention relates to the field of image processing and recognition technology, and in particular to a method, system, medium, and device for classifying target images based on complex backgrounds. Background Technology

[0002] The traditional von Neumann architecture has long dominated the field of computing. However, with continuous technological advancements and increasing demands, its limitations have become increasingly apparent. Due to its in-memory / compute separation design principle, the von Neumann architecture faces the dual challenges of the power wall and the memory wall, which limit the improvement of computing performance and gradually slow down the rate of increase in Moore's Law. To address these challenges, researchers are constantly exploring new computing technologies and architectures in an attempt to break through the limits of traditional microelectronic computing.

[0003] Diffractive Optical Neural Networks (DONNs), as an emerging computing technology, have shown broad application prospects in high-performance computing and artificial intelligence due to their advantages such as ultra-high speed, large bandwidth, and multi-dimensionality. DONNs fully integrate advanced technologies such as high-speed optical communication, optical interconnection, optical integration, and silicon-based optoelectronics, providing a highly competitive solution for breaking through the bottlenecks of traditional microelectronic computing in the post-Moore's Law era.

[0004] However, despite its many advantages, DONN still faces some challenges in practical applications. Currently, DONN is primarily used for target recognition of pre-processed, simple targets. When faced with targets in complex backgrounds, DONN often fails to achieve accurate recognition, significantly reducing its practicality in real-world scenarios. Therefore, improving DONN's target recognition capabilities in complex backgrounds has become an urgent problem to be solved. Summary of the Invention

[0005] Therefore, it is necessary to propose a target image classification method based on complex backgrounds to address the above problems.

[0006] A target image classification method based on complex backgrounds, the method comprising the following steps:

[0007] Obtain continuous and stable monochromatic p-polarized light;

[0008] Acquire a target image and load the target image into an amplitude modulation array. Use the amplitude modulation array after loading the target image to modulate the monochromatic p-polarized light to obtain the spatial-temporal optical signal of the target image containing a complex background.

[0009] The spatial-temporal optical signal of the target image containing a complex background is converted into a spatial-frequency domain signal, and the spatial-frequency domain signal is a spatial frequency distribution map containing spatial information of the target image.

[0010] Remove the background information from the spatial frequency domain signal to obtain the background-removed spatial frequency domain signal;

[0011] The background-removed spatial frequency domain signal is converted into a background-removed spatial temporal optical signal, which is a spatial intensity distribution map containing spatial information of the target image; the target image corresponding to the background-removed spatial temporal optical signal is then classified.

[0012] In the above scheme, obtaining continuous and stable monochromatic p-polarized light specifically includes:

[0013] The laser emits a continuous and stable monochromatic linearly polarized laser beam.

[0014] The monochromatic linearly polarized laser beam is passed through a pair of beam expanders and adjusted to obtain a parallel beam after beam expansion.

[0015] The expanded parallel beam is polarized and modulated according to a half-wave plate;

[0016] By controlling the angle of the half-wave plate, continuous and stable monochromatic p-polarized light can be obtained.

[0017] In the above scheme, the step of acquiring the target image and loading the target image into the amplitude modulation array further includes: polarization filtering of the continuous and stable monochromatic p-polarized light through a polarization beam splitter PBS.

[0018] In the above scheme, the step of using the amplitude modulation array after loading the target image to modulate the monochromatic p-polarized light to obtain the spatial-temporal optical signal of the target image containing a complex background further includes: adjusting the light field intensity of the reflected light according to the gray value distribution of the loaded grayscale image to obtain the spatial-temporal optical signal of the target image containing a complex background.

[0019] In the above scheme, the step of converting the spatial-temporal optical signal of the target image containing a complex background into a spatial-frequency domain signal specifically includes:

[0020] The amplitude modulation array of the reflective amplitude spatial light modulator (SLM) receives monochromatic p-polarized light reflected by the first polarizing beam splitter (PBS), and the monochromatic p-polarized light is reflected as monochromatic s-polarized light after amplitude modulation.

[0021] The monochromatic s-polarized light is reflected in a direction perpendicular to the incident light path by the first polarizing beam splitter PBS.

[0022] The reflected beam is passed from the amplitude modulation array through the first polarization beam splitter PBS, and the focal length f of the lens is determined according to the distance of the beam when it reaches the Fourier lens in its propagation path.

[0023] When the spatial-temporal optical signal of the target image with a complex background passes through a Fourier lens, the Fourier lens converts it to the spatial frequency domain to obtain a spatial frequency domain signal.

[0024] In the above scheme, the step of converting the spatial-temporal optical signal of the target image containing a complex background into a spatial-frequency domain signal specifically includes:

[0025] A two-dimensional Fourier transform is performed on the modulated s-polarized light to convert the spatial information of the target image into frequency information.

