System for implementing incoherent optical neural network

Abstract

The invention provides a system for realizing an incoherent optical neural network, and relates to the technical field of optical neural networks. The system comprises a coherent light module used for providing spatial coherent parallel beams; the incoherent optical module is used for providing an incoherent target image; the incoherent-coherent optical conversion module is used for converting the incoherent target image into a coherent optical image; the optical path multiplexing module is used for reflecting the incoherent target image to the incoherent-coherent optical conversion module and is also used for transmitting the coherent optical image; and the space optical neural network identification module is used for carrying out optical network identification on the transmitted coherent optical image. According to the system for realizing the incoherent optical neural network provided by the invention, the incoherent-coherent light conversion modulation device is innovatively adopted, so that incoherent intensity information is coupled to coherent light, and a spatial diffraction neural network image classification and recognition task under an incoherent light condition can be realized.

Description

Technical Field

[0001] This application relates to the field of optical neural network technology, and more specifically, to a system for implementing incoherent optical neural networks. Background Technology

[0002] Optical neural networks utilize optical signals for data processing and computation, offering significant advantages over traditional electrical neural networks. Limited by the inherent limitations of electronic devices and the von Neumann architecture, traditional electronic computing faces insurmountable bottlenecks in speed and energy efficiency. Optical neural networks fully integrate the performance advantages of optics, such as multidimensional multiplexing, high bandwidth, and low power consumption. They utilize complex optical fields as a carrier for neural network computation and data processing, possessing unique advantages such as low transmission latency and powerful parallel processing capabilities. Spatial diffraction optical neural networks utilize the phase spatial distribution of coherent light fields to achieve optical computation. In image classification tasks, the image input from a coherent light source passes through multiple phase masks, and the output light field intensity distribution is used for classification and recognition.

[0003] Therefore, most current spatial diffraction optical neural networks still rely on highly coherent light sources for operation, and cannot perform optical neural network recognition of real objects under incoherent light conditions, i.e., natural light. This greatly limits the expansion of practical applications of spatial diffraction optical neural networks. Summary of the Invention

[0004] To address at least one of the aforementioned problems, this application proposes a system for implementing incoherent optical neural networks.

[0005] According to a first aspect of this application, at least one embodiment of this application provides a system for implementing an incoherent optical neural network, comprising: a coherent optical module for providing a spatially coherent parallel beam; an incoherent optical module for providing an incoherent target image; an incoherent-to-coherent optical conversion module for converting the incoherent target image into a coherent optical image; an optical path multiplexing module for reflecting the incoherent target image to the incoherent-to-coherent optical conversion module and for transmitting the coherent optical image; and a spatial optical neural network recognition module for performing optical network recognition on the transmitted coherent optical image.

[0006] For example, in some embodiments of this application, the coherent light module includes: a coherent light source generating device for providing coherent light to generate a coherent light beam; and a coherent beam adjusting device for adjusting the coherent light beam to generate the spatially coherent parallel beam.

[0007] For example, in some embodiments of this application, the coherent light source generating device includes a laser.

[0008] For example, in some embodiments of this application, the coherent beam adjustment device includes a beam expander and / or an assembled lens.

[0009] For example, in some embodiments of this application, the incoherent optical module includes: an imaging device for imaging an incoherent target to generate an image of the incoherent target; and a filtering device for filtering the wavelength of the incoherent light so that the image of the incoherent target can pass through.

[0010] For example, in some embodiments of this application, the imaging device includes an imaging lens.

[0011] For example, in some embodiments of this application, the filtering device includes: a bandpass filter, an attenuation filter, a cutoff filter, and / or a beam splitter filter.

[0012] For example, in some embodiments of this application, the incoherent-to-coherent optical conversion module includes: an incoherent-to-coherent optical conversion device, used to couple the amplitude and phase information of the light carried in the incoherent target image to the spatially coherent parallel beam to generate the coherent light image.

