Hybrid optoelectronic encryption method and system based on physical diffraction layer

Abstract

The application discloses a hybrid optoelectronic encryption method and system based on a physical diffraction layer, relates to the technical field of optical computing and information security, and comprises the following steps: S1, input digital information is compressed and encoded into a low-resolution phase mask through a convolutional neural network and is loaded onto a wavefront modulator; S2, a double-wavelength illumination source is used to irradiate the low-resolution phase mask on the wavefront modulator, so that a wavelength-dependent diffraction pattern is generated; S3, the diffraction pattern is phase-modulated through a spatial light modulator of a diffraction layer, and optical decoding is completed, and the diffraction layer is provided with a physically embedded phase distribution; S4, the phase-modulated diffraction light is transmitted forwardly, and an optical intensity distribution is formed on a detection surface; and S5, a remote computer reads the optical intensity distribution transmitted by a positive detector and a negative detector, and calculates the type of the code through differential operation. The application overcomes the replicability and efficiency defects of the existing optical scheme, and provides a new high-security communication paradigm.

Description

Technical Field

[0001] This invention relates to the fields of optical computing and information security technology. More specifically, this invention relates to a hybrid optoelectronic encryption method and system based on a physical diffraction layer. Background Technology

[0002] Traditional encryption methods, such as symmetric encryption (AES) and asymmetric encryption (RSA, ECC), primarily rely on the complexity of mathematical computation to ensure security. The core of these methods lies in the confidentiality of the key and the irreversibility of the algorithm. For example, the RSA algorithm is based on the difficulty of factoring large integers, while AES relies on the complexity of multiple rounds of substitution and permutation operations. In the era of classical computing, these methods were widely used in digital communication, e-commerce, and network security, providing a certain level of protection. However, with the rapid development of computing technology, especially the rise of quantum computing, traditional encryption faces serious security risks.

[0003] The emergence of quantum computing marks a revolutionary shift in computing paradigms. Quantum computers, utilizing the principles of superposition and entanglement, can solve problems in polynomial time that require exponential time in classical computers. For example, Shor's algorithm can efficiently factor large integers, thus breaking public-key encryption systems such as RSA and ECC. Grover's algorithm accelerates symmetric key search, reducing the complexity of breaking AES-128 from 2^128 to 2^64, meaning that existing encryption standards will become vulnerable in the quantum era. Furthermore, traditional encryption suffers from side-channel attacks, key management vulnerabilities, and hardware implementation flaws, such as key theft through power analysis or timing attacks, further reducing security.

[0004] To address these challenges, optical encryption technology has emerged. Optical encryption leverages the wavefront modulation, multidimensional parallelism, and physical properties of light, offering potential advantages over electronic encryption. For example, optical systems can hide information through phase masks, speckle patterns, or holography, possessing high-speed processing (light-speed level) and high-dimensional encoding (amplitude, phase, polarization). Existing optical encryption schemes include Double Random Phase Encryption (DRPE) and Optical Neural Networks (ONN), which introduce physical randomness and improve resistance to quantum attacks. However, existing optical encryption still has significant limitations: first, it relies on a single optical modulation, making it vulnerable to reverse engineering attacks (such as reconstructing the phase through numerical simulation); second, the keys are often in digital form, which can be copied electronically; third, the lack of deep integration between electronics and optics leads to low encoding efficiency and an inability to handle complex data; and fourth, it ignores wavelength-dependent diffraction characteristics, resulting in insufficient key diversity and susceptibility to noise interference. Summary of the Invention

[0005] One object of the present invention is to solve at least the above-mentioned problems and / or defects, and to provide at least the advantages described below.

[0006] To achieve these objectives and other advantages of the present invention, a hybrid optoelectronic encryption method based on a physical diffraction layer is provided, comprising:

[0007] S1. The input digital information is compressed and encoded into a low-resolution phase mask through a convolutional neural network and loaded onto the wavefront modulator to complete the electronic encoding.

[0008] S2. Use dual-wavelength illumination sources to illuminate the low-resolution phase mask on the wavefront modulator to generate a wavelength-dependent diffraction pattern.

[0009] S3. The diffraction pattern is phase-modulated by the spatial light modulator of the diffraction layer to complete optical decoding. The diffraction layer has a physically embedded phase distribution.

