Visible light broadband edge detection system and diffractive optical element design method

Abstract

The invention discloses a visible light broadband edge detection system and a diffractive optical element design method, and relates to the technical field of optical neural networks. The system comprises an information input module which comprises a light source and a mask plate and is used for converting object information into intensity information of a light beam so as to obtain an input light field; the all-optical edge detection module comprises a 4f system and a diffractive optical element and is used for outputting a light field after phase modulation; and the information acquisition module acquires intensity information of an output light field through a photoelectric detector, and the output light field is subjected to phase modulation through a 4f system and a diffractive optical element, so that broadband edge detection is realized. According to the design of the diffractive optical element, a differentiable end-to-end light field transmission simulation model is established, differentiable mapping from diffraction element structure parameters to an output light field is achieved, and the diffraction element structure parameters are optimized based on the deep learning technology. According to the end-to-end reverse design method provided by the invention, all-optical broadband edge detection with high diffraction efficiency can be realized without physical priori knowledge.

Description

Technical Field

[0001] This invention relates to the field of optical neural network technology, and in particular to a visible light broadband edge detection system and a design method for diffractive optical elements. Background Technology

[0002] With the rapid development of artificial intelligence technology, image processing is playing an increasingly important role in various fields such as medical imaging, autonomous driving, and industrial inspection. However, traditional electronic computers typically employ the von Neumann architecture, inevitably suffering from the "memory wall" and "power wall" bottlenecks. Furthermore, the development of Moore's Law has slowed significantly, and the size of silicon-based chip transistors is gradually approaching their physical limits. These challenges make existing electronic computing increasingly unable to meet the exponentially growing computing demands of the AI ​​era. Compared to electronic computing, optical computing possesses inherent advantages such as high speed, low power consumption, high bandwidth, and massively parallel processing, thus providing a new computing paradigm for overcoming these limitations.

[0003] Optical image edge detection has become an emerging target recognition and detection technology. In recent years, with the development of micro-nano fabrication technology and materials science, metasurfaces have become compact and multifunctional all-optical edge detection platforms due to their ability to manipulate the amplitude, phase, and polarization of the light field. Generally, there are two methods for all-optical image edge detection: the Green's function method and the spatial Fourier phase shift method. Using the Green's function method, metasurface structures can be directly designed in the spatial domain to meet the specific optical transfer function required for all-optical spatial differentiation and edge detection. However, although existing methods have achieved system miniaturization and integration, they inevitably have limitations such as requiring specific incident angles, polarization dependence, narrow operating bandwidth, and complex device design, making it difficult to realize complex and varied practical application scenarios. As a classic alternative, optical 4f systems utilize the Fourier transform capability of traditional lenses, combined with spatial frequency domain filtering, to achieve angle-insensitive and polarization-independent broadband edge detection, exhibiting strong robustness and stability. However, most edge detection architectures based on 4f systems use amplitude-type filtering elements, resulting in low light energy utilization. In recent years, phase-modulated devices such as metasurfaces and diffractive optical elements (DOEs) have been integrated into edge detection systems, achieving miniaturization and integration. However, traditional phase-modulated schemes typically rely on forward design based on prior knowledge and can only operate at a single wavelength, failing to achieve broadband edge detection and limiting their potential applications in real-world scenarios.

[0004] Therefore, how to effectively achieve broadband edge detection that is polarization-independent, incident angle-insensitive, and has high diffraction efficiency by eliminating dependence on prior physical knowledge remains to be explored. Summary of the Invention

[0005] The technical problem to be solved by this invention is to provide a visible light broadband edge detection system and a design method for diffractive optical elements. By introducing diffractive optical elements (DOEs) into a 4f system and learning the phase modulation matrix through end-to-end reverse design, polarization independence, incident angle insensitivity, high diffraction efficiency, and broadband edge detection can be achieved without prior physical knowledge.

