A spectral encryption method based on diffraction deep neural network

Abstract

The application discloses a spectrum encryption method based on a diffraction deep neural network, which comprises the following steps: training a diffraction deep neural network model, building a spectrum codebook, i.e. under the condition of spatial coherence or spatial incoherence, the diffraction deep neural network maps each input wavelength to an output pattern, i.e. a spectrum base; generating a spectrum code of encrypted information, i.e. according to the encrypted information and the built spectrum codebook, a corresponding spectrum code is generated; and decrypting the information by using the spectrum codebook and the spectrum code, i.e. according to the positive and negative spectrum codes, a plane wave with a corresponding spectrum amplitude is input, the corresponding positive and negative encrypted information is synthesized through the diffraction deep neural network, and the encrypted information is decrypted. The spectrum encryption method has low complexity and high security level.

Description

Technical Field

[0001] This invention relates to the field of information encryption technology, and in particular to a spectral encryption method based on a diffraction deep neural network. Background Technology

[0002] Photonic neural networks (PNNs) utilize the properties of light for computation, offering speed and efficiency surpassing traditional electronic neural networks. Leveraging the parallel processing capabilities and high bandwidth of optical systems, they significantly reduce energy consumption and latency for tasks such as data processing, machine learning, and real-time analysis. Among various photonic neural networks, diffractive deep neural networks (D²NNs) have received sustained and significant attention due to their simplicity, high efficiency, strong generalization ability, and scalability. D²NNs consist of multiple modulation layers that adjust the amplitude or phase of the light field and are optimized using deep learning methods to achieve arbitrary input-output mapping functions. This approach addresses various artificial intelligence challenges at the speed of light, providing an innovative computational solution for the post-Moore's Law era.

[0003] Information encryption is crucial for protecting information security and is widely used in telecommunications, data storage, and financial transactions. Traditional electronic encryption methods face numerous challenges, including complex algorithm design, limited security levels, long latency, and high computational requirements. In contrast, optical encryption offers significant advantages such as higher speed, lower energy consumption, parallel processing capabilities, and numerous tunable optical dimensions, effectively addressing the limitations of electronic encryption. Encryption using optical neural networks further enhances these advantages, providing more flexible design options and intelligent control by leveraging the inherent properties of light. Summary of the Invention

[0004] To address the challenges of complex algorithm design, limited security levels, long latency, and high computational requirements in traditional electrical encryption technologies, this invention provides a spectral encryption method using a diffractive deep neural network. This method leverages the spectral dimension of light for optical encryption. A spectral codebook is constructed, mapping each wavelength to a specific, self-designed output pattern under conditions of spatial coherence and incoherence. Subsequently, the spectral cipher corresponding to any encrypted information within the specific spectral codebook is calculated, thus perfectly encrypting the information into both the spectral codebook and the spectral cipher. This invention provides a reference for the application of diffractive deep neural networks in spectral interactions.

[0005] This invention discloses a spectral encryption method based on a diffraction deep neural network, comprising:

[0006] Train a diffraction deep neural network model to build a spectral codebook, that is, under the conditions of spatial coherence or spatial incoherence, the diffraction deep neural network maps each input wavelength to an output pattern, i.e., the spectral basis.

[0007] Generate the spectral cipher for the encrypted information, that is, generate the corresponding spectral cipher based on the encrypted information and the constructed spectral cipherbook;

[0008] Information is decrypted using a spectral codebook and spectral cryptography. Specifically, based on positive and negative spectral cryptography, plane waves with corresponding wavelength amplitudes are input, and corresponding positive and negative encrypted information is synthesized through a diffraction deep neural network to decrypt the encrypted information. Further, the input wavelength selection of the spectral codebook involves choosing the number and resolution of the input working wavelengths; for the encrypted information... ,need There are 1 input wavelength, and the interval between the wavelengths is arbitrary;

[0009] The output pattern design of the spectral cryptography, that is, the design of the output pattern corresponding to each wavelength, serves as the spectral basis; for the encrypted information ,design There are spectral bases, which are linearly independent and constitute a rank of . maximal linearly independent set .

[0010] Furthermore, the input to the diffraction deep neural network is a unit amplitude surface source with spatial coherence or incoherence, which contains multiple spectral bands. The light source amplitude of each spectral channel is uniform and constant; for spatially coherent input, the phase is a real constant, while for spatially incoherent input, the phase varies randomly at different spatial locations.

