Image classification system and method based on asymmetric pumping

Abstract

This application belongs to the field of optical signal processing technology and discloses an image classification system and method based on asymmetric pumping, including an image acquisition module for acquiring a handwritten digital image and sending it to a nonlinear XOR module; a nonlinear XOR module for generating multiple serial signals based on the received handwritten digital image; delaying the serial signals to obtain a first signal and a second signal; sending the first signal and the second signal to a BIC resonant cavity to obtain a coupled XOR signal and send it to a linear classification module; the first signal is delayed by a preset time length compared to the second signal, and the difference between the power of the first signal and the power of the second signal is equal to a preset threshold; a linear classification module for classifying the handwritten digital image based on each coupled XOR signal and outputting a digital classification label. This application implements all-optical nonlinear transformation in an optical neural network, significantly improving data separability.

Description

Technical Field

[0001] The present application relates to the technical field of optical signal processing, and in particular to an image classification system and method based on asymmetric pumping. Background Art

[0002] Optical neural network technology has attracted much attention in recent years, with researchers gradually developing a series of network layers based on optical waveguides, lasers, and other components to circumvent the speed limitations of traditional electronic neural networks. Typical optical neural network architectures currently include silicon photonic neural networks based on on-chip Mach-Zehnder interferometers, diffraction-based neural networks based on 3D-printed phase plates, and hybrid optical neural networks combined with phase change materials. Most of these rely on linear optical responses such as interference, diffraction, and wavelength division multiplexing.

[0003] However, existing optical neural networks fail to introduce nonlinear terms, and the operational relationships between network layers are limited to linear operations. Although multi-layer diffraction networks can exhibit certain depth features, the constructed models can only process linear input-output relationships, which significantly reduces the generalization ability of diffraction neural networks and makes them difficult to apply to more complex machine learning tasks such as image classification. Summary of the Invention

[0004] The present application provides an image classification system and method based on asymmetric pumping, which can realize all-optical nonlinear transformation in optical neural networks, significantly improving data separability. In image classification tasks, the classification accuracy of optical neural networks is much higher than that of traditional linear classifiers, promoting further development in the field of optical computing.

[0005] In a first aspect, an embodiment of the present application provides an image classification system based on asymmetric pumping, comprising:

[0006] An image acquisition module is used to acquire a handwritten digital image and send it to a nonlinear XOR module;

[0007] a nonlinear XOR module configured to generate multiple serial signals based on a received handwritten digital image; delay the serial signals to obtain a first signal and a second signal; and transmit the first signal and the second signal to a BIC resonant cavity to obtain a coupled XOR signal, which is then transmitted to a linear classification module; wherein the first signal is delayed by a preset time period relative to the second signal, and the difference between the power of the first signal and the power of the second signal is equal to a preset threshold;

[0008] The linear classification module is used to receive the coupled XOR signals corresponding to the serial signals, classify the handwritten digital images according to the coupled XOR signals, and output digital classification labels.

[0009] Furthermore, the nonlinear XOR module includes a binarization unit, a signal generation unit, a delay unit, and a BIC resonant cavity;

[0010] The binarization unit is used to binarize the received handwritten digital image to obtain a digital grayscale image;

[0011] The signal generating unit is used to randomly select a pixel pair in the digital grayscale image and generate a corresponding serial signal according to the pixel values ​​of two pixels in the pixel pair;

[0012] The delay unit is used to delay the serial signal to obtain a first signal and a second signal, and send the first signal and the second signal to the coupled XOR unit;

[0013] The BIC resonant cavity is used to couple the first signal and the second signal to obtain a coupled XOR signal.

[0014] Furthermore, the BIC resonant cavity adopts an etching-free BIC quasi-two-dimensional perovskite photonic crystal structure.

[0015] Furthermore, the nonlinear XOR module further includes a frequency multiplication unit; the frequency multiplication unit is arranged after the signal generation unit;

[0016] The frequency multiplication unit is used to perform frequency multiplication processing on the serial signal and send the signal to the delay unit.

[0017] Furthermore, the linear classification module is a fully connected linear neural network.

[0018] Furthermore, the first signal is delayed by 30 picoseconds compared to the second signal.

[0019] Furthermore, the difference between the power of the first signal and the power of the second signal is equal to 0.3Pth.

[0020] Furthermore, the frequency doubling unit is a β-phase barium metaborate crystal.