[0026] Based on the spatial frequency distribution map formed on the back focal plane of the Fourier lens, the spatial frequency domain signal of the target image containing a complex background is determined by the signal after Fourier transform.

[0027] In the above scheme, the step of filtering and nonlinearly activating the spatial frequency domain signal to obtain the processed spatial frequency domain signal specifically includes:

[0028] The spatial frequency domain signal is filtered according to the first diffraction light neural network (DONN);

[0029] The spatial frequency domain signal is obtained by nonlinearly activating the filtered spatial frequency domain signal using a transmissive phase-type spatial light modulator (SLM).

[0030] This application also proposes a target image classification system based on complex backgrounds, the system comprising: a polarization acquisition unit, a temporal signal acquisition unit, a signal processing unit, and a classification unit;

[0031] The polarization acquisition unit is used to acquire continuous and stable monochromatic p-polarized light;

[0032] The time-domain signal acquisition unit is used to acquire a target image and load the target image into an amplitude modulation array. The amplitude modulation array after loading the target image is used to modulate the monochromatic p-polarized light to acquire the spatial-temporal optical signal of the target image containing a complex background.

[0033] The signal processing unit is configured to convert the spatial-temporal optical signal of the target image containing a complex background into a spatial-frequency domain signal, wherein the spatial-frequency domain signal is a spatial frequency distribution map containing spatial information of the target image; remove the background information of the spatial-frequency domain signal to obtain a background-removed spatial-frequency domain signal; convert the background-removed spatial-frequency domain signal into a background-removed spatial-temporal optical signal, wherein the background-removed spatial-temporal optical signal is a spatial intensity distribution map containing spatial information of the target image; and the classification unit is configured to classify the target image corresponding to the background-removed spatial-temporal optical signal.

[0034] This application also proposes a readable storage medium storing a computer program, which, when executed by a processor, causes the processor to perform the following steps:

[0035] Obtain continuous and stable monochromatic p-polarized light;

[0036] Acquire a target image and load the target image into an amplitude modulation array. Use the amplitude modulation array after loading the target image to modulate the monochromatic p-polarized light to obtain the spatial-temporal optical signal of the target image containing a complex background.

[0037] The spatial-temporal optical signal of the target image containing a complex background is converted into a spatial-frequency domain signal, and the spatial-frequency domain signal is a spatial frequency distribution map containing spatial information of the target image.

[0038] Remove the background information from the spatial frequency domain signal to obtain the background-removed spatial frequency domain signal;

[0039] The background-removed spatial frequency domain signal is converted into a background-removed spatial temporal optical signal, which is a spatial intensity distribution map containing spatial information of the target image; the target image corresponding to the background-removed spatial temporal optical signal is then classified.

[0040] This application also proposes a computer device, including a memory and a processor, wherein the memory stores a computer program, and the computer program is executed by the processor in the following steps:

[0041] Obtain continuous and stable monochromatic p-polarized light;

[0042] Acquire a target image and load the target image into an amplitude modulation array. Use the amplitude modulation array after loading the target image to modulate the monochromatic p-polarized light to obtain the spatial-temporal optical signal of the target image containing a complex background.

[0043] The spatial-temporal optical signal of the target image containing a complex background is converted into a spatial-frequency domain signal, and the spatial-frequency domain signal is a spatial frequency distribution map containing spatial information of the target image.

[0044] Remove the background information from the spatial frequency domain signal to obtain the background-removed spatial frequency domain signal;

[0045] The background-removed spatial frequency domain signal is converted into a background-removed spatial temporal optical signal, which is a spatial intensity distribution map containing spatial information of the target image; the target image corresponding to the background-removed spatial temporal optical signal is then classified.

[0046] The embodiments of this invention have the following beneficial effects: First, continuous and stable monochromatic p-polarized light is acquired; a target image is acquired and loaded onto an amplitude modulation array; the monochromatic p-polarized light is modulated using the amplitude modulation array after loading the target image to obtain a spatial-temporal optical signal of the target image containing a complex background; the spatial-temporal optical signal of the target image containing a complex background is converted into a spatial frequency domain signal, which is a spatial frequency distribution map containing spatial information of the target image; background information is removed from the spatial frequency domain signal to obtain a background-removed spatial frequency domain signal; the background-removed spatial frequency domain signal is converted into a background-removed spatial-temporal optical signal, which is a spatial intensity distribution map containing spatial information of the target image; the target image corresponding to the background-removed spatial-temporal optical signal is classified. This invention, through the processing of optical signals, can extract target images from complex backgrounds and perform accurate classification. Attached Figure Description

[0047] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0048] in:

[0049] Figure 1 This is a schematic diagram of a target image classification method based on complex backgrounds in one embodiment;

[0050] Figure 2 This is a schematic diagram of the structure of a target image classification system based on complex backgrounds in one embodiment.