[0006] In a first aspect, the present invention provides a visible light broadband edge detection system, comprising: The information input module, including a light source and a mask, is used to convert object information into the intensity information of the light beam, thereby obtaining the input light field; The all-optical edge detection module includes a 4f system and a diffractive optical element (DOE). The input light field undergoes a Fourier transform through the first lens in the 4f system. After the Fourier transform, the light field is phase-modulated by the DOE and then undergoes an inverse Fourier transform through the second lens in the 4f system to obtain the phase-modulated output light field. The phase modulation matrix of the DOE is obtained by end-to-end inverse learning using a differentiable optical field transmission simulation model and an electrical convolution operator. The information acquisition module is used to acquire the phase-modulated output light field through a photodetector and obtain the intensity information of the output light field.

[0007] Furthermore, the end-to-end reverse learning process of the phase modulation parameters of the diffractive optical element (DOE) includes: Design a diffractive optical system. Based on the light source type, detector, lens, and material of the diffractive optical element (DOE), establish a phase modulation model of the light field by the DOE using the forward propagation model of the light field and the Rayleigh-Sommerfeld formula. Given an input image set according to task requirements, a joint loss function is constructed based on the feature loss and diffraction efficiency ratio between the output light field and the target light field. The input image is used to obtain the target image through an electrical convolution operator, and the output image is obtained through a DOE phase modulation model of the light field. Based on the joint loss function, deep learning and error backpropagation algorithms are used to adjust the system structure during training and optimize the phase modulation matrix of the diffractive optical element (DOE).

[0008] Furthermore, constructing the joint loss function specifically includes: constructing a first loss function based on the mean square error loss between the predicted light field and the target light field; constructing a second loss function based on the difference between 1 and the diffraction efficiency ratio between the predicted light field and the target light field; and constructing a joint loss function based on the first loss function and the second loss function.

[0009] Furthermore, the information input module includes a laser, a pinhole filter, a collimating lens, and a mask; the laser emits laser light, which is expanded by the filter, then passes through the collimating lens, and finally the mask modulates the amplitude to obtain the input light field.

[0010] Secondly, the present invention provides a method for designing diffractive optical elements, comprising: Phase modulation model construction process: Design a diffractive optical system, and based on the light source type, detector, lens, and material of the diffractive optical element DOE, establish a phase modulation model of the diffractive optical element DOE on the light field using the forward propagation model of the light field and the Rayleigh-Sommerfeld formula; Loss function construction process: Construct a joint loss function based on the characteristic loss and diffraction efficiency ratio between the output light field and the target light field; Parameter learning process: The input image set is used to obtain the target image through the electrical convolution operator, and the output image is obtained through the phase modulation model of the diffractive optical element (DOE). Based on the joint loss function, the system structure is adjusted and the phase modulation matrix of the DOE is optimized during the training process using deep learning and error backpropagation algorithms.

[0011] Furthermore, a first loss function is constructed based on the mean square error loss between the predicted light field and the target light field, and a second loss function is constructed based on the difference between 1 and the diffraction efficiency ratio between the predicted light field and the target light field; a joint loss function is constructed based on the first loss function and the second loss function. The formula for the first loss function is:

[0012] Where n is the number of pixels in the image. Indicates the first i The output light field intensity at each sampling point I i Indicates the first i The target light field intensity at each sampling point; The formula for the second loss function is:

[0013] in, Indicates the first i The output light field intensity at each sampling point I imax Indicates the first i The maximum value of the target light field intensity at each sampling point.

[0014] Furthermore, the learned phase modulation matrix will be transformed into the step height matrix of the DOE to achieve phase modulation. The transformation formula is as follows:

[0015] in, Let λ be the difference between the refractive index of the DOE material and that of air, and λ be the wavelength of the light incident on the all-optical edge detection module. It is the phase matrix.

[0016] Furthermore, the DOE diffractive optical element is fabricated based on the step height matrix of the DOE.

[0017] Furthermore, the light source type is a coherent light source; the material of the diffractive optical element (DOE) is SK1300 material.

[0018] Furthermore, the diffractive optical element (DOE) is an eighth-order quantized DOE device.