[0010] S4. The phase-modulated diffracted light propagates forward and forms an intensity distribution on the detector surface;

[0011] S5. The remote computer reads the light intensity distribution transmitted by the positive and negative detectors and calculates the coded category through differential operation.

[0012] Preferably, in S1, the convolutional neural network is constructed using the PyTorch framework, the low-resolution phase mask is 20×20 pixels, and the value of each pixel ranges from 0 to 2π.

[0013] Preferably, in S3, the resolution of the spatial light modulator is 1920x1080;

[0014] The diffraction layer is a photosensitive resin phase plate obtained by 3D printing, and the accuracy of the diffraction layer is 0.1μm and the thickness is 200μm.

[0015] The value of each neuron on the diffraction layer ranges from 0 to 2π;

[0016] The diffraction layer is obtained through training a diffraction neural network, and the network parameters are optimized during the training of the diffraction neural network using the joint loss function Loss as follows:

[0017]

[0018] In the above formula, M Indicates the number of categories. g m The one-hot encoding representing the true class; if the sample belongs to class m, then... Otherwise, it equals zero. C mThis represents the predicted probability after the model output has been processed by Softmax. β Indicates the weighting coefficient. N Indicates the number of samples. W Indicates the number of pixels or feature points. y i,w Indicates the first i The sample at the th The true value of each position, Indicates the first i The sample at the th Predicted values ​​for each location.

[0019] Preferably, in S3, the light field modulated by the spatial light modulator f ( x , y , z = z 1) Characterized by the following formula:

[0020]

[0021] In the above formula, This represents the axial transmission distance between the input layer and the modulation layer; * represents the convolution operation. f 1( x , y , z = z 0) indicates z = z The complex optical field at position 0, Represents the impulse response function. z 0 indicates the position of the input plane of the diffraction network. z 1 indicates the axial position of the first diffraction modulation layer. z This indicates the propagation distance of the diffraction layer.

[0022] Preferably, in S4, the light intensity distribution I ( x , y It is characterized by the following formula:

[0023] In the above formula, Indicates the first m -1 diffraction layer and the first m The axial distance between each diffraction layer.

[0024] Preferably, in S4, the detection surface is divided into 10 sub-regions, and the size of each detection region is 10×10 pixels.

[0025] Preferably, in S5, the light intensity distribution captured independently by the positive and negative detectors after the low-resolution phase mask is illuminated by a dual-wavelength illumination source is set as follows: and ;

[0026] The remote computer then performs a difference calculation using the following formula to obtain the differentiated score. S Score :

[0027]

[0028] In the above formula, λa and λb are the different wavelengths corresponding to the dual-wavelength illumination source.

[0029] A hybrid optoelectronic encryption system, applied in a hybrid optoelectronic encryption method based on a physical diffraction layer, includes:

[0030] An electronic coding module that compresses and encodes input digital information into a low-resolution phase mask;

[0031] An optical decoding module with a non-copyable physical key;

[0032] A dual-wavelength illumination source that produces wavelength-dependent diffraction patterns by illuminating an coded phase mask.

[0033] A beam splitter that works in conjunction with a dual-wavelength illumination source to split the main optical path into two paths, with the two output optical paths of the beam splitter respectively matching the locations of the electronic encoding module and the optical decoding module;

[0034] Positive and negative detectors are positioned downstream of the optical decoding module to capture the intensity distribution of the diffracted light field.

[0035] The electronic coding module is configured to employ a wavefront modulator, and the optical decoding module is configured to include a spatial light modulator and a diffraction layer that works in conjunction with it.

[0036] Both the positive and negative detectors are configured to use CCD or CMOS sensors.

[0037] Preferably, the dual-wavelength illumination source is configured to include:

[0038] Laser I and Laser II, which generate two coherent light sources;

[0039] Optical component I, used in conjunction with laser I to construct the upper branch optical path;

[0040] Optical component II, used in conjunction with laser II to construct the lower branch optical path;

[0041] The optical component I includes:

[0042] Collimating lens I that works in conjunction with the output side of laser I;

[0043] The output light from collimating lens I is transmitted to the beam splitter of the beam splitter;

[0044] The optical component II includes:

[0045] Collimating lens II that works in conjunction with the output side of laser II;

[0046] It is positioned downstream of collimating lens II to reflect the output of lens II to the reflector of the beam splitter.

