Optical high-dimensional data time domain compression calculation method and system based on Hilbert mapping

Abstract

The invention provides an optical high-dimensional data time domain compression calculation method and system based on Hilbert mapping, and the method comprises the steps: employing a Hilbert mapping method, enabling original high-dimensional data to be expanded into a one-dimensional vector from a memory according to the dimension and size Size of the original high-dimensional data and a Hilbert curve with the order n of the same dimension, and carrying out the time domain compression calculation of the original high-dimensional data, and a digital-to-analog converter drives a high-speed optical modulator to load the vector to the time-varying intensity of the optical carrier so as to form signal light. Then, the signal light enters a basic time domain optical operation unit to be further processed, including weighted secondary intensity modulation, or optical difference (or differential) operation used for extracting a neighborhood information change relation and the like. And finally, vector information carried by the processed one-dimensional time domain signal light is extracted and stored, Hilbert inverse mapping of a corresponding dimension is carried out on the vector information, and the vector information is restored into high-dimensional feature information for subsequent operations such as judgment, segmentation and the like by an artificial intelligence algorithm.

Description

Technical Field

[0001] The present invention relates to the field of optical computing technology, and in particular to a method and system for calculating optical high-dimensional data time domain compression based on Hilbert mapping. Background Art

[0002] In the mainstream practice of current technology, photonic computing technology can map linear computing tasks in AI algorithms, such as matrix multiplication and addition operations and two-dimensional or higher-dimensional convolution operations, to single or multiple interacting physical dimensions of photons, or to the information flow of spatial light. Through this mapping, the technology can complete the above linear operations in the optical domain at high speed and low power consumption. The main advantages of this technology include large bandwidth, low latency and no ohmic heat loss. However, compared with the scalability of digital electronic AI computing accelerators, photonic computers have certain limitations in processing high-dimensional data (such as two-dimensional, three-dimensional and even four-dimensional data). These limitations are mainly reflected in the complexity and efficiency of actual operations.

[0003] Photonic computing technology is primarily categorized into spatial photonic computing and time-domain photonic computing. Spatial photonic computing offers high data throughput and is well-suited for processing data streams with a two-dimensional spatial distribution, such as spatial photonic convolutional neural networks (SPNs) that process image information. However, due to the stringent precision required of optical systems and the limitations of the limited aperture of paraxial optical systems and the limited refresh rate of light field manipulation devices, spatial photonic computing systems face significant challenges in stability, rapid reconfigurability, and integration. Furthermore, they are unable to process higher-dimensional data (three-dimensional, four-dimensional, etc.) in parallel. Advances in fiber optics, particularly integrated optics, have enabled time-domain photonic computing systems to achieve exceptional stability and integration. Leveraging high-speed and multi-dimensional modulation techniques, time-domain photonic computing systems can achieve rapid data throughput and parameter updates. However, basic time-domain optical computing units can only process one-dimensional or vectorized two-dimensional data. While multiplexing schemes such as large-scale spatial division multiplexing (integrating as many basic time-domain optical computing units as possible on a single chip) can be used to expand the data processing dimensionality of time-domain photonic computing systems, such implementation is complex and difficult. More importantly, conventional linear vectorization of high-dimensional data results in the loss of near-neighborhood information of the original high-dimensional data, significantly reducing the processing efficiency of AI algorithms. To compensate for this loss of near-neighborhood information, extensive electronic or optical caching technologies (such as optical delay line arrays) are required to preserve and process near-neighborhood information. Summary of the Invention

[0004] In view of the defects in the prior art, the object of the present invention is to provide an optical operation method and system for time-domain compression of high-dimensional data based on Hilbert mapping.

[0005] According to the present invention, an optical operation method for time-domain compression of high-dimensional data based on Hilbert mapping is provided, comprising:

[0006] The original high-dimensional data is expanded into a one-dimensional vector using the Hilbert mapping method. A high-speed optical modulator then loads the one-dimensional vector onto the time-varying intensity of an optical carrier to form a signal light.