[0013] For example, in some embodiments of this application, the incoherent-to-coherent optical conversion device includes a metasurface ethyl red thin film composite device, a BSO / liquid crystal cell composite device, a GaAs\BSO composite crystal device, a 6-thiophene (α-6T) thin film device, and / or a photorefractive crystal device.

[0014] For example, in some embodiments of this application, the optical path multiplexing module includes: a beam splitting device for reflecting the incoherent target image to the incoherent-to-coherent optical conversion module, and for transmitting the coherent light image to the spatial optical neural network recognition module.

[0015] For example, in some embodiments of this application, the beam splitting device includes a beam splitter and / or a beam splitter prism.

[0016] For example, in some embodiments of this application, the spatial optical neural network recognition module includes: an optical neural network device for performing calculations on the coherent light image at the speed of light; and a photodetector device for displaying the calculation results performed by the optical neural network device.

[0017] For example, in some embodiments of this application, the photodetector includes a charge-coupled device, a complementary metal-oxide-semiconductor detector, and / or a photodiode array.

[0018] Through the above example embodiments, this application provides a system for implementing incoherent optical neural networks. It innovatively employs an incoherent-to-coherent light conversion modulation device to couple incoherent intensity information to coherent light, enabling spatial diffraction neural network image classification and recognition tasks under incoherent light conditions. It incorporates a coherent light source module, which converts incoherent images of targets in natural light (i.e., incoherent light environments) into coherent light images using the incoherent-to-coherent light conversion modulation device for recognition by the spatial diffraction optical neural network. This allows for target classification and recognition using a spatial diffraction optical neural network architecture under incoherent natural light conditions, and can also be applied to any optical neural network constrained by coherent light conditions.

[0019] This application allows for the identification of incoherent light targets simply by changing the structure and devices of the front-end image feed system based on a spatial diffraction optical neural network. This expands the application of spatial diffraction optical neural networks beyond the laboratory, enabling optical neural network identification and classification of spatial targets under natural light conditions. This provides significant meaning and value for the application of spatial diffraction optical neural networks in real-world environments. Furthermore, this application can be applied to any optical neural network computing architecture based on coherent light, extending to incoherent light application scenarios, and further integrating it into edge computing scenarios such as autonomous driving, industrial robots, and unmanned equipment.

[0020] It should be understood that the above general description and the following detailed description are merely exemplary and do not limit this application. Attached Figure Description

[0021] The above and other objects, features, and advantages of this application will become more apparent from the detailed description of exemplary embodiments with reference to the accompanying drawings. The drawings described below are merely some embodiments of this application and are not intended to limit the scope of this application.

[0022] Figure 1 A schematic diagram of a system for implementing an incoherent optical neural network, as shown in an exemplary embodiment;

[0023] Figure 2 A schematic diagram illustrating the experimental structure and working principle of a system for implementing an incoherent optical neural network, as shown in an exemplary embodiment;

[0024] Figure 3 A schematic diagram illustrating the operation of an incoherent-to-coherent optical conversion device in an exemplary embodiment is shown.

[0025] Figure 4 A schematic diagram illustrating an example of an incoherent neural network application in the laboratory is shown;

[0026] Figure 5A schematic diagram illustrating an example of an outdoor incoherent neural network application is shown. Detailed Implementation

[0027] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the embodiments set forth herein; rather, they are provided so that this application will be thorough and complete, and will fully convey the concept of the exemplary embodiments to those skilled in the art. The same reference numerals in the drawings denote the same or similar parts, and therefore repeated descriptions of them will be omitted.

[0028] The described features, structures, or characteristics can be combined in any suitable manner in one or more embodiments. Numerous specific details are provided in the following description to give a full understanding of embodiments of this disclosure. However, those skilled in the art will recognize that the technical solutions of this disclosure can be practiced without one or more of these specific details, or other methods, components, materials, devices, etc. In these cases, well-known structures, methods, devices, implementations, materials, or operations will not be shown or described in detail.

[0029] The flowcharts shown in the accompanying drawings are merely illustrative and do not necessarily include all content and operations / steps, nor do they necessarily have to be performed in the described order. For example, some operations / steps can be broken down, while others can be combined or partially combined; therefore, the actual execution order may change depending on the specific circumstances.

