Optical computing system and construction method thereof

Abstract

The invention relates to an optical computing system and a construction method thereof, and the system comprises an optical compression module which is used for compressing redundant and invalid information in a scene after an optical signal of a natural scene enters an optical system, and obtaining compressed effective scene information which takes a target as a main part; the optical convolution module is used for processing the effective scene information to obtain a multi-channel output optical feature map; and the optical full-connection module is used for performing optical diffraction full-connection calculation according to the multi-channel output optical characteristic pattern and outputting a target classification result. According to the invention, all-optical calculation of optical remote sensing is realized through the optical compression module, the optical convolution module and the optical full-connection module, optical signals do not need to be converted into electric signals, the calculation rate, efficiency and other aspects are far higher than those of existing electronic calculation, and the processing speed and efficiency of the system are greatly improved; and the method is more suitable for optical remote sensing application with higher real-time requirement.

Description

Technical Field

[0001] The present invention relates to the field of optical computing technology, and in particular to an optical computing system and a construction method thereof. Background Art

[0002] Optical computing is the use of photons as physical carriers to express mathematical processes, which can be divided into digital optical computing and analog optical computing. The research on digital optical computing can be traced back to the 1950s. Its main idea is to design photonic transistors to replace electronic transistors based on Boolean logic to realize optical logic gate operations. Analog optical computing is the operation and processing of optical analog signals in linear optical systems, mainly covering two operations: analog signal operation and optical matrix calculation. Analog signal operation mainly uses spectrum shaping or other methods to process the frequency domain or time domain of optical analog signals to achieve specific signal transformations, such as optical differentiators, optical integrators, optical differential equation solving, Hilbert transform, Fourier transform, and optical arbitrary waveform generation; optical matrix calculation is to realize discrete parallel matrix multiplication calculations in the optical domain, such as linear equation solving, least squares problems, algebraic eigenvalue problems, and spatial Fourier transforms; in the linear part of convolutional neural networks, it is widely used in image recognition, speech recognition, intelligent driving and other fields. Analog optical computing can be realized through spatial domain simulation and integrated systems respectively.

[0003] For decades, researchers have been committed to finding effective ways to control light and use it for information transmission and processing. Early optical computing was mainly based on systems built on spatial optical computing methods. In 1964, Vander Lugt realized the first spatial optical filter with the support of the US Army Laboratory and applied it to point target extraction under complex backgrounds. After that, researchers explored a series of optical logic devices and model devices and proposed a large number of optical computing technologies, including optical pattern recognition, logic optical computing, optical neural networks, optical interconnection, optical holographic storage, etc.

[0004] However, due to the lack of clear application requirements and application scenarios for optical computing and the lack of optical hardware systems that match optical computing technology, the development of optical computing technology is relatively slow. In recent years, with the development of artificial intelligence technology, people's demand for optical computing technology has increased rapidly. The reasons are: First, with the gradual failure of Moore's Law and the continuous increase in the power consumption and speed requirements of computing systems in the era of big data, the high-speed, parallel and low-power characteristics of optical computing technology are increasingly valued by people. Second, the parallel computing characteristics of optical computing technology, as well as the development of algorithms and hardware architectures such as optical neural networks, provide the most potential solutions for the computing power needs of artificial intelligence technologies such as image recognition, speech recognition, and virtual reality. Especially in 2017, after MIT researchers demonstrated the first on-chip integrated optical neural network chip, optical computing technology entered a period of vigorous development, and a large number of artificial intelligence optical computing studies based on on-chip and spatial light were carried out.

[0005] Although optical computing has made great progress and has been verified in laboratory environments to be more energy efficient, cost-effective, and efficient than electrical computing systems in certain scenarios, laying the foundation for the practical application of optical computing, there is currently no effective optical computing solution for natural scenarios. Summary of the invention

[0006] In order to solve the technical problems existing in the above-mentioned prior art, the purpose of the present invention is to provide an optical computing system and a construction method thereof, which realizes all-optical computing by constructing an optical compression module, an optical convolution module and an optical fully-connected module, and realizes rapid detection and classification of targets and rapid detection and classification of targets on resource-constrained platforms.

[0007] To achieve the above-mentioned object, the present invention provides an optical computing system, comprising:

[0008] An optical compression module is used to compress redundant and invalid information in the scene after the light signal of the natural scene is incident on the optical system, so as to obtain compressed valid scene information mainly focusing on the target;

[0009] Optical convolution module, used to process effective scene information and obtain multi-channel output optical feature map;

[0010] The optical fully connected module is used to perform optical diffraction fully connected calculations based on the multi-channel output optical feature map and output the target classification results.