[0011] Furthermore, in the During the propagation process, the input light field of surface light sources of different wavelengths after passing through the rectangular aperture can be represented as follows:

[0012]

[0013] in, Indicates the input light field. Represents the function of rectangular holes. , Indicates the first Spatial location during secondary transmission Phase at; This represents the i-th wavelength channel;

[0014] For each wavelength channel The unit amplitude is uniformly distributed across the entire plane and remains constant over time. After transformation by the diffraction depth neural network, the light field distribution on the output plane is given by the following equation:

[0015]

[0016] in, Indicates the first The second transmission process Output light field intensity distribution of each channel This represents the modulation effect of the diffraction deep neural network, using pure phase modulation.

[0017] Furthermore, spectral bands The output intensity, i.e., the spectral basis of that spectral band, is expressed as:

[0018]

[0019] in, This represents the number of times the random phase is generated during the training of the diffraction deep neural network model. Let be the i-th spectral band.

[0020] Furthermore, two types of loss functions are used to co-train the diffraction deep neural network model:

[0021]

[0022] in, Represents the total loss function. It is a distributed loss function and the balance loss function The weighting coefficients, The goal is to minimize the mean square error between each spectral basis and its corresponding true value:

[0023]

[0024] in, It is a spectral base The corresponding true value, It is the maximum intensity value among all pixels on the output plane. This represents the total number of pixels for each spectral basis;

[0025] The goal is to maximize the intensity of each spectral basis. As close as possible:

[0026]

[0027] After the diffraction deep neural network model is trained, it serves as a spectral codebook.

[0028] Furthermore, after constructing the spectral codebook, each wavelength corresponds to a non-negative real-valued spectral basis. any vector Represented as:

[0029]

[0030] in, Represents the relationship with the spectral basis vectors Relevant weights;

[0031] Construct a matrix The spectral basis corresponding to each wavelength is taken as a column: ,but Represented as:

[0032]

[0033] in, , Indicates the first Spectral base;

[0034] matrix If it is reversible, then... There is a unique solution, namely ;

[0035] Weight vector Decomposed into positive weight vectors and negative weight vector ,in, and The components are defined as follows:

[0036]

[0037]

[0038] Decomposed into two components: and , Corresponding to the positive weight vector , Corresponding to the negative weight vector ;

[0039] and All are non-negative and formed by a non-negative linear combination of non-negative spectral bases. Spectral bands are simultaneously input into the input plane of the diffraction deep neural network, and the desired output intensity is synthesized in the output plane. and ;

[0040] Light intensity obtained through dual optical paths or two measurements and Obtain encrypted information ;

[0041] Positive branch and negative branches The corresponding spectral amplitude sequences, namely the positive spectral code and the negative spectral code, together constitute the spectral code.

[0042] Furthermore, Represented as:

[0043] .

[0044] Furthermore, output intensity and They are represented as follows:

[0045]

[0046] .

[0047] Furthermore, the positive spectral code and the negative spectral code are represented as follows: and .

[0048] Because of the adoption of the above technical solution, the present invention has the following advantages:

[0049] 1. This invention employs a diffraction-based deep neural network to convert information into a spectral codebook and spectral key using the spectral dimension of light, thereby achieving intelligent and high-level encryption of information.

[0050] 2. This invention uses a diffraction deep neural network for information decryption, achieving near-light-speed, low-energy-consumption, and high-quality information decryption.

[0051] 3. The method proposed in this invention has strong robustness and is suitable for a wide range of complex environments; it has strong flexibility and the codebook can be constructed in a flexible and varied manner; it has strong scalability and can be further combined with electrical modules to achieve higher encryption levels or simpler architectures. Attached Figure Description

[0052] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments recorded in the embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0053] Figure 1This is a flowchart illustrating a spectral encryption method based on a diffraction deep neural network according to an embodiment of the present invention.

[0054] Figure 2 This is a schematic diagram of the spectral codebook construction process according to an embodiment of the present invention;

[0055] Figure 3 This is a schematic diagram of the spectral key generation process and information reconstruction process in an embodiment of the present invention. Detailed Implementation

[0056] The present invention will be further described in conjunction with the accompanying drawings and embodiments. The described embodiments are only a part of the embodiments of the present invention, and not all of them. All other embodiments obtained by those skilled in the art should fall within the protection scope of the present invention.

[0057] See Figure 1 This invention provides an embodiment of a spectral encryption method based on a diffraction deep neural network, which includes the following steps:

[0058] S1, Training the diffraction neural network model to build a spectral codebook:

[0059] In this embodiment, the input and output of the spectral codebook include input wavelength selection and output pattern design;

[0060] Input wavelength selection, i.e., selecting the number of input working wavelengths and the resolution of the working wavelengths. For the encrypted information... ,need There are several input wavelengths, with arbitrary spacing between them; the output pattern is designed, that is, the output pattern corresponding to each wavelength is designed as a spectral basis. For the encrypted information... Design is required. There are spectral bases, which are linearly independent and constitute a rank of . maximal linearly independent set .