[0030] The terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish different objects, not to describe a specific order. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or apparatus that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products, or apparatuses.

[0031] Those skilled in the art will understand that the accompanying drawings are merely schematic diagrams of exemplary embodiments, and the modules or processes in the drawings are not necessarily essential for implementing this application, and therefore cannot be used to limit the scope of protection of this application.

[0032] The purpose of this application is to overcome the bottleneck that spatial diffractive optical neural networks can only perform target recognition based on coherent light sources, and to overcome the defect that spatial diffractive optical neural networks cannot directly recognize natural incoherent light targets, so as to provide a system for realizing incoherent optical neural networks.

[0033] Figure 1A schematic diagram of a system for implementing an incoherent optical neural network, as shown in an exemplary embodiment, is illustrated.

[0034] like Figure 1 As shown, the system for implementing an incoherent optical neural network includes: a coherent optical module 10, an incoherent optical module 20, an incoherent-to-coherent optical conversion module 30, an optical path multiplexing module 40, and a spatial optical neural network recognition module 50.

[0035] The system includes a coherent optical module 10 for providing a spatially coherent parallel beam, an incoherent optical module 20 for providing an incoherent target image, an incoherent-to-coherent optical conversion module 30 for converting the incoherent target image into a coherent optical image, an optical path multiplexing module 40 for reflecting the incoherent target image to the incoherent-to-coherent optical conversion module 30 and for transmitting the coherent optical image, and a spatial optical neural network recognition module 50 for performing optical network recognition on the transmitted coherent optical image.

[0036] like Figure 2 As shown, the coherent optical module 10 includes a coherent light source generating device 101 and a coherent beam adjustment device 102.

[0037] The coherent light source generating device 101 is used to provide coherent light to generate a coherent light beam. The coherent beam adjusting device 102 is used to adjust the coherent light beam to generate a spatially coherent parallel beam of a certain size.

[0038] According to some embodiments, the coherent light source generating device 101 refers to a device or apparatus that generates optical oscillation and emits coherent light through optical feedback formed by a resonant cavity or other means in a certain material that can generate stimulated emission amplification of photons, including but not limited to lasers of any wavelength.

[0039] According to some embodiments, the coherent beam adjustment device 102 refers to a lens assembly capable of changing the diameter and divergence angle of the beam, including but not limited to any beam expander, assembled lens, or other device that can provide beam scaling function.

[0040] like Figure 2 As shown, the incoherent optical module 20 includes an imaging device 201 and a filter device 202.

[0041] Imaging device 201 is used to image incoherent targets and generate incoherent target images. Filtering device 202 is used to filter incoherent light wavelengths so that incoherent light of a set wavelength, i.e., the incoherent target image, can pass through.

[0042] According to some embodiments, the imaging device 201 can image incoherent targets, collect light from incoherent light sources that are reflected or diffusely reflected by the object, and form an incoherent light image. An image with the same distribution as the object's outline can be obtained at the imaging point, including but not limited to various imaging lenses.

[0043] According to some embodiments, the filter device 202 can selectively transmit or reflect light of different radiation bands or different energy intensities, including but not limited to various filter devices such as bandpass filters, attenuation filters, cutoff filters, and beam splitters.

[0044] like Figure 2 As shown, the optical path multiplexing module 40 includes a beam splitter 401, which is used to realize the multiplexing of different beams in a unified optical path, causing the incoherent optical path to be redirected and propagated in the direction of the incoherent-to-coherent optical conversion device, acting on the surface of the incoherent-to-coherent optical conversion device 301, and modulating the characteristics of the coherent beam.

[0045] The beam splitter 401 is used to split an incident beam into two beams with a certain intensity ratio: a transmitted beam and a reflected beam. The incoherent target image is reflected to the incoherent-to-coherent optical conversion module 30, and the coherent light image is transmitted to the spatial optical neural network recognition module 50. The beam splitter 401 includes, but is not limited to, beam splitters and beam splitting prisms with any beam splitting ratio.