[0007] The basic time-domain optical operation unit performs time-domain compression processing on the optical signal to obtain a one-dimensional time-domain signal;

[0008] The vector information carried in the one-dimensional time domain signal is extracted, and the vector information is subjected to Hilbert inverse mapping of the corresponding dimension to restore it to high-dimensional feature information.

[0009] Preferably, the method of expanding the original high-dimensional data into a one-dimensional vector using the Hilbert mapping method includes: padding or clipping the original data according to the dimension and size of the original data to determine the data size and the Hilbert curve order of the corresponding dimension;

[0010] Map the original data according to the order of the Hilbert curve to obtain a one-dimensional vector

[0011] Preferably, the step of performing time domain compression processing on the optical signal by the basic time domain optical operation unit to obtain a one-dimensional time domain signal comprises:

[0012] Secondarily modulating the signal light using an optical intensity modulator or an adjustable attenuator;

[0013] The time-varying transmittance of the optical intensity modulator or variable attenuator used as weight is set to The clock of the DAC driving the optical intensity modulator or variable attenuator needs to be aligned with the clock of the DAC that implements the signal input, and the symbol rate must be consistent. Therefore, the intensity of the processed optical signal is expressed as: After being received and sampled by the light detector, the weighted vector is obtained Where ω is the center frequency of the optical carrier, is the initial phase of the optical carrier.

[0014] Preferably, the step of performing time domain compression processing on the optical signal by the basic time domain optical operation unit to obtain a one-dimensional time domain signal comprises:

[0015] Use optical filters to detect the rising and falling edges of signal light in the time domain;

[0016] Use an optical detector to record the intensity of the timing spike pulse in sequence and sample it to form a vector Among them, c k =abs(ak -a k-1 ), where abs(·) is the absolute value operation, and Compared to There is an element missing.

[0017] Preferably, the optical filter is an optical first-order differential operator, and is composed of an optical filter with a "V"-shaped transmission spectrum.

[0018] Preferably, the Hilbert inverse mapping of the corresponding dimension is performed on the vector information to restore it into high-dimensional feature information, including: vector Or a vector filled with 0 The kth element in is filled into the coordinates corresponding to the kth connection point on the selected Hilbert curve in the high-dimensional space, so as to realize the dimension increase of the processed vector data and the Hilbert inverse mapping.

[0019] Preferably, when the data size exceeds a preset condition, the original data is divided into blocks, the divided data is Hilbert mapped, and then modulated onto the time-varying intensity of light carriers with different center frequencies to form multiple signal lights.

[0020] According to the present invention, an optical computing system for high-dimensional data time-domain compression based on Hilbert mapping is provided, comprising:

[0021] The Hilbert mapping method is used to expand the original high-dimensional data into a one-dimensional vector. A high-speed optical modulator loads the one-dimensional vector onto the time-varying intensity of the optical carrier to form a signal light.

[0022] The basic time-domain optical operation unit performs time-domain compression processing on the optical signal to obtain a one-dimensional time-domain signal;

[0023] The vector information carried in the one-dimensional time domain signal is extracted, and the vector information is subjected to Hilbert inverse mapping of the corresponding dimension to restore it to high-dimensional feature information.

[0024] Preferably, the method of expanding the original high-dimensional data into a one-dimensional vector using the Hilbert mapping method includes: padding or clipping the original data according to the dimension and size of the original data to determine the data size and the Hilbert curve order of the corresponding dimension;

[0025] Map the original data in the order of the Hilbert curve to obtain a one-dimensional vector

[0026] Preferably, the step of performing time domain compression processing on the optical signal by the basic time domain optical operation unit to obtain a one-dimensional time domain signal comprises: performing weighted processing or differential processing on the optical signal by the basic time domain optical operation unit;

[0027] The weighting process includes:

[0028] Secondarily modulating the signal light using an optical intensity modulator or an adjustable attenuator;