[0047] The present invention has at least the following beneficial effects:

[0048] Firstly, this invention innovatively uses electronic convolutional neural networks (CNNs) for information encoding, with optical diffraction layers serving as the physical embedded key, and introduces a dual-wavelength differential detection mechanism. Through the physically non-replicable diffraction structure, secure key distribution and use are ensured, while simultaneously integrating efficient electronic compression with optical parallel computing to achieve quantum-secure encrypted transmission.

[0049] Secondly, this invention not only solves the computational dependency risk of traditional encryption, but also overcomes the easy replication and efficiency defects of existing optical schemes, providing a new high-security communication paradigm.

[0050] Other advantages, objectives and features of the present invention will become apparent in part from the following description, and in part from those skilled in the art through study and practice of the invention. Attached Figure Description

[0051] Figure 1 This is a block diagram of the hybrid optoelectronic encryption system in one embodiment of the present invention;

[0052] Among them, there are laser I-1, laser II-2, collimating lens I-3, collimating lens II-4, beam splitter-5, reflector-6, beam splitter-7, electronic encoding module-8, optical decoding module-9, and positive and negative detector-10. Detailed Implementation

[0053] The present invention will now be described in further detail with reference to the accompanying drawings, so that those skilled in the art can implement it based on the description.

[0054] This invention proposes a physical key decryption scheme for encoding encryption (i.e., a photoelectric hybrid encryption method). By using a physically non-replicable diffraction structure, it ensures the secure transmission of encrypted information, avoids the risk of electronic encoding and decoding being cracked, and provides a new paradigm for secure encryption and decryption.

[0055] An implementation of a hybrid optoelectronic encryption method based on a physical diffraction layer mainly includes the following aspects:

[0056] 1. System Hardware Construction

[0057] 1.1 Image and Encryption Encoding Module Based on Convolutional Neural Network: A convolutional neural network is constructed using the PyTorch framework, with an input size of 28x28, 3 hidden layers (3x3 convolutional kernels), and an output low-resolution phase mask (or low-resolution phase encoding).

[0058] The electronic encoder based on convolutional neural networks first encodes the original target to be transmitted into a two-dimensional low-resolution coded representation. The coded pattern consists of 20×20 pixels, and the value of each pixel ranges from 0 to 2π. This coded phase pattern serves as the light wavefront for task optimization, carrying the target discrimination information. This coded information cannot be directly recognized by the human eye or directly deciphered by a computer.

[0059] 1.2 Optical Module: The target encoding representation generated by the electronic convolutional neural network is then loaded onto a liquid crystal spatial light modulator. Dual-wavelength coherent light is used to illuminate the target on the modulator, allowing the target information to propagate forward in free space under dual-wavelength illumination. The optical interconnections between pixel values ​​follow Fresnel diffraction propagation.

[0060] (1)

[0061] in, , representing the distance between neurons, λ representing the illumination wavelength, j representing the imaginary unit, h ( x , y , z ) represents the impulse response function. z 1 indicates the axial position on the detection surface. z This represents the propagation distance of the diffraction layer. For a diffraction modulation layer, each neuron receives the optical signal emitted by each pixel in the input encoding and simultaneously emits an optical signal to each pixel unit on the output detector surface. Specifically, located at z= z The complex amplitude transmission coefficient t1 of the modulation layer at position 1 can be expressed as: , Let α represent the phase. Considering that typical spatial light modulators can only modulate amplitude or only modulate phase, but not both simultaneously, and since pure phase modulation has higher diffraction efficiency, we adopt pure phase modulation here, and set the amplitude coefficient of each neuron to a constant, i.e., α1=1, where α1 represents the amplitude coefficient, with a value ranging from 0 to 1. Therefore, the light field modulated by the diffraction modulation layer in this scheme can be expressed as:

[0062] (2)

[0063] in, This represents the axial transmission distance between the input layer and the modulation layer, and * represents the convolution operation.