[0046] like Figure 2 As shown, the incoherent-to-coherent optical conversion module 30 includes an incoherent-to-coherent optical conversion device 301. The incoherent-to-coherent optical conversion device 301 is used to couple the amplitude and phase information of the light carried in the incoherent target image into a spatially coherent parallel beam to generate a coherent light image.

[0047] According to some embodiments, the incoherent-to-coherent optical conversion device 301 can couple the amplitude and phase information of light carried in incoherent light into coherent light, and continue to propagate in the form of coherent light. The incoherent-to-coherent optical conversion device 301 includes, but is not limited to, metasurface ethyl red thin film composite devices, BSO / liquid crystal cell composite devices, GaAs\BSO composite crystal devices, 6-thiophene (α-6T) thin film devices, photorefractive crystal devices, and other devices that realize the upconversion from incoherent light to coherent light in any way.

[0048] The beam splitter 401 is also used to transmit coherent light carrying incoherent light information, and to obtain coherent light imaging of the target at the rear end of the beam splitter 401.

[0049] like Figure 2 As shown, the spatial optical neural network recognition module 50 includes: an optical neural network device 501 and a photodetector device 502.

[0050] The optical neural network device 501 is used to perform calculations on the coherent light image at the speed of light. The photodetector device 502 is used to display the results of the calculations performed by the optical neural network device 501.

[0051] According to some embodiments, the optical neural network device 501 refers to a device capable of utilizing the physical properties of light to simulate a traditional electronic neural network computing architecture and perform computational tasks at the speed of light. The optical neural network computing architecture includes, but is not limited to, spatial optical neural network architectures such as optical convolutional neural networks, optical diffraction neural networks, and optoelectronic hybrid neural networks, as well as on-chip optical neural network architectures. The computational methods used to perform the computational tasks include, but are not limited to, matrix multiplication and nonlinear activation functions.

[0052] According to some embodiments, the photodetector 502 can convert optical signals into electrical signals, including but not limited to charge-coupled devices (CCDs), complementary metal-oxide-semiconductor detectors (CMOS), photodiode arrays, and other detectors capable of photoelectric conversion.

[0053] like Figure 2 As shown, the coherent light source generating device 101, the coherent beam adjusting device 102, the beam splitting device 401, the optical neural network device 501, and the detector device 502 are arranged sequentially on the coherent optical path optical axis 001. The filter device 202, the imaging device 201, and the incoherent target 003 are arranged sequentially on the incoherent optical path optical axis 002, and the coherent optical path optical axis 001 and the incoherent optical path optical axis 002 form a 90° angle.

[0054] The working principle of the system used to implement incoherent optical neural networks is as follows:

[0055] A coherent light beam, via a coherent light source generator 101 and a coherent beam adjustment device 102, is incident on the surface of an incoherent-to-coherent optical converter 301 in the form of spatial light. Incoherent light emitted from a target in natural light is reflected by an imaging device 201, a filter device 202, and a beam splitter 401, and imaged onto the surface of the incoherent-to-coherent optical converter 301 in the form of narrowband light. Through modulation by the incoherent-to-coherent optical converter 301, the amplitude and phase information carried in the incoherent light are coupled into the coherent light, and the light continues to propagate in the form of coherent light. Figure 3 As shown; in the direction of coherent light propagation, the beam splitter 401 transmits coherent light carrying incoherent light information; at the rear end of the beam splitter 401, coherent light imaging of the target is obtained, realizing the coherent light path and the incoherent light path being coupled in the same light path through the beam splitter 401 for optical path multiplexing; the coherent light image plane coincides with the surface of the optical neural network device 501, and optical calculations are performed on the coherent light image of the target, and the calculated results are displayed on the detector device 502.

[0056] According to some embodiments, the size of the spatial light spot incident on the surface of the incoherent-to-coherent optical conversion device 301 by the coherent light beam is greater than or equal to the image plane size of the incoherent light target image on the surface of the incoherent-to-coherent optical conversion device 301, ensuring that the acquired incoherent target information can be completely coupled into the spatial optical neural network recognition module.