[0019] The technical solutions provided in the embodiments of the present invention have at least the following technical effects: 1. By establishing a differentiable end-to-end optical field transmission simulation model, the structural parameters of diffraction components are mapped to the output optical field in a differentiable manner. The Fourier plane DOE design parameters of the 4f system are optimized by an end-to-end reverse design method, and broadband edge detection in the visible light band that is polarization independent, incident angle insensitive, and can be achieved is successfully realized.

[0020] 2. A joint loss function is constructed based on the first and second loss functions, which can achieve broadband edge detection and high diffraction efficiency (the calculated diffraction efficiency ratio is 74.89% in simulation and 63.10% in experiment, while the existing scheme based on certain physical prior knowledge has a diffraction efficiency ratio of 37%).

[0021] 3. The innovative end-to-end reverse design method does not rely on prior physical knowledge, which greatly promotes intelligent design and scalability, and expands its applicability in real-world scenarios. It is of great significance for accelerating its practical application in biological microscopy imaging and machine vision.

[0022] 4. Utilizing photons instead of electrons for computation offers faster processing speeds and lower energy consumption compared to traditional neural networks driven by electronic computers. It also solves problems inherent in existing Green's function-based methods, such as incident angle dependence, polarization sensitivity, and limited operating bandwidth. Furthermore, it addresses the limitations of 4f-based spatial filtering methods, which typically employ amplitude-type optical elements leading to limited diffraction efficiency, and the limitation of traditional phase-type modulation devices operating only at a single wavelength, restricting their application scope.

[0023] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of the present invention more apparent and understandable, specific embodiments of the present invention are described below. Attached Figure Description

[0024] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0025] Figure 1 This is a schematic diagram of the system framework in Embodiment 1 of the present invention; Figure 2 This is a schematic diagram of the optical path of the system in Embodiment 1 of the present invention; Figure 3 This is a schematic diagram illustrating the learning principle of the phase modulation matrix of the DOE in Embodiment 1 of the present invention. Figure 4 This is a simulation result diagram of visible light broadband edge detection in Embodiment 1 of the present invention; Figure 5 This is an experimental result diagram of visible light broadband edge detection in Embodiment 1 of the present invention; Figure 6 This is a flowchart illustrating the overall process of designing diffractive optical elements in Embodiment 2 of the present invention. Figure 7 This is a flowchart of the DOE preparation process in Embodiment 2 of the present invention. Detailed Implementation

[0026] This invention provides a visible light broadband edge detection system and a design method for diffractive optical elements. By introducing a diffractive optical element (DOE) into a 4f system and learning the phase modulation matrix through end-to-end reverse design, polarization independence, incident angle insensitivity, high diffraction efficiency, and broadband edge detection are achieved without prior physical knowledge.

[0027] The overall concept of the technical solutions in the embodiments of the present invention is as follows: By introducing a diffractive optical element (DOE) into the 4f system, the phase of light at different wavelengths can be modulated through careful design of its surface structure. The DOE corresponds to a specific refractive index difference at different wavelengths, and phase compensation can be performed based on these differences. Furthermore, the DOE is a bulk material and is polarization-insensitive. This allows the system to meet the requirements of polarization independence, angle insensitivity, and broadband achromaticity.

[0028] At different wavelengths, the refractive index of a material exhibits different wavelength responses (meaning that micro / nano structures of the same height will have different phase responses at different wavelengths). Traditional lenses experience dispersion during beam focusing because the focusing position differs for different wavelengths. This application utilizes a DOE (Difference of Optical Array) to perform phase compensation for different wavelengths, enabling light from different bands to be focused to the same position.