[0064] 1.3 The transmitted wavefront is modulated by a phase modulator in space. The modulator is composed of an SLM with a resolution of 1920x1080 (Holoeye model). The diffraction layer is 3D printed from photosensitive resin (accuracy 0.1μm) with a thickness of 200μm. Its phase distribution is jointly optimized by deep learning methods, and the value of each neuron ranges from 0 to 2π.

[0065] 1.4 The modulated wavefront generates an intensity distribution on the detector surface:

[0066] (3)

[0067] The intensity distribution was captured by a CCD camera (1024×1024 resolution). To accurately identify the target category, the detection surface was divided into 10 sub-regions (corresponding to 10 categories), each with a size of 10×10 pixels. Finally, the intensity within each sub-region was summed to obtain a vector of size 10, corresponding to the score for each category. Considering the non-negativity of optical neural networks, this scheme employs dual-wavelength illumination to improve the accuracy of coded target recognition, using these as positive and negative signals on the detection surface respectively. Specifically, the system uses wavelengths of... as well as The light sources illuminate the targets respectively, and the corresponding intensity distributions are shown. and The light is captured independently by positive and negative detectors. After modulation, the generated light intensity is recorded in a sub-region of the detector, with each sub-region representing a type of label. The corresponding differential fraction is calculated as follows:

[0068] (4)

[0069] in, S Score This represents the fraction after normalized difference.

[0070] 1.5 This wavelength-domain differential operation can effectively amplify the discriminative features encoded in the phase modulation. The final classification decision is made by identifying the sub-region with the largest differential signal, which corresponds to the predicted target category.

[0071] 2. Detailed description of the training process

[0072] This scheme employs deep learning joint optimization, where the modulation parameters of the optics and the parameters of the electrical coding are completed in a data-driven manner during the deep learning process. Therefore, the optical and electrical modules can achieve end-to-end collaborative optimization. Through this joint training mechanism, the network can not only automatically learn the optimal phase modulation strategy and electrical coding weights, but also fully utilize the mapping relationship between input and output during training, thereby improving the overall robustness and accuracy of the system.

[0073] 2.1 Data preparation: The MNIST dataset (60,000 training samples) was selected. Before each iteration, the samples were randomly rotated by 0°, 90°, 180°, and 270°, randomly flipped, and subjected to contrast perturbation to improve robustness. The labels were converted into one-hot vectors (10-dimensional).

[0074] 2.2 Forward Propagation: The input image is encoded into a phase mask φ(x,y) by CNN; the mask is loaded with SLM, dual-wavelength illumination, and propagated through the diffraction layer at a distance of z=10cm. The light field U(x,y,z)=FFT[exp(iφ(x,y))] (Fourier transform to simulate diffraction).

[0075] 2.3 The detection surface is divided into 10 detection sub-regions, each with a size of 10×10. The sum of the light intensity within each sub-region is the corresponding detection score.

[0076] 2.4 Differential Calculation: Positive Detector negative detector Then the difference vector S m,out Characterized by the following formula:

[0077]

[0078] in, Indicates wavelength as The corresponding intensity distribution Indicates wavelength as The corresponding intensity distribution Indicates wavelength as The corresponding complex optical field, Indicates wavelength as The corresponding complex light field; after normalizing the difference vector, it is compared with the label. In this scheme, the difference vector calculation is used for one-hot decoding, which is suitable for multi-class information classification and has an accuracy of >90%.

[0079] 2.5 Loss Optimization:

[0080] To optimize network parameters, a joint loss function is used. During network training, the joint loss function (Loss) is minimized, and end-to-end learning is achieved by combining phase constraints, thereby optimizing the parameters of the optical modulation layer and the electrical coding layer. The specific joint loss function (LOSS) is represented by the following formula:

[0081]

[0082] In the above formula, the first term is to enable the network to correctly identify the category of the input target. M Indicates the number of categories. g m The one-hot encoding representing the true class (i.e., the target label), if the sample belongs to the m-th class, then Otherwise, it equals zero. C m The first term represents the predicted probability after the model output and the softmax effect. The second term aims to improve the detection efficiency of light intensity distribution in sub-regions on the detection surface. β This represents the weighting coefficient, used to adjust the importance of this part of the loss. N Indicates the number of samples. W Indicates the number of pixels or feature points. y i,w Indicates the first i The sample at the th The true value of each position, Indicates the first i The sample at the th Predicted values ​​for each location.