[0029] The time-varying transmittance of the optical intensity modulator or variable attenuator used as weight is set to The clock of the DAC driving the optical intensity modulator or variable attenuator needs to be aligned with the clock of the DAC that implements the signal input, and the symbol rate must be consistent. Therefore, the intensity of the processed optical signal is expressed as: After being received and sampled by the light detector, the weighted vector is obtained Where ω is the center frequency of the optical carrier, is the initial phase of the optical carrier;

[0030] The differential processing includes:

[0031] Use optical filters to detect the rising and falling edges of signal light in the time domain;

[0032] Use an optical detector to record the intensity of the timing spike pulse in sequence and sample it to form a vector Among them, c k =abs(a k -a k-1 ), where abs(·) is the absolute value operation, and Compared to There is an element missing.

[0033] Compared with the prior art, the present invention has the following beneficial effects:

[0034] 1. The present invention can realize the boundary information extraction of high-dimensional data at low cost and quickly;

[0035] 2. The basic optical computing unit of the present invention can be a passive device and does not require additional adjustment and refresh after locking;

[0036] 3. The present invention can achieve high computing power based on band multiplexing and time-frequency network;

[0037] 4. Compared with the photon convolution operation architecture based on digital mask convolution operation, the present invention has a significant speed advantage in edge information extraction speed: under the same four-wavelength multiplexing physical layer overhead conditions, the present invention can achieve boundary surface information extraction of 3D volumes with a resolution of 1024×1024×1024 (1K) at more than 60 frames per second (FPS). Under the same conditions, the speed of optical convolution to extract boundary surfaces of 1K 3D volumes using traditional 2×2 digital masks is only 3FPS;

[0038] 5. Through optoelectronic conversion and digital-analog collaboration, this invention can be closely integrated with current long-distance optical communication networks and well embedded in interactive networks for virtual reality technologies such as VR and AR. It can be used to quickly and massively compress and extract key 3D information, achieving an information compression rate of 93%. This reduces the storage and computing power requirements of 3D information processing terminals and paves the way for portable, real-time 3D information interaction systems.

[0039] 6. The present invention provides a Hilbert map-based method and system for time-domain compression of optical high-dimensional data. This method maximizes the preservation of near-neighborhood information of the original data while leveraging the inherent advantages of time-domain optical computing in terms of speed, power consumption, and parallel processing, thereby enabling efficient processing of high-dimensional data. This technology further provides a new upgrade and expansion solution for time-domain optical computing systems, paving the way for future technological evolution. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Other features, objects and advantages of the present invention will become more apparent upon reading the detailed description of non-limiting embodiments with reference to the following drawings:

[0041] Figure 1 Schematic diagram of two-dimensional 0 to 3 order Hilbert curve.

[0042] Figure 2a 、 Figure 2b Schematic diagrams of two-dimensional Hilbert mapping and three-dimensional Hilbert mapping respectively.

[0043] Figure 3 This is a flow chart of the calculation method for time-domain compression of optical high-dimensional data.

[0044] Figure 4 Schematic diagram of an optical edge information extraction device.

[0045] Figure 5 Schematic diagram of the experimental results of first-order time-domain differential operation on multi-wavelength signals.

[0046] Figure 6 Schematic diagram of edge information extraction results for chest two-dimensional CT data.

[0047] Figure 7 Schematic diagram of the results of boundary surface information extraction from chest 3D CT data. DETAILED DESCRIPTION

[0048] The present invention will be described in detail below with reference to specific embodiments. The following examples will help those skilled in the art to further understand the present invention, but are not intended to limit the present invention in any form. It should be noted that, for those skilled in the art, several changes and improvements can be made without departing from the scope of the present invention. These all fall within the scope of protection of the present invention.

[0049] Example 1

[0050] The present invention provides a time-domain compression calculation method and system for optical high-dimensional data based on Hilbert mapping, aiming to solve the difficulties encountered by existing time-domain optical operation units when processing high-dimensional data, and the problem of loss of neighborhood information caused by the vectorized input and processing of high-dimensional data.