[0029] Existing forward design methods for DOEs rely on manual experience (requiring numerous structural design attempts) or theoretical formula derivation (such as the Green's function method) to achieve broadband edge detection. However, this application creatively employs a reverse design approach. By simulating the optical field transmission process of an optical neural network composed of a 4f system and a DOE in a computer, a differentiable end-to-end optical field transmission simulation model is established. This enables the differentiable mapping of diffraction element structural parameters to the output optical field. Based on electrical convolution operators and deep learning techniques, only the joint loss function needs to be optimized to obtain the phase modulation matrix of the DOE, thus yielding the height of the DOE's micro / nano units without requiring prior physical knowledge. Ultimately, broadband phase modulation of the optical field can be achieved by reverse-designing the height dimensions of the micro / nano unit structure.

[0030] The overall implementation process can be summarized as follows: establish a simulation model, optimize the phase parameters of the diffractive optical element (DOE) through forward propagation, error calculation, backpropagation, gradient calculation and phase update processes based on the given training and test sets for the task; after the simulation is completed, fabricate the DOE, and then build an actual optical experimental system to complete the broadband edge detection task.

[0031] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0032] It should be noted that the following detailed descriptions are exemplary and intended to provide further explanation of this application. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains. Example

[0033] This embodiment provides a visible light broadband edge detection system, such as Figure 1 and Figure 2 As shown, including; The information input module, including a light source and a mask, is used to convert object information into beam intensity information, thereby obtaining the input light field. Specifically, the information input module may include a laser, a pinhole filter, a collimating lens, and a mask; the laser emits laser light, which is expanded by the filter, then passes through the collimating lens, and finally the mask modulates the amplitude to obtain the input light field.

[0034] The all-optical edge detection module includes a 4f system and a diffractive optical element (DOE). The input light field undergoes a Fourier transform through the first lens in the 4f system. After the Fourier transform, the light field is phase-modulated by the DOE and then undergoes an inverse Fourier transform through the second lens in the 4f system to obtain the phase-modulated output light field. The phase modulation matrix of the DOE is obtained by end-to-end inverse learning using a differentiable optical field transmission simulation model and an electrical convolution operator. The learning process can be described as follows: Figure 3 As shown.

[0035] The information acquisition module is used to acquire the phase-modulated output light field through a photodetector, thereby obtaining the intensity information of the output light field. Because the output light field is modulated by a 4f system and a DOE phase, it can achieve functions such as high diffraction efficiency and broadband edge detection.

[0036] Specifically, the end-to-end reverse learning process of the phase modulation parameters of the diffractive optical element (DOE) includes: Design a diffractive optical system. Based on the light source type, detector, lens, and material of the diffractive optical element (DOE), establish a phase modulation model of the light field by the DOE using the forward propagation model of the light field and the Rayleigh-Sommerfeld formula. Given an input image set according to task requirements, a joint loss function is constructed based on the feature loss and diffraction efficiency ratio between the output light field and the target light field. The input image is processed by an electrical convolution operator to obtain the target image, and the output image is obtained by using a DOE (Diffractive Optical Element) phase modulation model of the light field. A first loss function is constructed based on the mean square error loss between the predicted and target light fields, and a second loss function is constructed based on the difference between 1 and the ratio of diffraction efficiency between the predicted and target light fields. A joint loss function is constructed based on the first and second loss functions. Based on the joint loss function, deep learning and error backpropagation algorithms are used to adjust the system structure and optimize the phase modulation matrix of the DOE during training.

[0037] The optimization is performed iteratively by calculating the error between the output image and the target image. Training ends when the maximum number of iterations is reached. The diffraction efficiency is calculated using the following formula: η peak = ( I out ) / ( I inmax ).

[0038] After the simulation design is completed, the DOE phase modulation parameters are converted into a height matrix. Then, the DOE is fabricated. Finally, the actual system is built using the physical DOE to realize phase modulation.

[0039] Figure 4This figure shows the simulation results of broadband edge detection in visible light based on a learnable phase-type diffractive optical element (DOE) in this embodiment. Taking a rectangular input image as an example, simulations were performed at wavelengths of 633 nm, 532 nm, and 450 nm. The figure shows the edge detection output results of the rectangle at different wavelengths. At wavelengths of 633, 532, and 450 nm, the edge detection accuracies w1, w2, and w3 of the rectangle are 48 μm, 42 μm, and 36 μm, respectively. The full width at half maximum (FWHM) at different wavelengths demonstrates that the designed DOE can achieve broadband edge detection.