[0083] 2.6 Verification: The test set accuracy is 95%, and the accuracy drops to <5% after key error simulation (adding random phase noise σ=0.05π), proving the reliability.

[0084] 3. Encrypted transmission and physical decryption

[0085] 3.1 Encryption: The sender inputs the image to be encrypted, which is encoded by a convolutional neural network. The encoded pixel values ​​are converted into a phase distribution and loaded onto a wavefront modulator (such as a phase-type liquid crystal spatial light modulator) for long-distance transmission. Even if the optical information of the encoded pattern is intercepted midway, its encoding complexity ensures that the encryption remains intact. The information contained in the captured coded optical information is almost impossible to decipher using traditional electronic computers.

[0086] 3.2 Decoding: After receiving the encoded optical information, the receiver uses a specific physical diffraction layer to decrypt the encoding. Specifically, the optical information propagates through diffraction to reach the phase modulation layer (physical decoder), loads the mask, performs dual-wavelength illumination, captures the intensity with the detector, calculates the differential decoding class label, and calculates the residual using a custom loss function to update the network parameters.

[0087] Unlike traditional electronic decoding methods, the decoding process of the optical decoding module in the hybrid optoelectronic encryption method relies on the physical propagation of the light field (such as diffraction and interference), which is an irreversible physical process. Even if an attacker intercepts the input phase code, the decoding process will still be affected. Furthermore, it is impossible to directly deduce the original information from the intensity distribution. This is because light field propagation involves complex wavefront transformations, and here the angular spectrum method is used to represent the optical diffraction propagation function. Specifically:

[0088]

[0089] In the above formula, Let λ represent the transfer function over a propagation distance of Z, and let λ represent the wavelength of light. , denoted by and , respectively, representing spatial frequency coordinates, and j represents the imaginary root unit.

[0090] Because the decoding process of the optical decoding module in the hybrid optoelectronic encryption method requires precise knowledge of the parameters of the optical components (such as the thickness distribution of the phase plate). And refractive index parameters), which are typically highly customized and complex, making them difficult to measure in physical implementation. In terms of security, attackers cannot directly crack the decoding process through mathematical modeling or reverse engineering; they must physically access the optical network itself, significantly increasing the difficulty of the attack. The decoding process consists entirely of optical diffraction propagation and physical diffraction modulation, without involving electronic storage or computation, thus naturally evading traditional electronic attack methods.

[0091] The diffraction layer in an optical decoding module is typically fabricated using 3D printing or micro / nano fabrication techniques, and its thickness distribution... It is obtained through deep learning optimization and possesses high complexity and uniqueness. Its phase modulation layer is trained and optimized through deep learning to form a specific optical field mapping relationship. Replicating such an optical element requires precise manufacturing processes and original design parameters, which are usually kept secret. Even if an attacker obtains the phase code, they cannot build an equivalent decoding network on their own because they lack the specific parameters and manufacturing capabilities of the optical element.

[0092] A hybrid optoelectronic encryption system utilizes the physical diffraction layer of an optical diffraction neural network as an embedded key to achieve encrypted transmission of digital information. This system is particularly suitable for the high-security communication requirements of the quantum computing era, and specifically includes:

[0093] Electronic Encoding Module 8: Based on a Convolutional Neural Network (CNN), this module compresses and encodes input digital information (such as images) into a low-resolution phase mask. It utilizes multi-layer convolution and pooling operations to extract features and outputs a mask with phase values ​​ranging from [0, 2π], reducing data transmission while preserving key information.

[0094] Optical decoding module 9 includes a spatial light modulator (SLM) and a physically embedded diffraction layer. The diffraction layer is physically realized by training the phase and amplitude distribution (e.g., 3D printing or nanofabrication; atomic-level complexity ensures quantum security), serving as an uncopyable physical key. The complex structure of the key (atomic-level precision) ensures that even if the digital code is obtained, it cannot be reconstructed. Unlike digital keys, this module uses the diffraction modulation layer of an optical neural network as the physical entity key. Its phase / amplitude distribution is realized by high-precision processing (e.g., photolithography), possessing irreversible physical properties such as thermal noise and material defects. Even if an attacker obtains the encoded data or simulation model, they cannot accurately copy the key, resisting quantum parallel search and reverse engineering attacks. The physical distribution of the key (e.g., mailing) is used to resist electronic theft and reverse engineering attacks, further enhancing security. Furthermore, this invention innovatively integrates electronic compression and optical propagation through the combined use of the electronic encoding module and the optical decoding module, improving transmission efficiency.