[0057] This application provides a system for implementing incoherent optical neural networks. It innovatively employs an incoherent-to-coherent light conversion modulation device to couple incoherent intensity information to coherent light, enabling spatial diffraction neural network image classification and recognition tasks under incoherent light conditions. It incorporates a coherent light source module, which converts incoherent images of targets in natural light (i.e., incoherent light environments) into coherent light images for recognition by the spatial diffraction optical neural network. This allows for target classification and recognition under incoherent natural light conditions using a spatial diffraction optical neural network architecture, and can also be applied to any optical neural network constrained by coherent light conditions.

[0058] This application allows for the identification of incoherent light targets simply by changing the structure and devices of the front-end image feed system based on a spatial diffraction optical neural network. This expands the application of spatial diffraction optical neural networks beyond the laboratory, enabling optical neural network identification and classification of spatial targets under natural light conditions. This provides significant meaning and value for the application of spatial diffraction optical neural networks in real-world environments. Furthermore, this application can be applied to any optical neural network computing architecture based on coherent light, extending to incoherent light application scenarios, and further integrating it into edge computing scenarios such as autonomous driving, industrial robots, and unmanned equipment.

[0059] To illustrate the broad applicability of the incoherent optical neural network method described in this application, a target recognition and classification task under incoherent light in a laboratory setting is presented according to one embodiment of this application.

[0060] Figure 4 A schematic diagram illustrating an example of an incoherent neural network application in the laboratory is shown.

[0061] like Figure 4As shown, the coherent light source generating device 102 uses an 820nm laser light source; the incoherent-to-coherent optical conversion device 301 adopts a metasurface ethyl red thin film composite device; the incoherent target 003 adopts a digital micromirror array (DMD), with each micromirror control unit having an area of ​​0.8mm × 0.8mm; the imaging device 201 adopts a 4f lens, which is composed of two lenses placed at the sum of two focal lengths; the incoherent light source 004 adopts an incoherent LED light source; the optical neural network device 501 adopts a spatial optical diffraction neural network recognition chip; the detection device 502 adopts a complementary metal-oxide-semiconductor (CMOS) detector; the beam splitting device 401 adopts a beam splitter with a 50:50 splitting ratio; and the filter device 202 adopts a 532nm narrowband filter.

[0062] Incoherent light passes through a 532nm filter to form 532nm wavelength incoherent light, which is then transmitted to the DMD to load an image. The output incoherent image is imaged onto the surface of the metasurface ethyl red composite device through a 4f lens. Light emitted from an 820nm laser source is expanded by a beam expander and transmitted to the surface of the metasurface ethyl red composite device. Coupling of incoherent light to coherent light is achieved at the metasurface ethyl red composite device. The generated coherent image continues to propagate to a spatial optical diffraction neural network recognition chip for calculation. The calculation result is transmitted to a CMOS to realize incoherent light target classification and recognition in the laboratory.

[0063] To illustrate the broad applicability of the incoherent optical neural network method described in this application, a target recognition and classification task under outdoor natural light is presented according to one embodiment of this application.

[0064] Figure 5 A schematic diagram illustrating an example of an outdoor incoherent neural network application is shown.

[0065] like Figure 5 As shown, the coherent light source generating device 102 uses an 820nm laser light source; the incoherent-to-coherent optical conversion device 301 adopts a metasurface ethyl red composite device; the road sign is used as an incoherent target 003; the imaging device 201 adopts a 50mm focal length imaging objective lens; natural light is used as an incoherent light source 004; the optical neural network device 501 adopts a spatial optical diffraction neural network recognition chip; the detection device 502 adopts a complementary metal-oxide-semiconductor (CMOS) detector; the beam splitting device 401 adopts a beam splitter with a 50:50 splitting ratio; and the filter device 202 adopts a 532nm narrowband filter.