[0051] The Hilbert mapping-based optical high-dimensional data time-domain compression calculation method and system include: expanding the original high-dimensional data from memory into a one-dimensional vector based on its dimension and size according to a Hilbert curve of order n with the same dimension, and loading the vector onto the time-varying intensity of an optical carrier through a digital-to-analog converter to form a signal light. Subsequently, the signal light enters a basic time-domain optical operation unit for time-domain compression processing, including but not limited to weighted quadratic intensity modulation, or optical differential (or differential) operations for extracting the changing relationship of neighborhood information. Finally, the vector information carried by the one-dimensional time-domain signal is extracted and stored, and an inverse Hilbert mapping of the corresponding dimension is performed on it to restore it to high-dimensional feature information for subsequent artificial intelligence algorithms to perform judgment, segmentation, and other operations.

[0052] This invention maximizes the preservation of the original data's near-neighborhood information while leveraging the inherent advantages of time-domain optical computing in terms of speed, power consumption, and parallel processing, thereby enabling efficient processing of high-dimensional data. Preserving near-neighborhood information means that two or more data points that are adjacent in the original data dimension remain adjacent or close to each other after vectorization. This technology further provides new upgrade and expansion solutions for time-domain optical computing systems, paving the way for future technological evolution.

[0053] Specifically, the optical high-dimensional data time domain compression calculation method based on Hilbert mapping is as follows: Figures 1 to 7 Shown, including:

[0054] Fill or crop the original data according to the dimension and size of the original data to determine the data size and the Hilbert curve order of the corresponding dimension;

[0055] Perform Hilbert mapping on the original data to obtain a one-dimensional vector;

[0056] Specifically, the Hilbert curve order n determines the size of the Hilbert curve to be 2 n ×2 n ×2 h , where h=0 in the two-dimensional Hilbert curve and h=n in the three-dimensional Hilbert curve. When quantizing high-dimensional data, the original data should be filled with 0 or cropped to make its size=2 n , and then vectorize it with reference to the n-th order Hilbert curve of the same dimension as the original data.

[0057] The signal light is obtained by loading the reduced one-dimensional vector onto the time-varying dimension of the photon through a high-speed optical modulator.

[0058] The basic optical time domain operation unit performs time domain compression processing on the signal light to obtain a one-dimensional time domain signal;

[0059] The one-dimensional time domain signal is detected, stored and restored to high-dimensional data through the Hilbert inverse map.

[0060] The high-dimensional data information generally refers to two-dimensional data and three-dimensional data, such as images and three-dimensional voxels, point cloud information, etc. having two-dimensional and three-dimensional spatial relationships respectively.

[0061] The Hilbert map is proposed based on the Hilbert curve.

[0062] The Hilbert curve, a typical fractal geometry construct, can theoretically be extended to three-dimensional space through a continuous and infinitely subdivided sequence of discrete units in a limited two-dimensional space. The curve can also be extended to three-dimensional space through iterative construction. The connection points that sequentially form the Hilbert curve in two-dimensional or three-dimensional space also correspond to a two-dimensional or three-dimensional coordinate in a higher-dimensional space. The values ​​corresponding to the coordinate positions in the original data are stored in a one-dimensional vector according to the sequence of the Hilbert connection points, thus forming a Hilbert map.

[0063] The LS grammar of the Hilbert curve in two-dimensional space is expressed as:

[0064] Initial: X;

[0065] X=+YF-XFX-FY+;

[0066] Y=-XF+YFY+FX-;

[0067] Here, X represents clockwise subgraph drawing (when order n = 1), Y represents counterclockwise subgraph drawing, + represents upward direction, - represents downward direction, and F represents drawing connecting lines. By designing a recursive program based on this grammar, we can draw a two-dimensional Hilbert curve of the corresponding order. A similar grammar can also be used to construct a three-dimensional Hilbert curve.