[0095] Dual-wavelength illumination source: produces two coherent light sources (such as...) Laser I1 The laser (II2) illuminates the coded phase mask, generating a wavelength-dependent diffraction pattern to enhance key diversity and anti-interference capability. This is achieved through the separate use of dual-wavelength illumination sources and the introduction of a differential detection mechanism. This innovatively utilizes wavelength-dependent diffraction to enhance key diversity and noise immunity. The dual-wavelength illumination sources are configured to include:

[0096] Laser I1 and Laser II2, which generate two coherent light sources;

[0097] Optical component I, used in conjunction with laser I to construct the upper branch optical path;

[0098] Optical component II, used in conjunction with laser II to construct the lower branch optical path;

[0099] The optical component I includes:

[0100] Collimating lens I3 that works in conjunction with the output side of laser I1;

[0101] The output light from collimating lens I3 is transmitted to beam splitter 5 of beam splitter 7;

[0102] The optical component II includes:

[0103] Collimating lens II4 that works in conjunction with the output side of laser II2;

[0104] It is positioned downstream of collimating lens II 4 to reflect the output of lens II to the reflector 6 of beam splitter 5.

[0105] A beam splitter 7 is used in conjunction with a dual-wavelength illumination source to split the main optical path into two paths, and the two output optical paths of the beam splitter 7 are respectively matched with the positions of the electronic encoding module 8 and the optical decoding module 9.

[0106] Positive and negative detectors 10: CCD or CMOS sensors are divided into positive and negative sub-regions, which respectively capture the intensity distribution of the diffracted light field to form a differential vector for information decoding.

[0107] The remote computer is mainly used to transmit the encoded mask to the electronic channel and is responsible for the physical distribution of optical keys. That is, the control unit of the remote computer coordinates the training and inference process of optical keys. Its workflow is mainly as follows: the electronic encoding module encodes the input information electronically (i.e., low-resolution phase mask), and then transmits it through diffraction after being illuminated by a dual-wavelength illumination source; the receiving end loads the mask to the optical decoding module, decodes it through physical key, and then captures it through positive and negative detectors, and then calculates differential decoding through the remote computer.

[0108] The benefits of hybrid optoelectronic encryption methods and systems are mainly reflected in the following aspects:

[0109] Security is significantly improved: physical keys cannot be electronically copied, and the complexity of quantum attacks increases exponentially; differential detection resists side-channel leakage.

[0110] Improved transmission efficiency: Encoding compression reduces bandwidth requirements, and optical parallel decoding accelerates processing (100 times faster than pure electronic).

[0111] Highly robust and compatible: noise and interference resistant; compatible with existing communication protocols, suitable for scenarios such as image recognition and data storage.

[0112] The above solution is merely an illustration of a preferred example and is not limited thereto. When implementing this invention, appropriate substitutions and / or modifications can be made according to the user's needs.

[0113] Although embodiments of the present invention have been disclosed above, they are not limited to the applications listed in the specification and embodiments. It can be applied to various fields suitable for the present invention. Other modifications can be readily made by those skilled in the art. Therefore, without departing from the general concept defined by the claims and their equivalents, the present invention is not limited to the specific details and examples shown and described herein.