[0066] Under diffuse reflection of natural light, the road sign is imaged through a 50mm imaging lens and then through a 532nm narrowband filter to form an incoherent image at a wavelength of 532nm, which is then imaged onto the surface of a metasurface ethyl red composite device. Light emitted from an 820nm laser source is expanded by a beam expander and transmitted to the surface of the metasurface ethyl red composite device. At the metasurface ethyl red composite device, incoherent light is coupled to coherent light. The generated coherent image continues to propagate to a spatial optical diffraction neural network recognition chip for calculation. The calculation result is transmitted to the CMOS detector device to achieve incoherent light target classification and recognition in outdoor natural light.

[0067] It should be clearly understood that this application describes how specific examples are formed and used, but this application is not limited to any details of these examples. Rather, based on the teachings of the disclosure of this application, these principles can be applied to many other embodiments.

[0068] Furthermore, it should be noted that the above figures are merely illustrative representations of the processes included in the method according to exemplary embodiments of this application, and are not intended to be limiting. It is readily understood that the processes shown in the above figures do not indicate or limit the temporal order of these processes. Additionally, it is readily understood that these processes may be executed synchronously or asynchronously, for example, in multiple modules.

[0069] Exemplary embodiments of this application have been specifically shown and described above. It should be understood that this application is not limited to the detailed structures, arrangements, or implementation methods described herein; rather, this application is intended to cover various modifications and equivalent arrangements that fall within the objectives and scope of the appended claims.

Claims

1. A system for implementing incoherent optical neural networks, characterized in that, include: Coherent optical modules are used to provide spatially coherent parallel beams; Incoherent optical modules are used to provide incoherent target images; An incoherent-to-coherent optical conversion module is used to convert the incoherent target image into a coherent light image; The optical path multiplexing module is used to reflect the incoherent target image to the incoherent-to-coherent optical conversion module, and also to transmit the coherent light image; The spatial optical neural network recognition module is used to perform optical network recognition on the transmitted coherent light image.

2. The system as described in claim 1, characterized in that, The coherent optical module includes: A coherent light source generating device, used to provide coherent light to generate a coherent light beam; A coherent beam adjustment device is used to adjust the coherent light beam to generate the spatially coherent parallel beam.

3. The system as described in claim 2, characterized in that, The coherent light source generating device includes a laser.

4. The system as described in claim 2, characterized in that, The coherent beam adjustment device includes a beam expander and / or an assembled lens.

5. The system as described in claim 1, characterized in that, The incoherent optical module includes: An imaging device for imaging incoherent targets and generating an image of the incoherent targets; A filter device is used to filter the wavelengths of incoherent light so that the incoherent target image can pass through.

6. The system as described in claim 5, characterized in that, The imaging device includes an imaging lens.

7. The system as described in claim 5, characterized in that, The filtering device includes: a bandpass filter, an attenuation filter, a cutoff filter, and / or a beam splitter filter.

8. The system as described in claim 1, characterized in that, The incoherent-to-coherent optical conversion module includes: An incoherent-to-coherent optical conversion device is used to couple the amplitude and phase information of the light carried in the incoherent target image into the spatially coherent parallel beam to generate the coherent light image.

9. The system as described in claim 8, characterized in that, The incoherent-to-coherent optical conversion devices include metasurface ethyl red thin film composite devices, BSO / liquid crystal cell composite devices, GaAs\BSO composite crystal devices, 6-thiophene (α-6T) thin film devices and / or photorefractive crystal devices.

10. The system as claimed in claim 1, characterized in that, The optical path multiplexing module includes: A beam splitter is used to reflect the incoherent target image to the incoherent-to-coherent optical conversion module, and also to transmit the coherent light image to the spatial optical neural network recognition module.

11. The system as claimed in claim 10, characterized in that, The beam splitting device includes a beam splitter and / or a beam splitting prism.

12. The system as claimed in claim 1, characterized in that, The space optical neural network recognition module includes: An optical neural network device for performing calculations on the coherent light image at the speed of light; A photodetector device for displaying the results of calculations performed by the optical neural network device.

13. The system as described in claim 12, characterized in that, The photodetector includes a charge-coupled device, a complementary metal-oxide-semiconductor detector, and / or a photodiode array.

Images