[0068] Among them, the Hilbert curve can continuously pass through discrete points in two-dimensional or three-dimensional space. Taking the two-dimensional Hilbert curve as an example, the sequence number k of the Hilbert curve connection point corresponds to the coordinate (x k ,y k ), and construct a one-dimensional vector with the same length as the number of corresponding Hilbert connection points make Among them, O is the original data in the two-dimensional space, which realizes the vectorization of high-dimensional data, namely the Hilbert map.

[0069] The vector is modulated by time domain intensity The high-speed optical modulator driven by the digital-to-analog converter is loaded onto the time-varying intensity of the photon to form a signal light. Its light intensity in each symbol time can be expressed as: Where ω is the center frequency of the optical carrier, is the initial phase of the optical carrier.

[0070] Sending the signal light into the optical time domain basic operation unit for time domain compression processing, including: performing weighted or differential processing on the signal light to obtain a one-dimensional time domain signal;

[0071] The weighted processing includes: re-modulating the intensity of the signal light according to the weighted coefficient obtained by algorithm training, and performing a mathematical dot product operation on the intensity;

[0072] Specifically, a dot product operation is performed on the signal to achieve weighting, and an optical intensity modulator is used, or an adjustable attenuator is used to perform secondary modulation on the signal light: the time-varying transmittance of the optical modulator or adjustable attenuator used for weighting is set to in, The value of is learned by machine learning algorithm training, and the clock of the digital-to-analog converter driving the weighted optical modulator needs to be aligned with the clock of the digital-to-analog converter that realizes the signal input, and the symbol rate is consistent. Therefore, the intensity of the processed optical signal can be expressed as:

[0073] After being received and sampled by the light detector, the weighted vector is obtained

[0074] The differential processing includes: using a time domain differential (differential) operator composed of an optical filter device, such as a microring and a microring array, to detect the rising and falling edges of the signal light in the time domain;

[0075] Specifically, the signal light is sent to an optical first-order differential operator for time-domain differential operation: because the transmission spectrum of an ideal first-order differential operator for time-domain signals is mathematically "V"-shaped with the center at 0 frequency, the device performing the optical time-domain differential operation is composed of an optical filter with a similar "V"-shaped transmission spectrum, such as a silicon-based microring resonator with appropriate parameters, whose operating (center) frequency should be consistent with the center frequency ω of the light carriers.

[0076] If the intensity of two adjacent code elements in the time domain signal light is inconsistent, that is, a k ≠a k-1 When the optical first-order differential operator filters, the position of the time-varying rising or falling edge between the two code elements will form a spike pulse, otherwise the light intensity here will become 0. The intensity of the timing spike pulse is recorded and sampled in sequence using an optical detector to form a vector It records the vector The changes between elements, that is, their difference operations:

[0077] c k =abs(a k -a k-1 ), where abs(·) is the absolute value operation, and Compared to There is an element missing.

[0078] The vector Or a vector filled with 0 The kth element in is filled into the coordinates corresponding to the kth connection point on the selected Hilbert curve in the high-dimensional space, realizing the dimensionality increase of the processed vector data, that is, the Hilbert inverse mapping.

[0079] vector Actually, it is a vector The difference operation result of It is obtained by vectorizing the high-dimensional original data through the Hilbert map. The adjacent elements in the vector are actually adjacent in the high-dimensional space. After being restored to a high-dimensional space through the Hilbert inverse map, it records the difference between each data in the original data O and the adjacent data, and can obtain the edge information of the two-dimensional image data and the boundary surface information of the three-dimensional volume data.

[0080] The present invention also provides an optical high-dimensional data time domain compression calculation system based on Hilbert mapping. The optical high-dimensional data time domain compression calculation system based on Hilbert mapping can be implemented by executing the process steps of the optical high-dimensional data time domain compression calculation method based on Hilbert mapping. That is, those skilled in the art can understand the optical high-dimensional data time domain compression calculation method based on Hilbert mapping as a preferred implementation of the optical high-dimensional data time domain compression calculation system based on Hilbert mapping.