[0061] In this embodiment, the training of the diffraction neural network model in S1 and the construction of the spectral codebook are based on diffraction deep neural networks (D²NNs) to map each input wavelength to an output pattern (spectral basis) under the conditions of spatial coherence or spatial incoherence.

[0062] like Figure 2 As shown, the input is a single-amplitude surface source with spatial coherence or incoherence, containing multiple spectral bands ( Across the entire plane, the light source amplitude for each spectral channel is uniformly constant. For spatially coherent inputs, the phase is a real constant, while for spatially incoherent inputs, the phase varies randomly at different spatial locations. In the... During the propagation process, the input light field after passing through the rectangular aperture can be represented as:

[0063]

[0064] in, , Indicates the first Spatial location during secondary transmission The phase at that point. For each wavelength channel. The element amplitude is uniformly distributed across the entire plane and remains constant over time. After D... 2 After the NN transformation, the light field distribution on the output plane is given by the following equation:

[0065]

[0066] in, Indicates the first The second transmission process Output light field intensity distribution of each channel D represents 2 The modulation effect of the neural network employs pure phase modulation. For spatially coherent samples, the output remains constant over time. However, for spatially incoherent samples, the phase changes over time, causing a corresponding change in the output. During training, to simulate the spatial incoherence effect, the output intensity distribution is calculated as the average of the output intensity distributions across multiple forward propagations based on the principle of intensity superposition. Since the modulation layer remains constant over time, the inputs in these forward propagations pass through the same modulation function. Let... This indicates the number of times the random phase is generated during training. The output intensity of a spectral band, i.e., the spectral basis of that band, can be expressed as:

[0067]

[0068] For ease of subsequent explanation, when considered as a vector, Output light intensity of spectral band Represented as .

[0069] To achieve a specified, self-designed spectral basis while balancing the intensity differences between spectral basis bases, two types of loss functions are used for co-training:

[0070]

[0071] in, Represents the total loss function. It is a distributed loss function and the balance loss function The weighting coefficient is set to . The goal is to minimize the mean square error between each spectral basis and its corresponding true value:

[0072]

[0073] in, It is the corresponding true value. This refers to the maximum intensity value among all pixels on the output plane. This represents the total number of pixels for each spectral basis. The goal is to maximize the intensity of each spectral basis. As close as possible:

[0074]

[0075] After training, D 2 NN acts as a spectral codebook, mapping each input spectral band to a unique output, constructing a codebook containing... indivual Spectral cryptography of linearly independent vectors of dimension.

[0076] S2, generate the spectral cipher for the encrypted information:

[0077] In this embodiment, the spectral cryptography for generating encrypted information in S2 is based on the encrypted information. The corresponding spectral code is generated from the constructed spectral codebook.

[0078] After constructing the spectral codebook, each wavelength corresponds to a non-negative real-valued spectral basis. These bases are linearly independent and together span a real space. For ease of explanation, when the spectral basis is used as the output surface, the output light intensity is not bolded, and is represented as... When the spectral basis is used as a vector, it is represented in bold, i.e., as... , , and Similarly;

[0079] Any vector All of these can be represented by these baseline characteristics:

[0080]

[0081] in, Indicates the relationship with spectral base Relevant weights. Construct a matrix. Its spectral basis is used as an example: ,but It can be represented as:

[0082]

[0083] in, Since these spectral bases are linearly independent, the matrix... If it is reversible, then... There is a unique solution, namely Therefore, the weight vector It can be solved directly without the need for training.

[0084] Weight vector It can be decomposed into a positive weight vector and negative weight vector ,in and The components are defined as follows:

[0085]

[0086]

[0087] Similarly, It can also be decomposed into two components: , corresponding to the positive weight vector ,and , corresponding to the negative weight vector .therefore, It can also be expressed as:

[0088]

[0089] like Figure 3 As shown, due to and All are nonnegative and formed by nonnegative linear combinations of nonnegative spectral bases, therefore they can be found in D. 2 The neural network (NN) simultaneously inputs spectral bands with specific amplitudes onto its input plane, thereby synthesizing the desired output intensity onto its output plane. and :

[0090]

[0091]

[0092] Light intensity obtained through dual optical paths or two measurements and Encrypted information can be obtained by performing a simple electronic subtraction. The spectral amplitude sequences corresponding to the positive and negative branches, i.e., the positive spectral code and the negative spectral code, can be represented as follows: and Together, these constitute the spectral code. Therefore, a diffraction deep neural network is used to achieve spectral encryption of information, where the information is hidden as a spectral codebook and a spectral cipher. Only by simultaneously obtaining both the spectral codebook and the spectral cipher can the encrypted information be reconstructed.