Claims

1. A hybrid optoelectronic encryption method based on a physical diffraction layer, characterized in that, include: S1. The input digital information is compressed and encoded into a low-resolution phase mask through a convolutional neural network and loaded onto the wavefront modulator to complete the electronic encoding. S2. Use dual-wavelength illumination sources to illuminate the low-resolution phase mask on the wavefront modulator to generate a wavelength-dependent diffraction pattern. S3. The diffraction pattern is phase-modulated by the spatial light modulator of the diffraction layer to complete optical decoding. The diffraction layer has a physically embedded phase distribution. S4. The phase-modulated diffracted light propagates forward and forms an intensity distribution on the detector surface; S5. The remote computer reads the light intensity distribution transmitted by the positive and negative detectors and calculates the coded category through differential operation; In S3, the spatial light modulator has a resolution of 1920x1080; The diffraction layer is a photosensitive resin phase plate obtained by 3D printing, and the accuracy of the diffraction layer is 0.1μm and the thickness is 200μm. The value of each neuron on the diffraction layer ranges from 0 to 2π; The phase on the diffraction layer is obtained through training a diffraction neural network, and the network parameters are optimized using the joint loss function Loss as shown in the following formula during the training of the diffraction neural network: In the above formula, M Indicates the number of categories. g m The one-hot encoding representing the true class; if the sample belongs to class m, then... Otherwise, it equals zero. C m This represents the predicted probability after the model output has been processed by Softmax. β Indicates the weighting coefficient. N Indicates the number of samples. W Indicates the number of pixels or feature points. y i,w Indicates the first i The sample at the th The true value of each position, Indicates the first i The sample at the th Predicted values ​​for each location; In S3, the light field modulated by the spatial light modulator f ( x , y , z = z 1) Characterized by the following formula: In the above formula, This represents the axial transmission distance between the input layer and the modulation layer; * represents the convolution operation. f 1( x , y , z = z 0) indicates z = z The complex optical field at position 0, Represents the impulse response function. z 0 indicates the position of the input plane of the diffraction network. z 1 indicates the axial position of the first diffraction modulation layer. z This indicates the propagation distance of the diffraction layer.

2. The hybrid optoelectronic encryption method based on a physical diffraction layer as described in claim 1, characterized in that, In S1, the convolutional neural network is constructed using the PyTorch framework, and the low-resolution phase mask consists of 20×20 pixels, with each pixel having a value ranging from 0 to 2π.

3. The hybrid optoelectronic encryption method based on a physical diffraction layer as described in claim 1, characterized in that, In S4, the light intensity distribution I ( x , y It is characterized by the following formula: In the above formula, Indicates the first m -1 diffraction layer and the first m The axial distance between each diffraction layer.

4. The hybrid optoelectronic encryption method based on a physical diffraction layer as described in claim 3, characterized in that, In S4, the detection surface is divided into 10 sub-regions, and the size of each detection region is 10×10 pixels.

5. The hybrid optoelectronic encryption method based on a physical diffraction layer as described in claim 4, characterized in that, In S5, the light intensity distribution captured independently by the positive and negative detectors after the low-resolution phase mask is illuminated by a dual-wavelength illumination source is set as... and ; The remote computer then performs a difference calculation using the following formula to obtain the differentiated score. S Score : In the above formula, λa and λb are the different wavelengths corresponding to the dual-wavelength illumination source.

6. A hybrid optoelectronic encryption system, applied in the hybrid optoelectronic encryption method based on a physical diffraction layer as described in any one of claims 1-5, characterized in that, include: An electronic coding module that compresses and encodes input digital information into a low-resolution phase mask; An optical decoding module with a non-copyable physical key; A dual-wavelength illumination source that produces wavelength-dependent diffraction patterns by illuminating an coded phase mask. A beam splitter that works in conjunction with a dual-wavelength illumination source to split the main optical path into two paths, with the two output optical paths of the beam splitter respectively matching the locations of the electronic encoding module and the optical decoding module; Positive and negative detectors are positioned downstream of the optical decoding module to capture the intensity distribution of the diffracted light field. The electronic coding module is configured to employ a wavefront modulator, and the optical decoding module is configured to include a spatial light modulator and a diffraction layer that works in conjunction with it. Both the positive and negative detectors are configured to use CCD or CMOS sensors.

7. The hybrid optoelectronic encryption system as described in claim 6, characterized in that, The dual-wavelength illumination source is configured to include: Laser I and Laser II, which generate two coherent light sources; Optical component I, used in conjunction with laser I to construct the upper branch optical path; Optical component II, used in conjunction with laser II to construct the lower branch optical path; The optical component I includes: Collimating lens I that works in conjunction with the output side of laser I; The output light from collimating lens I is transmitted to the beam splitter of the beam splitter; The optical component II includes: Collimating lens II that works in conjunction with the output side of laser II; It is positioned downstream of collimating lens II to reflect the output of lens II to the reflector of the beam splitter.

Images