[0021] In a second aspect, an embodiment of the present application provides an image classification method based on asymmetric pumping, comprising:

[0022] The image acquisition module acquires the handwritten digital image and sends it to the nonlinear XOR module;

[0023] The nonlinear XOR module generates multiple serial signals based on the received handwritten digital image; delays the serial signals to obtain a first signal and a second signal; transmits the first signal and the second signal to the BIC resonant cavity to obtain a coupled XOR signal, which is then transmitted to the linear classification module; wherein the first signal is delayed by a preset time length relative to the second signal, and the difference between the power of the first signal and the power of the second signal is equal to a preset threshold;

[0024] The linear classification module receives the coupled XOR signals corresponding to the serial signals, classifies the handwritten digital images according to the coupled XOR signals, and outputs digital classification labels.

[0025] Furthermore, the nonlinear XOR module generates a plurality of serial signals according to the received handwritten digital image, including:

[0026] Binarize the received handwritten digital image to obtain a digital grayscale image;

[0027] Randomly select pixel pairs in the digital grayscale image;

[0028] Generate corresponding serial signals according to the pixel values ​​of two pixels in a pixel pair.

[0029] Furthermore, the method further comprises:

[0030] Before the delay, the serial signal is frequency-multiplied using a β-phase barium metaborate crystal.

[0031] In summary, compared with the prior art, the technical solutions provided by the embodiments of the present application have at least the following beneficial effects:

[0032] An embodiment of the present application provides an image classification system based on asymmetric pumping. The system delays the serial signal generated by a handwritten digital image into a first signal and a second signal with different delay times, and then utilizes the coupling effect of a BIC resonant cavity to nonlinearly map the first signal and the second signal to a high-dimensional space, thereby achieving all-optical nonlinear processing. At the same time, due to the delay and power gap between the first signal and the second signal, an all-optical exclusive-OR (XOR) logic gate can be implemented during mapping to obtain coupled XOR signals, significantly improving data separability. Linear classification is then performed based on each coupled XOR signal, with an accuracy far greater than that of traditional neural network classification methods, demonstrating the application potential of optical neural networks in the field of image classification. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 A flowchart of an image classification system based on asymmetric pumping is provided in accordance with one embodiment of the present application.

[0034] Figure 2 A schematic diagram of an optical neural network computing architecture provided for one embodiment of the present application.

[0035] Figure 3 This is a diagram of the internal structure of a delay unit provided in one embodiment of the present application.

[0036] Figure 4 Schematic diagram of the incident light of a BIC resonant cavity provided in one embodiment of the present application.

[0037] Figure 5A schematic diagram of implementing classification based on XOR logic gate operations is provided for one embodiment of the present application.

[0038] Figure 6 A schematic diagram of signals performing XOR coupling under different pump powers is provided for one embodiment of the present application. DETAILED DESCRIPTION

[0039] The technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments.

[0040] Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making any creative work shall fall within the scope of protection of this application.

[0041] See Figure 1 , an embodiment of the present application provides an image classification system based on asymmetric pumping, comprising:

[0042] The image acquisition module is used to acquire the handwritten digital image and send it to the nonlinear XOR module.

[0043] Among them, MNIST images can be used as handwritten digital images, and this image set can be called using pytorch.

[0044] The nonlinear XOR module is used to generate multiple serial signals based on the received handwritten digital image; delay the serial signals to obtain a first signal and a second signal; send the first signal and the second signal to the BIC resonant cavity to obtain a coupled XOR signal and send it to the linear classification module; wherein the first signal is delayed by a preset time length compared to the second signal, and the difference between the power of the first signal and the power of the second signal is equal to a preset threshold.

[0045] The BIC resonant cavity can adopt an etching-free BIC quasi-two-dimensional perovskite photonic crystal structure. The delay of the first signal and the second signal is performed after the serial signal is separated, and the power control is achieved when the serial signal is generated.

[0046] It can be understood that the problem to be solved by this application is to realize nonlinear transformation in optical neural networks. The core of optical nonlinearity is to realize the transformation of light regulation from low dimension to high dimension, and the fundamental of light regulation lies in the interaction between light and matter. Therefore, the key to optical nonlinearity is to solve the interaction between light and matter.

[0047] The BIC (Bound States in the Continuum) used in this application is a periodic photonic crystal structure that suppresses radiation loss through interference effects. When the structure supports symmetry in a specific direction, modes with the same symmetry degenerate at the Γ point (the symmetry point with wave vector k = 0), causing radiation fields from different propagation directions to cancel each other, thus forming a radiation-free bound state. This process not only suppresses energy leakage but also localizes the modes, achieving an extremely high quality factor (Q).