[0011] According to a technical solution of the present invention, the optical compression module compresses the optical signal through spatial coding to obtain compressed scene information mainly based on the target;

[0012] The optical compression module includes an optical compression encoding plate composed of a metasurface structure.

[0013] According to a technical solution of the present invention, the optical convolution module performs convolution calculation on the effective scene information through an optical 4F system to obtain a multi-channel output optical feature map;

[0014] The optical 4F system includes two lenses with equal focal lengths and a spatial filter between the two lenses.

[0015] According to a technical solution of the present invention, the optical fully-connected module adopts an optical diffraction unit to map different feature information to different spatial positions or angular directions through diffraction and interference phenomena.

[0016] According to a technical solution of the present invention, the metasurface structure is composed of sub-wavelength-scale unit structures, and the encoding and compression of optical signals are achieved by changing the shape, size, spacing and arrangement of the unit structures. The unit structures are obtained through data-driven training and have spatial light field intensity and phase modulation functions.

[0017] According to one aspect of the present invention, a method for constructing an optical computing system as described in any one of the above technical solutions is proposed, comprising:

[0018] According to typical scenarios and indicator requirements, numerical models of optical compression module, optical convolution module and optical fully connected module are constructed, and the relevant parameters of optical compression template, convolution kernel and fully connected diffraction plate are completed and optimized through numerical simulation methods.

[0019] According to a technical solution of the present invention, a metasurface structure is designed based on relevant parameters of an optical compression template;

[0020] Design and create spatial filters based on the relevant parameters of the convolution kernel;

[0021] Based on the relevant parameters of the fully connected diffraction plate, an optical diffraction unit is designed and manufactured.

[0022] Compared with the prior art, the present invention has the following beneficial effects:

[0023] The present invention proposes an optical computing system and a construction method thereof, which realizes all-optical computing of optical remote sensing through an optical compression module, an optical convolution module and an optical fully connected module, without converting optical signals into electrical signals. The computing rate and efficiency are much higher than those of existing electronic computing, which greatly improves the processing speed and efficiency of the system and is more suitable for optical remote sensing applications with high real-time requirements.

[0024] All-optical scene compression is achieved through the optical compression module. First, the large-scale scene is compressed to a small scale, which not only reduces the scene scale, but also increases the proportion of the target in the scene, providing a guarantee for the subsequent optical convolution and optical full-connection calculations, so as to effectively process large-scale natural scene data for optical remote sensing. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention, and for ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0026] Figure 1 A schematic diagram schematically shows the architecture of an optical computing system in an embodiment of the present invention;

[0027] Figure 2 A schematic diagram schematically shows the composition of an optical computing system according to an embodiment of the present invention;

[0028] Figure 3 Schematically showing an optical convolution calculation kernel of an optical convolution module according to an implementation example of the present invention;

[0029] Figure 4 (ac) schematically show the input and output results of the optical convolution calculation kernel in an embodiment of the present invention. DETAILED DESCRIPTION

[0030] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0031] like Figure 1 and Figure 2 As shown, an optical computing system of the present invention comprises:

[0032] An optical compression module is used to compress redundant and invalid information in the scene after the light signal of the natural scene is incident on the optical system, so as to obtain compressed valid scene information mainly focusing on the target;

[0033] Optical compression is located at the forefront of the optical computing system. After the light signal of the natural scene is incident on the optical system, the optical compression module compresses the light signal through spatial coding to obtain compressed scene information mainly based on the target. The optical compression module includes an optical compression coding version composed of a metasurface structure. The metasurface structure is composed of a sub-wavelength-scale unit structure. The encoding and compression of the light signal are achieved by changing the shape, size, spacing and arrangement of the unit structure. The unit structure is obtained through data-driven training and has the function of spatial light field intensity and phase modulation.

[0034] In typical scenarios, high-rate optical compression can be achieved through feature extraction.

[0035] An optical convolution module is used to process effective scene information to obtain a multi-channel output optical feature map; the optical convolution module performs convolution calculation on the effective scene information through an optical 4F system to obtain a multi-channel output optical feature map; the optical 4F system includes two lenses with equal focal lengths and a spatial filter between the two lenses.