[0040] Figure 5 This figure shows the experimental results of broadband edge detection in visible light based on a learnable phase-type diffractive optical element (DOE) according to an embodiment of the present invention. At wavelengths of 633, 532, and 450 nm, the edge detection accuracies of the rectangle's full width at half maximum (FWHM) w4, w5, and w6 are 48 μm, 45.6 μm, and 38.4 μm, respectively. The experimental results further demonstrate that the learnable phase-type diffractive optical element (DOE) designed in this application can achieve broadband edge detection. Example

[0041] This embodiment provides a method for designing diffractive optical elements, such as Figure 6 As shown, it includes: S1-1, Phase Modulation Model Construction Process: Design a diffractive optical system, and based on the light source type, detector, lens, and material of the diffractive optical element DOE, establish a phase modulation model of the diffractive optical element DOE on the light field using the forward propagation model of the light field and the Rayleigh-Sommerfeld formula. S1-2, Loss function construction process: Construct a joint loss function based on the ratio of characteristic loss and diffraction efficiency between the output light field and the target light field; S2. Parameter Learning Process: The input image set is processed by an electrical convolution operator to obtain the target image, and the output image is obtained by the phase modulation model of the DOE (Diffractive Optical Element). The root mean square error between the predicted light field and the target light field is constructed as the first loss function, and the difference between the ratio of the diffraction efficiency of the predicted light field and the target light field is constructed as the second loss function. A joint loss function is constructed based on the first and second loss functions. According to the joint loss function, deep learning and error backpropagation algorithms are used to adjust the system structure and optimize the phase modulation matrix of the DOE during the training process.

[0042] In one specific embodiment, the diffractive optical element design method is implemented through the following steps: Step 1. Design the optical system according to the mission, selecting the light source type, detector, lens, and DOE material. A coherent light source is selected as the light source; SK1300 material is selected for the diffractive optical element (DOE).

[0043] Step 2. To achieve end-to-end intelligent optimization design, this embodiment establishes a forward propagation model of the optical field. In this work, this embodiment uses a quadratic phase shift to represent the modulation of the optical field by a conventional lens. After passing through a conventional lens with a focal length of f, its expression is:

[0044] in, k =2π / λ represents the wave vector, and λ represents the wavelength. These are the position coordinates of pixels in the input image, where the zero coordinate is located at the center. It is through the input light field in front of the lens. It is the output light field modulated by the lens. j Represents an imaginary number.

[0045] Step 3. Based on the principles of Fourier optics, the Rayleigh-Sommerfeld formula is used to describe the propagation of light in free space. Its expression is:

[0046] in FFT and FFT -1 These represent the Fourier transform and the inverse Fourier transform, respectively. The light field distribution in the L-layer plane, For the light field distribution in the L+1 layer plane, H f ( f x , f y ) is the angular spectral transfer function, and its expression is:

[0047] in f x and f y It is the spatial frequency, and z is the distance that light travels in free space.

[0048] Step 4. Set the phase parameters of the DOE based on the selected material and mission objectives. Select phase as the modulation parameter and set the amplitude modulation coefficient to 1. Its expression is:

[0049] In this work, a phase-type mask is designed, with amplitude a(x,y) = 1, and optical losses are ignored. Therefore, the modulation of the optical field by the DOE can be described as follows:

[0050] Step 5. Perform a simulation of visible light broadband edge detection based on a learnable phase-type diffraction optical element using a given training set, test set, and objective function. During the simulation, this embodiment will output an image. and target image I i The root mean square error is used as the first loss function, and its expression is:

[0051] Where n is the number of pixels in the image. Indicates the first i The output light field intensity at each sampling point I i Indicates the first i The target light field intensity at each sampling point.

[0052] The difference between 1 and the ratio of diffraction efficiency of the output image and the target image is used as the second loss function, and its expression is:

[0053] in, Indicates the first i The output light field intensity at each sampling point I imax Indicates the first i The maximum value of the target light field intensity at each sampling point.