[0048] Therefore, the BIC structure has two significant characteristics: a theoretically near-infinite Q factor and strong field confinement, which significantly enhances the interaction between light and matter. By using a BIC photonic crystal as a resonant cavity and combining it with etching-free micro-nanotechnology, the interaction between light and gain material can be further enhanced.

[0049] The dipole-dipole interaction in the BIC resonant cavity can be analyzed by dipole approximation. During the optical excitation process, the dipole collective resonance excited by the pump light is usually limited to a small area near the excitation area. In order to establish the correlation between the dipole collectives, this application uses two pump lasers (first signal and second signal) for excitation. When the two pump beams excite the sample, a coupling channel is formed between the BIC dipole resonances, realizing the dipole-dipole correlation coupling between different dipole collectives. In non-Hermitian systems, the relationship between gain and loss can be analyzed by studying PT symmetry (Parity-Timesymmetry). When the BIC metasurface is pumped by two spatially separated laser beams of finite size, a coupled PT symmetric system can be formed between the discrete quasi-BIC nodes. The coupled mode equation in the PT symmetric system can be used to model and accurately describe the transient coupling process of dipole oscillations in the BIC resonant cavity under femtosecond pulse laser pumping:

[0050]

[0051]

[0052] The Hamiltonian is as follows:

[0053]

[0054] in, and represent two asymmetric pump laser beams, represents the natural frequency, the gain term and and loss items and affects the amplitude evolution of each mode separately, while the coupling term and This reflects the interaction between the two modes.

[0055] The linear classification module is used to receive the coupled XOR signals corresponding to each serial signal, classify the handwritten digital image according to each coupled XOR signal, and output a digital classification label. The linear classification module is a fully connected linear neural network.

[0056] Specifically, a simple fully connected linear neural network model can be trained using the PyTorch framework. During the training process, the model performs forward propagation, loss calculation, backpropagation, and parameter update in sequence.

[0057] In the model architecture, the number of neurons in the input layer equals the total number of pixels in the image, and the output layer corresponds to different classification labels. In the first linear layer, self.fc1 = nn.Linear(input_size, num_classes) , where input_size is the size of the input image. If the input handwritten digit image contains only 10 digits, 0-9, then num_classes will have 10 output labels. The hyperparameters EPOCH, BATCH_SIZE, and learning rate are also set.

[0058] See Figure 2 It can be understood that both the serial signal and the coupled XOR signal are optical signals. After being processed by the nonlinear XOR module, the coupled XOR signals output can be regarded as images containing nonlinear information after the XOR operation. They are input into the trained linear classification module to complete the classification task of the numbers 0-9 in the image.

[0059] The above-mentioned embodiment provides an image classification system based on asymmetric pumping. By delaying the serial signal generated by a handwritten digital image into two first and second signals with different delay times, and then utilizing the coupling effect of the BIC resonant cavity to nonlinearly map the first and second signals to a high-dimensional space, all-optical nonlinear processing is achieved. At the same time, due to the delay and power gap between the first and second signals, an all-optical exclusive-OR (XOR) logic gate can be implemented during mapping to obtain coupled XOR signals, significantly improving data separability. Linear classification is then performed based on each coupled XOR signal, with an accuracy far exceeding that of traditional neural network classification methods, demonstrating the application potential of optical neural networks in the field of image classification.

[0060] In addition, some current optical nonlinear research focuses on intensity-type nonlinear transformations, but this usually relies on devices such as semiconductor optical amplifiers and photodetectors to achieve the nonlinear relationship between input and output, or is achieved through the nonlinear absorption of materials. However, intensity-type nonlinear transformations are often accompanied by optical power loss, which also limits the application of optical neural networks in complex learning tasks such as image classification. This application combines the coupling dynamics between quasi-BIC microlasers with optical signal processing, and proposes a new idea of ​​using optical loss as a control degree of freedom, which solves the power loss problem in nonlinear transformations and provides a new technical path for the practical application of all-optical nonlinear signal processing.

[0061] In some embodiments, the nonlinear XOR module includes a binarization unit, a signal generation unit, a delay unit, and a BIC resonant cavity; the binarization unit is used to binarize the received handwritten digital image to obtain a digital grayscale image.

[0062] The signal generation unit is used to randomly select pixel pairs in the digital grayscale image and generate corresponding serial signals based on the pixel values ​​of the two pixels in the pixel pairs. Since the handwritten digital image has been binarized, the pixel values ​​in the digital grayscale image are only two, 0 and 1. The signals corresponding to the two pixel values ​​are the first signal and the second signal, respectively. Together, they form the serial signal. The difference between the power of the first signal and the power of the second signal is equal to 0.3Pth, where Pth = 250 nJ / cm².