[0051] Explanation of reference numerals in the attached figures

[0052] 1: Laser; 2: Beam expander lens; 3: HWP; 4: PBS; 5: Reflective amplitude-type SLM; 6: Target acquisition detector; 7: Processing system; 8: Fourier lens; 9: First DONN; 10: Transmissive phase-type SLM; 11: Inverse Fourier lens; 12: Second DONN; 13: Classification and recognition detector. Detailed Implementation

[0053] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0054] In the following description, numerous specific details are set forth in order to provide a more thorough understanding of the invention; however, it will be apparent to those skilled in the art that the invention may be practiced without one or more of these details; in other instances, certain technical features well-known in the art have not been described in order to avoid confusion with the invention. It should be understood that the invention can be practiced in different forms and should not be construed as limited to the embodiments set forth herein; rather, these embodiments are provided to make the disclosure thorough and complete and to fully convey the scope of the invention to those skilled in the art.

[0055] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the invention. When used herein, the singular forms “a,” “an,” and “the” are also intended to include the plural forms, unless the context clearly indicates otherwise. The terms “comprising” and / or “including,” when used in this specification, identify the presence of said features, integers, steps, operations, elements, and / or components, but do not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups. When used herein, the term “and / or” includes any and all combinations of the associated listed items.

[0056] To address this issue, this application proposes a novel method: a 4f system is used to perform a Fourier transform on the real-time acquired optical signal in the spatial domain, converting it into an optical signal in the spatial frequency domain. Then, a DONN (Discrete Array Neural Network) and a Spatial Light Modulator (SLM) are used to filter and perform nonlinear activation operations on the optical signal in the spatial frequency domain. This method effectively filters out complex backgrounds while retaining the target to be identified, thereby improving the target recognition capability of the DONN in complex backgrounds. Finally, another DONN is used to identify the target image after background filtering to achieve accurate target classification. This method not only improves the practicality of the DONN but also provides broader possibilities for its application in daily life.

[0057] To fully understand the present invention, a detailed structure will be presented in the following description in order to illustrate the technical solution proposed by the present invention; optional embodiments of the present invention are described in detail below, however, in addition to these detailed descriptions, the present invention may have other embodiments.

[0058] like Figure 1 As shown, in one embodiment, a target image classification method based on complex backgrounds is provided. This method includes steps S101 to S106, which are detailed below:

[0059] S101. Obtain continuous and stable monochromatic p-polarized light;

[0060] In some embodiments, obtaining continuously stable monochromatic p-polarized light specifically includes:

[0061] The laser emits a continuous and stable monochromatic linearly polarized laser beam.

[0062] A monochromatic linearly polarized laser beam is passed through a pair of beam expanders and adjusted to obtain a parallel beam after beam expansion.

[0063] Polarization modulation of the expanded parallel beam is performed using a half-wave plate;

[0064] By controlling the angle of the half-wave plate, continuous and stable monochromatic p-polarized light can be obtained.

[0065] Preferably, the laser emits a continuous and stable monochromatic linearly polarized laser beam, which is then expanded in diameter and adjusted to a parallel beam by a beam expander lens. At this point, the incident light wave is a plane wave, and its complex amplitude U(r) is expressed as:

[0066] U(r)=u0e ikz

[0067] Where u0 is the real part of the complex amplitude, which determines the magnitude of the light intensity and the phase function. This indicates that the phase changes only with the z-direction and is independent of the x and y directions. The wave vector k represents the direction of wave propagation and is expressed as k = 2π / λ.

[0068] Since SLM requires p-polarized light (the component whose electric vector vibrates parallel to the incident plane is called the p component; if linearly polarized light only has the p component, it is called p-polarized light) to be incident normally, the beam after beam expansion needs to be polarized and modulated by HWP to obtain continuous and stable monochromatic p-polarized light.

[0069] The polarization modulation process involves rotating a half-wave plate so that its fast axis makes an angle θ with the original polarization direction. Since the half-wave plate adds a phase delay π to the slow axis component, the linearly polarized light remains linearly polarized after passing through the half-wave plate, and the orientation of the vibration plane has rotated by an angle of 2θ compared to the incident light. Therefore, the polarization direction of the linearly polarized light can be adjusted to make it p-polarized light simply by changing the size of the angle θ.