[0054] Then, a joint loss function is constructed based on the first loss function and the second loss function (the ratio of the first loss function to the second loss function is set to 99:1).

[0055] Step 6. Backpropagate the calculated joint loss error and calculate the gradient. Use the Adaptive Moments Estimation (ADAM) method for gradient descent, with a learning rate of 0.01 and 30 iterations.

[0056] Step 7. After the simulation training reaches the set maximum number of iterations, calculate the diffraction efficiency ratio between the output light field and the target light field, and the training ends.

[0057] Step 8. After the simulation training is completed, the phase modulation parameters of the DOE are converted into step heights and then fabricated.

[0058]

[0059] in, Let λ be the difference between the refractive index of the DOE material and that of air, and λ be the wavelength of the light incident on the all-optical edge detection module. It is the phase matrix. It is the height matrix of the DOE.

[0060] The DOE (Diffractive Optical Element) is fabricated based on the step height matrix of the DOE. Figure 7 This is a flowchart illustrating the fabrication process of DOE in this embodiment of the invention. The fabrication process includes steps such as cleaning, spin coating, photolithography and development, and etching.

[0061] In this embodiment, SK1300 material was selected for the DOE. The fabrication process specifically includes: first, SK1300 was used for substrate cleaning and photoresist homogenization; next, a high-precision electron beam lithography system was used to expose the pattern onto the photoresist; then, after development, an ion beam etching process was used to transfer the pattern to the SK1300 substrate at a precise depth. After three overlay and etching cycles, an eighth-order quantization DOE device was finally fabricated. After cleaning, a profilometer and diffraction efficiency testing system were used to rigorously inspect the morphology, size, and optical performance of the micro / nano structure to ensure compliance with standards.

[0062] By building a practical system and placing the prepared DOE into the Fourier plane of the 4f system to achieve phase modulation, polarization independence, incident angle insensitivity, high diffraction efficiency, and broadband edge detection can be achieved.

[0063] The technical solutions provided in the embodiments of the present invention have at least the following technical effects: 1. By establishing a differentiable end-to-end optical field transmission simulation model, the structural parameters of diffraction components are mapped to the output optical field in a differentiable manner. The Fourier plane DOE design parameters of the 4f system are optimized by an end-to-end reverse design method, and broadband edge detection in the visible light band that is polarization independent, incident angle insensitive, and can be achieved is successfully realized.

[0064] 2. A joint loss function is constructed based on the first and second loss functions, which can achieve broadband edge detection and high diffraction efficiency (the calculated diffraction efficiency ratio is 74.89% in simulation and 63.10% in experiment, while the existing scheme based on certain physical prior knowledge has a diffraction efficiency ratio of 37%).

[0065] 3. The innovative end-to-end reverse design method does not rely on prior physical knowledge, which greatly promotes intelligent design and scalability, and expands its applicability in real-world scenarios. It is of great significance for accelerating its practical application in biological microscopy imaging and machine vision.

[0066] 4. Utilizing photons instead of electrons for computation offers faster processing speeds and lower energy consumption compared to traditional neural networks driven by electronic computers. It also solves problems inherent in existing Green's function-based methods, such as incident angle dependence, polarization sensitivity, and limited operating bandwidth. Furthermore, it addresses the limitations of 4f-based spatial filtering methods, which typically employ amplitude-type optical elements leading to limited diffraction efficiency, and the limitation of traditional phase-type modulation devices operating only at a single wavelength, restricting their application scope.