[0063] Specifically, if the pixel value is "0", the corresponding signal pump power is 0.8 Pth, and if the pixel value is "1", the corresponding signal pump power is 1.2 Pth. That is, this application uses the signal pump power to represent different pixel values.

[0064] The delay unit is used to delay the serial signal to obtain a first signal and a second signal, and send the first signal and the second signal to the coupled XOR unit, wherein the first signal is delayed by 30 picoseconds compared to the second signal.

[0065] Furthermore, the nonlinear XOR module also includes a frequency multiplication unit; the frequency multiplication unit is arranged after the signal generation unit; the frequency multiplication unit is used to perform frequency multiplication processing on the serial signal and send it to the delay unit.

[0066] The frequency doubling unit may specifically be a β-phase barium borate crystal (β-BaB2O4), referred to as BBO.

[0067] Specifically, BBO is used to frequency-double 800nm ​​light to 400nm, meaning that all subsequent processes use 400nm light.

[0068] See Figure 3The delay unit can use a beam splitter prism and multiple reflectors to separate and delay the first signal and the second signal in the serial signal. After the serial signal is separated into the first signal and the second signal by the beam splitter prism, they enter different reflection paths respectively and undergo different numbers of reflections to achieve different signal delay times.

[0069] The BIC resonant cavity is used to couple the first signal and the second signal to obtain a coupled XOR signal.

[0070] Specifically, see Figure 4 、 Figure 5 and Figure 6 , first mark the first signal and the second signal as A and B respectively, and optimize the incident light spot to a Gaussian distribution with a diameter of about 5μm. When the preset time length of the second signal delay compared to the first signal is 30 picoseconds and the power difference is 0.3Pth, the constructed all-optical XOR logic gate is the most reasonable and accurate. Because in the coupled mode theory, the coupling coefficient of the dual resonant cavity depends on the relative distance and relative time between the cavities. By adjusting the relative time delay of the two beams of pump light, the coupling of the dual-cavity system can be effectively controlled. In the image classification system of the present application, the delay unit converts the time delay into spatial resolution, adjusts the relative intensity of the pump light, and realizes the control of the system gain and asymmetric pump coupling.

[0071] Specifically, a pixel pair in the image is randomly selected, and an exclusive-OR (XOR) operation is performed based on its value to generate a coupled XOR signal; if both pixel pairs are 0, the output is [0, 0, 0]; if one is 0 and the other is 1, the output is [0, 1, 1]; if one is 1 and the other is 0, the output is [1, 0, 1]; when both are 1, the output is [1, 1, 0], that is, the output coupled XOR signal exists in the form of a vector. This logic gate can effectively classify on a two-dimensional plane by introducing nonlinear features, thereby separating the output results of "0" and "1". Combined with the trained linear classification module, the construction of an all-optical neural network for image classification can be realized.

[0072] In some embodiments, the nonlinear XOR module further includes a white light generating unit; the white light generating unit is configured to generate a white light signal and add the white light signal to the first signal and the second signal before being sent to the BIC resonant cavity.

[0073] The above embodiment adds a white light generating unit before the BIC resonant cavity, which enables researchers to observe the transient optical response of the BIC sample at different time points, and helps to understand the transient characteristics of the BIC sample.

[0074] Another embodiment of the present application provides an image classification method based on asymmetric pumping, comprising:

[0075] Step S1: The image acquisition module acquires a handwritten digital image and sends it to a nonlinear XOR module.

[0076] In step S2, a nonlinear XOR module generates multiple serial signals based on the received handwritten digital image; delays the serial signals to obtain a first signal and a second signal; and sends the first signal and the second signal to a BIC resonant cavity to obtain a coupled XOR signal and sends it to a linear classification module; wherein the first signal is delayed by a preset time length compared to the second signal, and the difference between the power of the first signal and the power of the second signal is equal to a preset threshold.

[0077] Step S3: The linear classification module receives the coupled XOR signals corresponding to the serial signals, classifies the handwritten digital image according to the coupled XOR signals, and outputs a digital classification label.

[0078] Furthermore, the nonlinear XOR module generates a plurality of serial signals according to the received handwritten digital image, including:

[0079] Step S21 : binarize the received handwritten digital image to obtain a digital grayscale image.

[0080] Step S22: randomly selecting pixel pairs in the digital grayscale image.