[0070] S102. Acquire the target image and load the target image into the amplitude modulation array. Use the amplitude modulation array after loading the target image to modulate the monochromatic p-polarized light to obtain the spatial-temporal optical signal of the target image containing a complex background.

[0071] In some embodiments, acquiring a target image and loading the target image onto an amplitude modulation array further includes: polarization filtering of continuous and stable monochromatic p-polarized light using a polarization beam splitter PBS.

[0072] Specifically, the PBS performs polarization filtering on the HWP-modulated outgoing beam, which can transmit p-polarized light and reflect s-polarized light, so that only p-polarized light is incident on the amplitude modulation array of the reflective amplitude type SLM. At the same time, according to the actual use requirements, the processing system loads the grayscale image of the target with a complex background, which is acquired in real time by the target acquisition detector, onto the amplitude modulation array of the reflective amplitude type SLM, so that the light field intensity of the reflected light follows the grayscale value distribution of the loaded grayscale image.

[0073] The working principle of PBS is to select a suitable adhesive (n e <n<n o Two negative uniaxial crystals (n e <n o A right-angle prism made of s is glued together along the inclined plane. The optical axes of the two s crystals are parallel to the right-angle plane through which light passes and the incident plane. At this time, the e-light is p-polarized and the o-light is s-polarized (the component of the electric vector that vibrates perpendicular to the incident plane is called the s-component. If linearly polarized light has only the s-component, it is called s-polarized light).

[0074] The working principle of an amplitude-type SLM is as follows: When no voltage is applied to the electrodes, the liquid crystal arrangement gradually twists by 90° along the direction of the light incident plane. Natural light passes through polarizer p1 (parallel to the incident plane) to obtain polarized light, i.e., p-polarized light, with its vibration direction parallel to the incident plane. After the beam enters the light valve, it is reflected by a dielectric mirror and then exits through the light valve. The polarization direction of the outgoing light does not change, so no light passes through the orthogonal analyzer p2 (perpendicular to the incident plane). When the electrodes of the liquid crystal light valve are energized, the liquid crystal arrangement twists by 45°. The vibration direction of the outgoing light reflected by the dielectric mirror is perpendicular to the incident plane, and at this time, all the outgoing light can pass through the analyzer p2.

[0075] Based on actual usage requirements, the grayscale image of the target acquired by the target acquisition detector is loaded onto the amplitude modulation array via the processing module. The amplitude modulation array changes the resistance of the photosensitive layer according to the grayscale value of the pixels written into the image, causing a spatially varying electric field to be formed on the liquid crystal layer. The change in the distribution of the electric field on the liquid crystal layer causes a change in the distortion of the liquid crystal arrangement. In this way, the vibration direction of the polarized light emitted through the light valve changes, and the intensity of the light emitted through the analyzer is spatially distributed, thereby achieving the purpose of spatial light modulation. At this time, the reflected light field can be expressed as:

[0076]

[0077] Where u(x,y) is the gray value of the pixel at (x,y) in the image being written, and δ(x0-x,y0-y) is the pulse function. The light field at this time can be regarded as a linear combination of several point light sources, and its complex amplitude U(x0,y0) is distributed according to the gray value u(x,y) of the loaded image.

[0078] The complex amplitude of the reflected light field of a reflection-type amplitude SLM is:

[0079] U(x0,y0)=A(x0,y0)r(x0,y0)

[0080] Where A(x0,y0) is the light wave that is irradiated onto the amplitude modulation array, and r(x0,y0) is the amplitude reflection coefficient, which is the gray-scale distribution of the target image with a complex background that has been horizontally inverted and loaded onto the amplitude modulation array.

[0081] Since the light source illuminating the amplitude modulation array is planar light, and its amplitude is assumed to be 1, the complex amplitude of the reflected light field of the reflection type is:

[0082] U(x0,y0)=r(x0,y0)

[0083] Where r(x0,y0) is the amplitude reflection coefficient.

[0084] In some embodiments, modulating monochromatic p-polarized light using an amplitude modulation array after loading the target image to obtain the spatial-temporal optical signal of the target image containing a complex background further includes: adjusting the light field intensity of the reflected light according to the gray value distribution of the loaded grayscale image to obtain the spatial-temporal optical signal of the target image containing a complex background.

[0085] S103. Convert the spatial-temporal optical signal of the target image containing a complex background into a spatial-frequency domain signal. The spatial-frequency domain signal is a spatial frequency distribution map containing spatial information of the target image.