[0081] Example 2

[0082] Example 2 is a preferred example of Example 1

[0083] The invention also discloses an optical high-dimensional data boundary information extractor based on Hilbert mapping.

[0084] The device includes: a multi-wavelength signal source, an optical high-speed modulator array, a multi-wavelength optical first-order differential operator, and a photodetector array;

[0085] The optical high-speed modulator array is used to load the vector A obtained by Hilbert mapping high-dimensional information onto the time-varying intensity of the multi-wavelength signal source to obtain signal light. The multi-wavelength optical first-order differential operator then performs a time-domain differential operation on the signal light. Finally, the signal light is converted into an electrical signal by the photodetector array, sampled by an analog-to-digital converter, and stored to obtain a vector C.

[0086] The high-dimensional data boundary information extraction includes edge information extraction of two-dimensional image data and boundary surface information extraction of three-dimensional volume data.

[0087] The two-dimensional image data is converted into a one-dimensional vector through a two-dimensional Hilbert map.

[0088] The three-dimensional volume data is converted into a one-dimensional vector through a three-dimensional Hilbert map.

[0089] The multi-wavelength first-order differential operator array can expand computing power through frequency division multiplexing at the band level.

[0090] When processing data of a relatively large size, the device can divide the original data into blocks, perform Hilbert mapping on the divided data, and then modulate the data onto the time-varying intensities of photocarriers with different center frequencies.

[0091] The photon parallel time-domain first-order differential operator of the device is composed of a silicon-based micro-ring resonant cavity network with a "V"-shaped transmission spectrum line.

[0092] The detector array is a high-speed photodetector array.

[0093] Those skilled in the art will appreciate that, in addition to implementing the system and its various devices, modules, and units provided by the present invention in purely computer-readable program code, it is entirely possible to implement the same functions of the system and its various devices, modules, and units provided by the present invention in the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers, and embedded microcontrollers by logically programming the method steps. Therefore, the system and its various devices, modules, and units provided by the present invention can be considered a hardware component, and the devices, modules, and units included therein for implementing various functions can also be considered as structures within the hardware component; the devices, modules, and units for implementing various functions can also be considered as both software modules implementing the method and structures within the hardware component.

[0094] The above describes specific embodiments of the present invention. It should be understood that the present invention is not limited to the specific embodiments described above, and those skilled in the art may make various changes or modifications within the scope of the claims, which do not affect the essence of the present invention. The embodiments of this application and the features in the embodiments may be combined with each other in any manner unless there is a conflict.

Claims

1. An optical operation method for time-domain compression of high-dimensional data based on Hilbert mapping, characterized in that: include: The original high-dimensional data is expanded into a one-dimensional vector using the Hilbert mapping method. A high-speed optical modulator then loads the one-dimensional vector onto the time-varying intensity of an optical carrier to form a signal light. The basic time-domain optical operation unit performs time-domain compression processing on the optical signal to obtain a one-dimensional time-domain signal; The vector information carried in the one-dimensional time domain signal is extracted, and the vector information is subjected to Hilbert inverse mapping of the corresponding dimension to restore it to high-dimensional feature information.

2. The optical operation method for high-dimensional data time domain compression based on Hilbert map according to claim 1, characterized in that: The method of using the Hilbert mapping method to expand the original high-dimensional data into a one-dimensional vector includes: padding or clipping the original data according to the dimension and size of the original data to determine the data size and the Hilbert curve order of the corresponding dimension; Map the original data according to the order of the Hilbert curve to obtain a one-dimensional vector 3. The optical operation method for high-dimensional data time-domain compression based on Hilbert map according to claim 1, characterized in that: The basic time-domain optical operation unit performs time-domain compression processing on the optical signal to obtain a one-dimensional time-domain signal, including: Secondarily modulating the signal light using an optical intensity modulator or an adjustable attenuator; The time-varying transmittance of the optical intensity modulator or variable attenuator used as weight is set to The clock of the DAC driving the optical intensity modulator or variable attenuator needs to be aligned with the clock of the DAC that implements the signal input, and the symbol rate must be consistent. Therefore, the intensity of the processed optical signal is expressed as: After being received and sampled by the light detector, the weighted vector is obtained Where ω is the center frequency of the optical carrier, is the initial phase of the optical carrier.