[0067] While specific embodiments of the present invention have been described above, those skilled in the art should understand that the specific embodiments described are merely illustrative and not intended to limit the scope of the present invention. Equivalent modifications and variations made by those skilled in the art in accordance with the spirit of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A visible light broadband edge detection system, characterized in that, include: The information input module, including a light source and a mask, is used to convert object information into the intensity information of the light beam, thereby obtaining the input light field; The all-optical edge detection module includes a 4f system and a diffractive optical element (DOE). The input light field undergoes a Fourier transform through the first lens in the 4f system. After the Fourier transform, the light field is phase-modulated by the DOE and then undergoes an inverse Fourier transform through the second lens in the 4f system to obtain the phase-modulated output light field. The phase modulation matrix of the DOE is obtained by end-to-end inverse learning using a differentiable optical field transmission simulation model and an electrical convolution operator. The information acquisition module is used to acquire the phase-modulated output light field through a photodetector and obtain the intensity information of the output light field.

2. The system according to claim 1, characterized in that: The end-to-end reverse learning process of the phase modulation parameters of the diffractive optical element (DOE) includes: Design a diffractive optical system. Based on the light source type, detector, lens, and material of the diffractive optical element (DOE), establish a phase modulation model of the light field by the DOE using the forward propagation model of the light field and the Rayleigh-Sommerfeld formula. Given an input image set according to task requirements, a joint loss function is constructed based on the feature loss and diffraction efficiency ratio between the output light field and the target light field. The input image is used to obtain the target image through an electrical convolution operator, and the output image is obtained through a DOE phase modulation model of the light field. Based on the joint loss function, deep learning and error backpropagation algorithms are used to adjust the system structure during training and optimize the phase modulation matrix of the diffractive optical element (DOE).

3. The system according to claim 2, characterized in that, The construction of the joint loss function specifically includes: constructing a first loss function based on the mean square error loss between the predicted light field and the target light field; constructing a second loss function based on the difference between 1 and the diffraction efficiency ratio between the predicted light field and the target light field; and constructing a joint loss function based on the first loss function and the second loss function.

4. The system according to claim 1, characterized in that: The information input module includes a laser, a pinhole filter, a collimating lens, and a mask. The laser emits laser light, which is expanded by the filter, then passes through the collimating lens, and finally the mask modulates the amplitude to obtain the input light field.

5. A method for designing diffractive optical elements, characterized in that, include: Phase modulation model construction process: Design a diffractive optical system, and based on the light source type, detector, lens, and material of the diffractive optical element DOE, establish a phase modulation model of the diffractive optical element DOE on the light field using the forward propagation model of the light field and the Rayleigh-Sommerfeld formula; Loss function construction process: Construct a joint loss function based on the characteristic loss and diffraction efficiency ratio between the output light field and the target light field; Parameter learning process: The input image set is used to obtain the target image through the electrical convolution operator, and the output image is obtained through the phase modulation model of the diffractive optical element (DOE). Based on the joint loss function, the system structure is adjusted and the phase modulation matrix of the DOE is optimized during the training process using deep learning and error backpropagation algorithms.

6. The method according to claim 5, characterized in that: A first loss function is constructed based on the mean square error loss between the predicted and target light fields; a second loss function is constructed based on the difference between 1 and the diffraction efficiency ratio between the predicted and target light fields; a joint loss function is constructed based on the first and second loss functions. The formula for the first loss function is: Where n is the number of pixels in the image. Indicates the first i The output light field intensity at each sampling point I i Indicates the first i The target light field intensity at each sampling point; The formula for the second loss function is: in, Indicates the first i The output light field intensity at each sampling point I imax Indicates the first i The maximum value of the target light field intensity at each sampling point.

7. The method according to claim 5, characterized in that: The learned phase modulation matrix is ​​transformed into the step height matrix of the DOE to achieve phase modulation. The transformation formula is as follows: in, Let λ be the difference between the refractive index of the DOE material and that of air, and λ be the wavelength of the light incident on the all-optical edge detection module. It is the phase matrix.

8. The method according to claim 7, characterized in that: The DOE (Diffractive Optical Element) is fabricated based on the step height matrix of the DOE.

9. The method according to claim 5, characterized in that: The light source type is a coherent light source; the material of the diffractive optical element (DOE) is SK1300 material.

10. The method according to claim 5, characterized in that: The diffractive optical element (DOE) is an eighth-order quantized DOE device.

Images