[0081] Step S23 : generating a corresponding serial signal according to the pixel values ​​of the two pixels in the pixel pair.

[0082] Furthermore, the method further comprises:

[0083] Before the delay, the serial signal is frequency-multiplied using a β-phase barium metaborate crystal.

[0084] Furthermore, the method further comprises:

[0085] Before being sent to the BIC resonant cavity, a white light signal is added to the first signal and the second signal.

[0086] For the specific limitations of the image classification method based on asymmetric pumping provided in this embodiment, please refer to the above embodiment of an image classification system based on asymmetric pumping, which will not be repeated here.

[0087] The technical features of the above embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0088] The above-described embodiments merely represent several implementation methods of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that a person skilled in the art could make various modifications and improvements without departing from the spirit of the present application, all of which fall within the scope of protection of the present application. Therefore, the scope of protection of the present patent application shall be determined by the appended claims.

Claims

1. An image classification system based on asymmetric pumping, characterized in that: include: An image acquisition module is used to acquire a handwritten digital image and send it to a nonlinear XOR module; a nonlinear XOR module, configured to generate a plurality of serial signals according to the received handwritten digital image; Delaying the serial signal to obtain a first signal and a second signal; sending the first signal and the second signal to a BIC resonant cavity to obtain a coupled XOR signal and sending the coupled XOR signal to a linear classification module; wherein the first signal is delayed by a preset time length compared to the second signal, and the difference between the power of the first signal and the power of the second signal is equal to a preset threshold; The linear classification module is used to receive the coupled XOR signals corresponding to the serial signals, classify the handwritten digital image according to the coupled XOR signals, and output a digital classification label.

2. The image classification system based on asymmetric pumping according to claim 1, characterized in that: The nonlinear XOR module includes a binarization unit, a signal generation unit, a delay unit and a BIC resonant cavity; The binarization unit is used to binarize the received handwritten digital image to obtain a digital grayscale image; The signal generating unit is configured to randomly select a pixel pair in the digital grayscale image and generate the corresponding serial signal according to the pixel values ​​of two pixels in the pixel pair; The delay unit is used to delay the serial signal to obtain the first signal and the second signal, and send the first signal and the second signal to the coupled XOR unit; The BIC resonant cavity is used to couple the first signal and the second signal to obtain a coupled XOR signal.

3. The image classification system based on asymmetric pumping according to claim 1, characterized in that: The BIC resonant cavity adopts an etching-free BIC quasi-two-dimensional perovskite photonic crystal structure.

4. The image classification system based on asymmetric pumping according to claim 2, characterized in that: The nonlinear XOR module further includes a frequency multiplication unit; the frequency multiplication unit is arranged after the signal generation unit; The frequency multiplication unit is used to perform frequency multiplication processing on the serial signal and send the result to the delay unit.

5. The image classification system based on asymmetric pumping according to claim 1, characterized in that: The linear classification module is a fully connected linear neural network.

6. The image classification system based on asymmetric pumping according to claim 1, characterized in that: The first signal is delayed by 30 picoseconds compared to the second signal.

7. The image classification system based on asymmetric pumping according to claim 1, characterized in that: A difference between the power of the first signal and the power of the second signal is equal to 0.3Pth.

8. The image classification system based on asymmetric pumping according to claim 4, characterized in that: The frequency doubling unit is a β-phase barium metaborate crystal.

9. An image classification method based on asymmetric pumping, characterized in that: include: The image acquisition module acquires the handwritten digital image and sends it to the nonlinear XOR module; The nonlinear XOR module generates a plurality of serial signals according to the received handwritten digital image; Delaying the serial signal to obtain a first signal and a second signal; sending the first signal and the second signal to a BIC resonant cavity to obtain a coupled XOR signal and sending the coupled XOR signal to a linear classification module; wherein the first signal is delayed by a preset time length compared to the second signal, and the difference between the power of the first signal and the power of the second signal is equal to a preset threshold; The linear classification module receives the coupled XOR signals corresponding to the serial signals, classifies the handwritten digital image according to the coupled XOR signals, and outputs a digital classification label.

10. The image classification method based on asymmetric pumping according to claim 9, characterized in that: The nonlinear XOR module generates a plurality of serial signals according to the received handwritten digital image, including: Binarizing the received handwritten digital image to obtain a digital grayscale image; randomly selecting pixel pairs in the digital grayscale image; The corresponding serial signal is generated according to the pixel values ​​of two pixels in the pixel pair.

Images