4. The optical operation method for high-dimensional data time domain compression based on Hilbert map according to claim 1, characterized in that: The basic time-domain optical operation unit performs time-domain compression processing on the optical signal to obtain a one-dimensional time-domain signal, including: Use optical filters to detect the rising and falling edges of signal light in the time domain; Use an optical detector to record the intensity of the timing spike pulse in sequence and sample it to form a vector Among them, c k =abs(a k -a k-1 ), where abs(·) is the absolute value operation, and Compared to There is an element missing.

5. The optical operation method for high-dimensional data time-domain compression based on Hilbert mapping according to claim 4, characterized in that: The optical filter is an optical first-order differential operator, and is composed of an optical filter with a "V"-shaped transmission spectrum line.

6. The optical operation method for high-dimensional data time-domain compression based on Hilbert mapping according to claim 3 or 4, characterized in that: The vector information is subjected to the Hilbert inverse mapping of the corresponding dimension to restore it to high-dimensional feature information, including: vector Or a vector filled with 0 The kth element in is filled into the coordinates corresponding to the kth connection point on the selected Hilbert curve in the high-dimensional space, so as to realize the dimension increase of the processed vector data and the Hilbert inverse mapping.

7. The optical operation method for high-dimensional data time-domain compression based on Hilbert map according to claim 1, characterized in that: When the data size exceeds the preset condition, the original data is divided into blocks, the divided data is Hilbert mapped, and then modulated to the time-varying intensity of light carriers with different center frequencies to form multiple signal lights.

8. An optical computing system for high-dimensional data time-domain compression based on Hilbert mapping, characterized in that: include: The original high-dimensional data is expanded into a one-dimensional vector using the Hilbert mapping method. A high-speed optical modulator then loads the one-dimensional vector onto the time-varying intensity of an optical carrier to form a signal light. The basic time-domain optical operation unit performs time-domain compression processing on the optical signal to obtain a one-dimensional time-domain signal; The vector information carried in the one-dimensional time domain signal is extracted, and the vector information is subjected to Hilbert inverse mapping of the corresponding dimension to restore it to high-dimensional feature information.

9. The optical computing system for high-dimensional data time-domain compression based on Hilbert mapping according to claim 8, characterized in that: The method of using the Hilbert mapping method to expand the original high-dimensional data into a one-dimensional vector includes: padding or clipping the original data according to the dimension and size of the original data to determine the data size and the Hilbert curve order of the corresponding dimension; Map the original data in the order of the Hilbert curve to obtain a one-dimensional vector 10. The optical computing system for high-dimensional data time-domain compression based on Hilbert mapping according to claim 8, characterized in that: The step of performing time domain compression processing on the optical signal by the basic time domain optical operation unit to obtain a one-dimensional time domain signal comprises: performing weighted processing or differential processing on the optical signal by the basic time domain optical operation unit; The weighting process includes: Secondarily modulating the signal light using an optical intensity modulator or an adjustable attenuator; The time-varying transmittance of the optical intensity modulator or variable attenuator used as weight is set to The clock of the DAC driving the optical intensity modulator or variable attenuator needs to be aligned with the clock of the DAC that implements the signal input, and the symbol rate must be consistent. Therefore, the intensity of the processed optical signal is expressed as: After being received and sampled by the light detector, the weighted vector is obtained Where ω is the center frequency of the optical carrier, is the initial phase of the optical carrier; The differential processing includes: Use optical filters to detect the rising and falling edges of signal light in the time domain; Use an optical detector to record the intensity of the timing spike pulse in sequence and sample it to form a vector Among them, c k =abs(a k -a k-1 ), where abs(·) is the absolute value operation, and Compared to There is an element missing.

Images