[0086] In some embodiments, converting the spatial-temporal optical signal of a target image containing a complex background into a spatial-frequency domain signal specifically includes:

[0087] The amplitude modulation array of the reflective amplitude spatial light modulator (SLM) receives monochromatic p-polarized light reflected by the first polarizing beam splitter (PBS). The monochromatic p-polarized light is then reflected as monochromatic s-polarized light after amplitude modulation.

[0088] According to the first polarizing beam splitter PBS, the monochromatic s-polarized light is reflected in a direction perpendicular to the incident light path;

[0089] The reflected beam passes from the amplitude modulation array through the first polarization beam splitter PBS, and the focal length f of the lens is determined based on the distance of the beam when it reaches the Fourier lens in its propagation path.

[0090] When the spatial-temporal optical signal of a target image with a complex background passes through a Fourier lens, the Fourier lens converts it to the spatial frequency domain to obtain the spatial frequency domain signal.

[0091] Specifically, the incident p-polarized light is reflected as s-polarized light after being amplitude modulated by the reflective amplitude-type SLM. It is then reflected by the first PBS in a direction perpendicular to the incident light path. The distance from the amplitude-modulated array to the Fourier lens after passing through the PBS is exactly the focal length f of the Fourier lens.

[0092] When a light beam passes through multiple optical elements or systems, the image of the preceding element or system becomes the object of the following element or system. When the light field reflected by the amplitude modulation array is reflected by the PBS, it is equivalent to passing through a parallel plate. At this time, the image of the target is located in front of the amplitude modulation array, at a distance of:

[0093]

[0094] Where L is the side length of PBS and n is the refractive index of PBS, the distance from the reflected light field to the Fourier lens through the PBS should be equal to the focal length f of the Fourier lens to form a 4f system.

[0095] The relationship between the distance l1 from the PBS surface to the Fourier lens, the PBS side length L, the distance l2 from the amplitude modulation array to the PBS surface, and the focal length f of the Fourier lens can be obtained from the above formula:

[0096] l1+L+Δl=f

[0097] Substituting Δl into the above equation, we get...

[0098]

[0099] When the above relationships are satisfied, a 4f system is formed;

[0100] The reflected light field is reflected by the PBS and then passes through a Fourier lens to illuminate the front surface of the first DONN, where Fresnel diffraction occurs. The complex amplitude on the front surface of the first DONN is obtained using the Fresnel diffraction formula:

[0101] U(ξ,η)=c·FT{r(x0,y0)}=c·R(ξ,η)

[0102] Where c is a complex constant, ξ = x / λf, η = y / λf, which correspond to the positions of x and y in the spatial frequency domain, respectively. This transformation is called the standard Fourier transform, in which the target spatial domain optical signal with a complex background is converted to the spatial frequency domain.

[0103] In some embodiments, converting the spatial-temporal optical signal of a target image containing a complex background into a spatial-frequency domain signal specifically includes:

[0104] A two-dimensional Fourier transform is performed on the modulated s-polarized light to convert the spatial information of the target image into frequency information.

[0105] Based on the spatial frequency distribution map formed on the focal plane, the spatial frequency domain signal of the target image containing a complex background is determined by the signal after Fourier transform.

[0106] S104. Remove the background information of the spatial frequency domain signal and obtain the spatial frequency domain signal after background removal.

[0107] In some embodiments, removing background information from the spatial frequency domain signal to obtain the background-removed spatial frequency domain signal specifically includes:

[0108] The spatial frequency domain signal is filtered according to the first diffraction light neural network DONN;

[0109] The spatial frequency domain signal is obtained by nonlinearly activating the filtered spatial frequency domain signal using a transmissive phase-type spatial light modulator (SLM).

[0110] Specifically, the optical signal converted to the spatial frequency domain will be filtered by the first DONN. The parameters of the first DONN are trained on a computer using deep learning methods based on a specific filtering task dataset. Let the filtering function of the first DONN in the spatial frequency domain be H(ξ,η), then the complex amplitude of the optical field after filtering by the first DONN is:

[0111] U′(ξ,η)=U(ξ,η)H(ξ,η)

[0112] After filtering, some of the light signals containing background information will be removed;

[0113] Specifically, H(ξ,η) can be composed of the superposition of a vortex phase function, a random matrix phase function, and a Fresnel lens phase function. The vortex phase function modulates the light field, causing its wavefront phase to exhibit a spiral distribution. In this case, the beam carries a certain orbital angular momentum, and the light intensity also exhibits a ring-shaped distribution. For the target light signal converted to Fourier space, the fundamental frequency information located at the center of the spatial frequency domain is filtered out by the vortex phase function, so that the final output retains only mid-to-high frequency information.