[0093] S3, decrypt the information using a spectral codebook and spectral cryptography:

[0094] In this embodiment, in S3, the spectral codebook and spectral code are used to decrypt the information. That is, according to the positive and negative spectral codes, a plane wave with the corresponding wavelength amplitude is input, and the corresponding positive and negative encrypted information is synthesized through a diffraction deep neural network, thereby decrypting the encrypted information by subtraction.

[0095] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the protection scope of the claims of the present invention.

Claims

1. A spectral encryption method based on a diffraction deep neural network, characterized in that, include: Train a diffraction deep neural network model to build a spectral codebook, that is, under the conditions of spatial coherence or spatial incoherence, the diffraction deep neural network maps each input wavelength to an output pattern, i.e., the spectral basis. Generate the spectral cipher for the encrypted information, that is, generate the corresponding spectral cipher based on the encrypted information and the constructed spectral cipherbook; Information is decrypted using a spectral codebook and spectral cryptography. Specifically, based on positive and negative spectral cryptography, plane waves with corresponding wavelength amplitudes are input, and corresponding positive and negative encrypted information is synthesized through a diffraction deep neural network to decrypt the encrypted information. The input wavelength selection for the spectral codebook involves choosing the number and resolution of the input working wavelengths. For the encrypted information... ,need There are 1 input wavelength, and the interval between the wavelengths is arbitrary; The output pattern design of the spectral cryptography, that is, the design of the output pattern corresponding to each wavelength, serves as the spectral basis; for the encrypted information ,design There are spectral bases, which are linearly independent and constitute a rank of . maximal linearly independent set ; The input to the diffraction deep neural network is a single-amplitude surface source with spatial coherence or incoherence, which contains multiple spectral bands. The light source amplitude of each spectral channel is uniform and constant; for spatially coherent input, the phase is a real constant, while for spatially incoherent input, the phase varies randomly at different spatial locations. Two types of loss functions are used to co-train the diffraction deep neural network model: in, Represents the total loss function. It is a distributed loss function and the balance loss function The weighting coefficients, The goal is to minimize the mean square error between each spectral basis and its corresponding true value: in, It is a spectral base The corresponding true value, It is the maximum intensity value among all pixels on the output plane. This represents the total number of pixels for each spectral basis; The goal is to maximize the intensity of each spectral basis. As close as possible: After the diffraction deep neural network model is trained, it serves as a spectral codebook.

2. The method according to claim 1, characterized in that, In the During the propagation process, the input light field of different wavelength surface light sources after passing through the rectangular aperture can be represented as follows: in, Indicates the input light field. Represents the function of rectangular holes. , Indicates the first Spatial location during secondary transmission Phase at; This represents the i-th wavelength channel; For each wavelength channel The unit amplitude is uniformly distributed across the entire plane and remains constant over time. After transformation by the diffraction depth neural network, the light field distribution on the output plane is given by the following equation: in, Indicates the first The second transmission process Output light field intensity distribution of each channel This represents the modulation effect of the diffraction deep neural network, using pure phase modulation.

3. The method according to claim 2, characterized in that, spectral bands The output intensity, i.e., the spectral basis of that spectral band, is expressed as: in, This represents the number of times the random phase is generated during the training of the diffraction deep neural network model. Let be the i-th spectral band.

4. The method according to claim 1, characterized in that, After constructing the spectral codebook, each wavelength corresponds to a non-negative real-valued spectral basis. any vector Represented as: in, Indicates the relationship with spectral base Relevant weights; Construct a matrix The spectral basis corresponding to each wavelength is used as the column vector of the matrix: ,but Represented as: in, ; matrix If it is reversible, then... There is a unique solution, namely ; Weight vector Decomposed into positive weight vectors and negative weight vector ,in, and The components are defined as follows: Decomposed into two components: and , Corresponding to the positive weight vector , Corresponding to the negative weight vector ; and All are non-negative and formed by a non-negative linear combination of non-negative spectral bases. Spectral bands are simultaneously input into the input plane of the diffraction deep neural network, and the desired output intensity is synthesized in the output plane. and ; Light intensity obtained through dual optical paths or two measurements and Obtain encrypted information ; Positive branch and negative branches The corresponding spectral amplitude sequences, namely the positive spectral code and the negative spectral code, together constitute the spectral code.

5. The method according to claim 4, characterized in that, Represented as: 。 6. The method according to claim 4 or 5, characterized in that, Output strength and They are represented as follows: 。 7. The method according to claim 4, characterized in that, The positive and negative spectral codes are represented as follows: and .

Images