[0036] like Figure 3 and Figure 4 (ac) as shown, Figure 3 This is a typical optical convolution kernel designed for the MNIST dataset. Figure 4 (a) is the input number 5, Figure 4 (b) 25-channel tiled feature map output by simulation, Figure 4 (c) is the experimental output of the 25-channel tiled feature map result.

[0037] The optical fully connected module is used to perform optical diffraction fully connected calculations based on the multi-channel output optical feature map and output the target classification result. The optical fully connected module uses an optical diffraction unit to map different feature information to different spatial positions or angular directions through diffraction and interference phenomena.

[0038] According to one aspect of the present invention, a method for constructing an optical computing system as described in any one of the above technical solutions is proposed, comprising:

[0039] According to typical scenarios and indicator requirements, numerical models of optical compression module, optical convolution module and optical fully connected module are constructed, and the relevant parameters of optical compression template, convolution kernel and fully connected diffraction plate are completed and optimized through numerical simulation. Specifically, based on the relevant parameters of optical compression template, the metasurface structure is designed; based on the relevant parameters of convolution kernel, the spatial filter is designed and manufactured; based on the relevant parameters of fully connected diffraction plate, the optical diffraction unit is designed and manufactured.

[0040] In view of the large-scale natural scene characteristics and specific detection and classification index requirements of optical remote sensing, numerical models of optical compression, optical convolution and optical full connection are established. When constructing the optical compression module, the spectral characteristics, spatial distribution and other factors of different objects in the natural scene are fully considered to achieve effective identification and compression of redundant and invalid information. For the optical convolution module, according to the feature extraction requirements of different targets, appropriate convolution kernel parameters are designed so that the key features of the target can be accurately extracted. The optical fully connected module optimizes the parameters of the fully connected diffraction plate according to the number of categories and accuracy requirements of target classification to ensure that different targets can be accurately classified and displayed. Through numerical simulation methods, the parameters in the above models are repeatedly adjusted and optimized to achieve the best performance.

[0041] For example, when simulating the processing of optical remote sensing data of sea surface scenes, the parameters of the optical compression model are adjusted according to the spectral reflection characteristics of different objects in the scene, such as aircraft carriers, destroyers, cargo ships, fishing boats, etc., so that the system can highlight the information of key targets while compressing the scene information, providing more accurate data support for subsequent monitoring and early warning.

[0042] For the optical compression module, the specific scheme of the metasurface structure is designed according to the optimized optical compression template parameters. The metasurface structure is usually composed of a series of sub-wavelength-scale unit structures, which can accurately control the phase, amplitude, polarization and other characteristics of the incident light. For example, metal nanoantennas, dielectric resonators or plasma structures are used as metasurface units, and the specific encoding and compression functions of natural scene light signals are realized by changing the shape, size, spacing and arrangement of the units.

[0043] Using micro-nano processing technologies such as electron beam lithography, focused ion beam etching, nanoimprinting and other processes, the designed metasurface structure is prepared on a suitable substrate material, such as silicon wafers, glass or flexible polymers. During the preparation process, the process parameters need to be strictly controlled to ensure the accuracy and quality of the metasurface structure in order to achieve the expected optical compression effect.

[0044] The prepared optical compression coding board is integrated into the optical system, so that the light signal of the natural scene can be accurately incident on the metasurface structure. After the encoding and compression of the metasurface, the compressed scene information is output and transmitted to the subsequent optical convolution module for further processing.

[0045] For the optical convolution module, an optical 4F system is built according to the design requirements. The optical 4F system mainly consists of two lenses with equal focal lengths and a spatial filter located between the two lenses. According to the optimized convolution kernel parameters, the corresponding spatial filter is designed and manufactured. The spatial filter can be an amplitude type, a phase type, or a hybrid filter, and its transmittance function should correspond to the mathematical expression of the convolution kernel.

[0046] The compressed scene information output by the optical compression module is used as the input of the optical 4F system. The optical signal undergoes Fourier transformation after passing through the first lens, and multiplies with the spatial filter on the spectrum plane to achieve spectrum domain processing of convolution operation. Then, the spectrum information is converted back to the spatial domain through the inverse Fourier transformation through the second lens to obtain the optical feature map of multi-channel output.