[0114] The random matrix phase function, randomly distributed in two-dimensional space and ranging from 0 to 2π, is used to precisely modulate spatial frequency domain information to filter out complex backgrounds, retaining only the mid-to-high frequency information of the target. In the spatial domain, it is represented as the target shape. The Fresnel lens phase function is used to simulate a Fresnel lens, focusing or diverging the light beam. Since some energy is diffracted outside the system's field of view when the light field passes through the DONN, leading to a decrease in final image intensity, a Fresnel lens phase function needs to be added to the first DONN to converge its energy and ensure the final processing effect.

[0115] S105. Convert the background-removed spatial frequency domain signal into a background-removed spatial time domain optical signal. The background-removed spatial time domain optical signal is a spatial intensity distribution map containing spatial information of the target graphic.

[0116] In this process, after nonlinear activation of the transmissive phase-type SLM, the optical signal containing target information is enhanced while the remaining optical signal containing background information is weakened. At this point, the optical signal contains almost only target information. The basic principle of nonlinear activation is to set the gamma correction curve of the transmissive phase-type SLM to a pre-selected nonlinear activation function Φ(ξ,η)=e f(ξ,η) When the light field undergoes nonlinear activation, its complex amplitude can be expressed as:

[0117] U′(ξ,η)=U(ξ,η)H(ξ,η)Φ(ξ,η)=c·e f(ξ,η) R(ξ,η)H(ξ,η)

[0118] Where c is a complex constant and H(ξ,η) is the filter function.

[0119] The distances from the rear surface of the transmissive phase-type SLM and the front surface of the second DONN to the inverse Fourier lens are both equal to the focal length f of the inverse Fourier lens. At this point, the optical signal in the spatial frequency domain will undergo another Fourier transform. If the coordinates are reversed, it is equivalent to the nonlinearly activated optical signal undergoing an inverse Fourier transform and being transferred to the spatial domain. At this point, the optical signal in the spatial domain almost only contains target information, and its optical wave field can be expressed as:

[0120] U i (x i ,y i ) = FT -1 {c·e f(ξ,η) R(ξ,η)H(ξ,η)}

[0121] Where c is a complex constant and H(ξ,η) is the filter function.

[0122] S106. Classify the target image corresponding to the spatial-temporal optical signal after removing the background.

[0123] Specifically, the second DONN classifies the spatial domain light signal after removing the background. Its parameters are trained on a computer using deep learning methods based on the specific classification task dataset. The classification and recognition detector acquires the classified light signal and transmits it to the processing system for recognition, thereby realizing real-time classification and recognition of targets in complex backgrounds.

[0124] like Figure 2 As shown, this application also proposes a target image classification system based on complex backgrounds, the system comprising:

[0125] A laser is used to provide a continuous and stable light source for the entire system. The light it emits is monochromatic and linearly polarized with a fixed polarization direction.

[0126] A beam expander is used to diverge the monochromatic beam emitted by the laser, while collimating and shaping the beam so that it eventually passes through the subsequent system as parallel light. It also disperses the beam energy to avoid excessive power damage to other components and ensures that the beam illuminating the reflective amplitude-modulated SLM can fill the entire amplitude modulation array.

[0127] HWP is used to change the polarization direction of the linearly polarized light output by the laser and to modulate the polarization state of the beam incident on the amplitude modulation array into p-polarized light according to the requirements of the reflective amplitude modulation SLM, so as to ensure that the reflective amplitude modulation SLM can work normally.

[0128] PBS is used to split incident light into two perpendicular linearly polarized beams; the p-polarized beam passes through completely, while the s-polarized beam is reflected at a 45° angle and the outgoing direction is at a 90° angle to the p-polarized beam. Both the p-polarized beam and the s-polarized beam are linearly polarized and their polarization directions are perpendicular to each other.

[0129] The reflective amplitude-modulated liquid crystal lamp (SLM) is used to modulate the amplitude of incident p-polarized light. It consists of a sandwich structure composed of a photoconductive layer, a dielectric mirror, a liquid crystal layer, and a transparent conductive electrode (ITO) on a glass substrate. It is composed of many basic independent units to form a two-dimensional array liquid crystal panel. The processing module can control the voltage to adjust the rotation direction of the liquid crystal molecules through software, thereby achieving amplitude control of a single pixel. The reflected light from the modulation array is s-polarized light.

[0130] The target acquisition detector is used to acquire target images in the application scenario in real time for subsequent system processing;

[0131] The processing system is used to transmit the target image acquired by the target acquisition detector to the reflective amplitude-type SLM and the preset phase modulation parameters to the transmissive phase-type SLM, while acquiring the classification and recognition signal obtained by the classification and recognition detector.

[0132] A Fourier lens is used to convert optical signals in the spatial domain into optical signals in the spatial frequency domain, so that subsequent modulation of the optical signal can be performed in the spatial frequency domain. It has the same focal length f as the inverse Fourier lens, forming a 4f system.

[0133] The first DONN is used to filter and modulate optical signals in the spatial frequency domain. Its specific parameters are obtained through computer simulation training based on the dataset and manufactured by technologies such as 3D printing or photolithography.

[0134] Transmissive phase-type SLM is used to perform nonlinear activation operation on the spatial frequency domain optical signal after the first DONN filter.

[0135] An inverse Fourier lens is used to convert optical signals in the spatial frequency domain into optical signals in the spatial domain, so that subsequent processing and recognition of the optical signals can be carried out in the spatial domain. It has the same focal length f as the Fourier lens, forming a 4f system.

[0136] The second DONN is used to identify the filtered spatial domain optical signal. Its specific parameters are obtained through computer simulation training based on the dataset and are manufactured by technologies such as 3D printing or photolithography.

[0137] The classification and recognition detector is used to collect the light signal after classification and recognition by the second DONN and transmit the light signal to the processing system for recognition and processing.

[0138] This application also proposes a readable storage medium storing a computer program, which, when executed by a processor, causes the processor to perform the following steps:

[0139] Obtain continuous and stable monochromatic p-polarized light;

[0140] The target image is acquired and loaded into an amplitude modulation array. The monochromatic p-polarized light is modulated by the amplitude modulation array after loading the target image to obtain the spatial-temporal optical signal of the target image containing a complex background.

[0141] The spatial-temporal optical signal of a target image containing a complex background is converted into a spatial-frequency domain signal. The spatial-frequency domain signal is a spatial frequency distribution map containing spatial information of the target image.

[0142] Remove the background information from the spatial frequency domain signal to obtain the background-removed spatial frequency domain signal;

[0143] The background-removed spatial frequency domain signal is converted into a background-removed spatial time domain optical signal, which is a spatial intensity distribution map containing spatial information of the target graphic.

[0144] Classify the target image corresponding to the spatial-temporal optical signal after background removal.

[0145] This application also proposes a computer device, including a memory and a processor, wherein the memory stores a computer program, and the computer program is executed by the processor in the following steps:

[0146] Obtain continuous and stable monochromatic p-polarized light;

[0147] The target image is acquired and loaded into an amplitude modulation array. The monochromatic p-polarized light is modulated by the amplitude modulation array after loading the target image to obtain the spatial-temporal optical signal of the target image containing a complex background.

[0148] The spatial-temporal optical signal of a target image containing a complex background is converted into a spatial-frequency domain signal. The spatial-frequency domain signal is a spatial frequency distribution map containing spatial information of the target image.

[0149] Remove the background information from the spatial frequency domain signal to obtain the background-removed spatial frequency domain signal;

[0150] The background-removed spatial frequency domain signal is converted into a background-removed spatial time domain optical signal, which is a spatial intensity distribution map containing spatial information of the target graphic.

[0151] Classify the target image corresponding to the spatial-temporal optical signal after background removal.

[0152] Those skilled in the art will understand that implementing all or part of the processes in the above embodiments can be accomplished by instructing related hardware through a computer program. The program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0153] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0154] The embodiments described above are merely illustrative of several implementation methods of this application, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of this application's patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. The embodiments disclosed above are merely preferred embodiments of the present invention and should not be construed as limiting the scope of the present invention. Therefore, equivalent variations made according to the claims of this invention are still within the scope of this invention.

Claims

1. A method for classifying a target image based on a complex background, characterized in that, The method comprises: acquiring continuous and stable monochromatic p-polarized light; collecting a target image and loading the target image to an amplitude modulation array, modulating the monochromatic p-polarized light by using the amplitude modulation array loaded with the target image, and acquiring a spatial time-domain light signal of the target image containing a complex background; converting the spatial time-domain light signal of the target image containing the complex background into a spatial frequency domain signal, the spatial frequency domain signal being a spatial frequency distribution map containing spatial information of the target image, and when the spatial time-domain light signal of the target image containing the complex background passes through a Fourier lens, the Fourier lens converts it into the spatial frequency domain signal; removing background information of the spatial frequency domain signal to acquire a spatial frequency domain signal after removing the background, specifically including: filtering processing the spatial frequency domain signal according to a first diffraction light neural network (DONN); and performing nonlinear activation on the spatial frequency domain signal after filtering processing according to a transmission type phase spatial light modulator (SLM) to acquire the spatial frequency domain signal after removing the background; converting the spatial frequency domain signal after removing the background into a spatial time-domain light signal after removing the background, the spatial time-domain light signal after removing the background being a spatial intensity distribution map containing spatial information of the target image; classifying the target image corresponding to the spatial time-domain light signal after removing the background.