[0047] The feature map obtained by the optical convolution calculation is inspected and analyzed to evaluate its quality and accuracy. The convolution calculation can be checked to see if it has achieved the expected results by comparing it with a known standard feature map or by using an image recognition algorithm to extract and identify the target features in the feature map. If the feature map is found to be defective or inaccurate, the parameters of the optical 4F system, the design of the spatial filter, or the output of the optical compression module need to be adjusted and optimized.

[0048] For the optical fully connected module, an optical diffraction unit is designed and manufactured according to the optimized fully connected diffraction plate parameters. The optical diffraction unit can use a holographic grating, a Fresnel lens array or other diffraction optical elements, and its diffraction characteristics should meet the requirements of optical fully connected calculations, and can perform spatial optical diffraction on the input feature map to achieve classified display of different targets.

[0049] A photodetector array is set at the output end of the optical fully connected module to detect and collect the intensity distribution of the diffracted light signal. The classification results of different targets are determined by processing and analyzing the data collected by the photodetector array. For example, threshold judgment, cluster analysis or machine learning algorithms can be used to classify and identify the signals output by the photodetector array, and finally output the classification results of the target, such as the target's category, location, shape and other information.

[0050] An optical computing system and a construction method thereof of the present invention realize all-optical computing of optical remote sensing through an optical compression module, an optical convolution module and an optical fully connected module, without converting optical signals into electrical signals. The computing rate and efficiency are far higher than those of existing electronic computing, which greatly improves the processing speed and efficiency of the system and is more suitable for optical remote sensing applications with high real-time requirements.

[0051] All-optical scene compression is achieved through the optical compression module. First, the large-scale scene is compressed to a small scale, which not only reduces the scene scale, but also increases the proportion of the target in the scene, providing a guarantee for the subsequent optical convolution and optical full-connection calculations, so as to effectively process large-scale natural scene data for optical remote sensing.

[0052] It should be noted that, in the present invention, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or terminal device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or terminal device. In the absence of further restrictions, the elements defined by the sentence "include one..." do not exclude the existence of other identical elements in the process, method, article or terminal device including the elements.

[0053] It should also be noted that the above is a preferred embodiment of the present invention. It should be noted that although the preferred embodiment of the present invention has been described, once the basic creative concept of the present invention is known, a number of improvements and modifications can be made by a person skilled in the art without departing from the principles of the present invention, and these improvements and modifications should also be regarded as the protection scope of the present invention. Therefore, the attached claims are intended to be interpreted as including the preferred embodiment and all changes and modifications that fall within the scope of the embodiments of the present invention.

Claims

1. An optical computing system, characterized in that: include: An optical compression module is used to compress redundant and invalid information in the scene after the light signal of the natural scene is incident on the optical system, so as to obtain compressed valid scene information mainly focusing on the target; Optical convolution module, used to process effective scene information and obtain multi-channel output optical feature maps; The optical fully connected module is used to perform optical diffraction fully connected calculations based on the multi-channel output optical feature map and output the target classification results.

2. The optical computing system according to claim 1, characterized in that The optical compression module compresses the optical signal through spatial coding to obtain compressed scene information mainly based on the target; The optical compression module includes an optical compression encoding plate composed of a metasurface structure.

3. The optical computing system according to claim 1, characterized in that The optical convolution module performs convolution calculation on the effective scene information through an optical 4F system to obtain a multi-channel output optical feature map; The optical 4F system includes two lenses with equal focal lengths and a spatial filter between the two lenses.

4. The optical computing system according to claim 1, characterized in that The optical fully connected module adopts an optical diffraction unit to map different feature information to different spatial positions or angular directions through diffraction and interference phenomena.

5. The optical computing system according to claim 2, characterized in that The metasurface structure is composed of sub-wavelength-scale unit structures, and encoding and compression of optical signals are achieved by changing the shape, size, spacing and arrangement of the unit structures. The unit structures are obtained through data-driven training and have spatial light field intensity and phase modulation functions.

6. A method for constructing an optical computing system according to any one of claims 1 to 5, characterized in that: include: According to typical scenarios and indicator requirements, numerical models of optical compression module, optical convolution module and optical fully connected module are constructed, and the relevant parameters of optical compression template, convolution kernel and fully connected diffraction plate are completed and optimized through numerical simulation methods.

7. The method for constructing an optical computing system according to claim 6, characterized in that: Design the metasurface structure based on the relevant parameters of the optical compression module; Design and create spatial filters based on the relevant parameters of the convolution kernel; Based on the relevant parameters of the fully connected diffraction plate, an optical diffraction unit is designed and manufactured.

Images