2. The method of claim 1, wherein, The acquisition of the continuous and stable monochromatic p-polarized light specifically comprises: collecting a continuous and stable monochromatic linearly polarized laser beam emitted by a laser; adjusting the monochromatic linearly polarized laser beam through a beam expander lens to acquire a parallel light beam after expansion; polarization modulation of the parallel light beam after expansion according to a half-wave plate; controlling the angle of the half-wave plate to acquire the continuous and stable monochromatic p-polarized light.

3. The method of claim 2, wherein, The collection of the target image and the loading of the target image to the amplitude modulation array further comprises polarization filtering of the continuous and stable monochromatic p-polarized light by a polarization beam splitter (PBS).

4. The method of claim 3, wherein, The modulation of the monochromatic p-polarized light by using the amplitude modulation array loaded with the target image to acquire the spatial time-domain light signal of the target image containing the complex background further comprises obtaining the spatial time-domain light signal of the target image containing the complex background according to the gray value distribution of the loaded gray scale image.

5. The method of claim 4, wherein, The conversion of the spatial time-domain light signal of the target image containing the complex background into the spatial frequency domain signal specifically comprises: the amplitude modulation array of the reflective amplitude type spatial light modulator (SLM) receives the monochromatic p-polarized light reflected by the first polarization beam splitter (PBS), and the monochromatic p-polarized light is reflected as monochromatic s-polarized light after amplitude modulation; the first polarization beam splitter (PBS) reflects the monochromatic s-polarized light to a direction perpendicular to the incident light path; the reflected light beam passes through the first polarization beam splitter (PBS) from the amplitude modulation array, and the focal length f of the lens is determined according to the distance of the light beam in the propagation path when reaching the Fourier lens; when the spatial time-domain light signal of the target image containing the complex background passes through the Fourier lens, the Fourier lens converts it to the spatial frequency domain to acquire the spatial frequency domain signal.

6. The method for classifying target images based on complex background according to claim 5, characterized in that, The converting the spatial time domain light signal of the target image containing complex background into a spatial frequency domain signal specifically comprises: performing two-dimensional Fourier transform on the modulated s-polarization to convert the spatial information of the target image into frequency information; forming a spatial frequency distribution map on the focal plane, and determining the spatial frequency domain signal of the target image containing complex background through the signal after Fourier transform.

7. The method of claim 6, wherein, The removing the background information of the spatial frequency domain signal to obtain a spatial frequency domain signal after removing the background specifically comprises: filtering the spatial frequency domain signal according to a first diffractive light neural network (DONN); performing nonlinear activation on the filtered spatial frequency domain signal according to a transmissive phase spatial light modulator (SLM) to obtain a processed spatial frequency domain signal.

8. A complex background based target image classification system, comprising: The system comprises a polarization acquisition unit, a time domain signal acquisition unit, a signal processing unit and a classification unit. The polarization acquisition unit is configured to acquire continuous and stable p-polarization. The time domain signal acquisition unit is configured to collect a target image, load the target image to an amplitude modulation array, modulate the p-polarization by using the amplitude modulation array loaded with the target image, and acquire a spatial time domain light signal of the target image containing complex background. The signal processing unit is configured to convert the spatial time domain light signal of the target image containing complex background into a spatial frequency domain signal, the spatial frequency domain signal being a spatial frequency distribution map containing spatial information of the target image, and when the spatial time domain light signal of the target image containing complex background passes through a Fourier lens, the Fourier lens converts the spatial time domain light signal into a spatial frequency domain signal; the signal processing unit is further configured to remove the background information of the spatial frequency domain signal to obtain a spatial frequency domain signal after removing the background, specifically by filtering the spatial frequency domain signal according to a first diffractive light neural network (DONN), performing nonlinear activation on the filtered spatial frequency domain signal according to a transmissive phase spatial light modulator (SLM) to obtain a spatial frequency domain signal after removing the background, converting the spatial frequency domain signal after removing the background into a spatial time domain light signal after removing the background, and the spatial time domain light signal after removing the background being a spatial intensity distribution map containing spatial information of the target image; and the classification unit is configured to classify the target image corresponding to the spatial time domain light signal after removing the background. 9.A readable storage medium storing a computer program, wherein the computer program is executed by a processor to make the processor perform the steps of the method in any one of claims 1 to 7. 10.A computer device comprising a memory and a processor, wherein the memory stores a computer program, and the computer program is executed by the processor to make the processor perform the steps of the method in any one of claims 1 to 7.

Images