A soft-hard cooperative raman microscope real-time spectrum signal processing method and system

Abstract

This invention discloses a hardware-software co-processing method for real-time spectral signal processing in Raman microscopy, belonging to the field of spectral signal processing technology. The method includes: establishing strict temporal constraints between physical displacement and cloud inference, and initiating asynchronous communication using the physical gaps in stage movement; simultaneously, deploying a deep learning model (F2P) pre-trained with a self-supervised masking strategy in the cloud. This pre-trained model utilizes the structural correlation between the Raman broadband background and narrowband characteristic peaks to directly perform real-time forward inference on the input single-frame noisy spectrum, achieving millisecond-level high-fidelity reconstruction of a single frame without any complex parameter tuning or the need for a clean control spectrum. This invention completely eliminates the physical lag in multi-frame processing and the general technical bias of supervised learning algorithms, achieving true "synchronization between acquisition and display," and significantly improving the real-time processing performance and cross-device automated deployment capabilities of in-situ Raman imaging.

Description

Technical Field

[0001] This invention relates to the field of signal processing technology at the intersection of spectral analysis and artificial intelligence, specifically to a real-time imaging noise reduction method and system for Raman microscopy based on a software-hardware co-training and self-supervised pre-training model, which aims to overcome the physical time constraints of single-point scanning. Background Technology

[0002] Raman spectroscopy, as a molecular vibrational spectroscopy detection technique, is widely used in real-time characterization systems such as cell life process monitoring and photoluminescence reactions of two-dimensional materials due to its non-invasive nature and rich fingerprint information. In practical Raman microscopy imaging, data is typically acquired continuously via point scanning or line-by-line scanning, placing high demands on temporal resolution. However, spontaneous Raman scattering signals are extremely weak and easily interfered with by ambient fluorescence, cosmic rays, and inherent detector noise during acquisition. Although configuring a high-sensitivity cooled CCD and extending the exposure time can improve the signal-to-noise ratio, this often comes at the cost of sacrificing temporal resolution, making it difficult to meet real-time imaging requirements. Therefore, introducing efficient noise reduction algorithms without changing the instrument hardware is crucial to overcoming physical limitations and achieving online signal processing and reconstruction.

[0003] Most existing real-time spectral noise reduction schemes are based on traditional filtering methods (such as bilateral filtering) to establish a sliding window. These methods rely on frequency separation and manual parameter settings (such as window width or basis function selection), which can easily lead to characteristic peak distortion in complex noise environments. Although they are easy to embed into Raman microscopy systems, the need for repeated parameter tuning across different instruments makes efficient automated deployment difficult. Deep learning methods have advantages in signal reconstruction, but they mostly follow an offline "acquisition first, processing later" model, unable to provide real-time feedback. Furthermore, their model training phase heavily relies on clean reference spectra (Ground Truth) to calculate reconstruction loss. However, in real-world physical or chemical dynamic monitoring experiments, obtaining completely noise-free, homologous experimental "true" data is physically impossible, making it difficult for traditional supervised learning models to achieve automated, universal deployment across unknown samples or different devices.

[0004] Therefore, how to break through the traditional offline mode of "acquiring first and then processing", build an architecture that can perform real-time processing through deep learning algorithms during the acquisition process, and at the same time ensure its automated deployment and versatility in different device environments, has become a key technical problem that urgently needs to be solved in the field of real-time Raman spectroscopy imaging. Summary of the Invention

[0005] The purpose of this invention is to solve the problems in the prior art.

[0006] The technical solution adopted by this invention to solve its technical problem is: to provide a hardware-software co-processing method for real-time spectral signal processing in Raman microscopy, comprising the following steps:

[0007] An asynchronous communication mechanism is established between the instrument and the computing end. After the instrument collects the spectrum at position t, it moves from position t to position t+1. During this process, the spectrum collected at position t is sent from the instrument to the computing end.

[0008] A pre-trained deep learning model F2P is deployed on the computing side, and single-frame forward inference is performed based on the noisy spectral data sent by the instrument to output the denoised reconstructed spectrum.

[0009] By combining asynchronous communication mechanism with deep learning model F2P, real-time imaging noise reduction of Raman microscopy is achieved.

[0010] Preferably, the asynchronous communication mechanism between the instrument and the computing end satisfies timing constraints:

[0011] ;

[0012] in, For the time it takes for data to travel to and from the cloud, This refers to the model's ultra-fast forward inference time in the cloud. This refers to the mechanical motion time for the displacement stage to move to the next pixel. For the spectral acquisition time of the next pixel; when the device is in The location has completed spectral acquisition, at this time The location's spectrum has been denoised and returned to the device.

[0013] Preferably, the step of deploying a pre-trained deep learning model F2P at the computing end, performing single-frame forward inference based on noisy spectral data sent from the instrument, and outputting a denoised reconstructed spectrum includes the following steps:

[0014] Obtain a large number of noisy Raman spectral samples y;

[0015] For each sample y, randomly select 10%–20% of the wavenumber points, replace their spectral intensity values ​​with Gaussian noise or zero values, and generate a masked spectrum ym;

[0016] Input ym into a neural network based on the Transformer architecture and output the reconstructed spectrum f(ym);

[0017] The mean square error (MSE) between f(ym) and the original y at the mask position is calculated as the loss function. The network parameters are optimized through backpropagation, so that the model learns the structural continuity between the broadband background and the narrow band characteristic peaks in the Raman spectrum.

[0018] Freeze the parameters of the trained model and deploy it as a lightweight model that only supports forward inference for real-time online noise reduction.

[0019] Preferably, the asynchronous communication mechanism is implemented through a background thread in the instrument control software. This thread immediately triggers data upload via HTTP or TCP / IP protocol after spectral acquisition is completed, without blocking the main control thread's motion control commands to the displacement stage.

[0020] Preferably, the computing end is a server deployed on a high-performance computing unit, using a cloud server or a local workstation, on which a RESTful inference service built based on the Python-based FastAPI or Flask framework runs, supporting concurrent reception of multi-channel spectral data streams.

[0021] Preferably, the method of combining asynchronous communication mechanism with deep learning model F2P to achieve real-time imaging noise reduction for Raman microscopy includes the following steps:

[0022] During Raman microscopy point scanning imaging, the instrument acquires spectra at the i-th scanning point;

[0023] After completing the spectral acquisition at the i-th scanning point, the displacement stage moves to the (i+1)-th scanning point. During the movement of the displacement stage, the noisy Raman spectral data of the i-th scanning point is simultaneously sent to the cloud computing server through an asynchronous communication mechanism.

[0024] After receiving the noisy spectral data, the cloud computing server calls the pre-trained deep learning model F2P to perform single-frame forward inference, outputs the denoised reconstructed spectrum, and returns the denoising result to the instrument before the displacement stage completes its movement to the (i+1)th point and begins the next exposure.

[0025] The instrument receives and displays the denoised reconstructed spectrum at point i, achieving real-time imaging denoising.

[0026] This invention also provides a hardware-software co-processing Raman microscope real-time spectral signal processing system, comprising:

[0027] The asynchronous communication mechanism establishment module establishes an asynchronous communication mechanism between the instrument and the computing end. During the process of moving from position t to position t+1 after the instrument has collected the spectrum at position t, the spectrum collected at position t will be sent from the instrument to the computing end.

[0028] The self-supervised model training module deploys a pre-trained deep learning model F2P on the computing end, performs single-frame forward inference based on noisy spectral data sent from the instrument, and outputs a denoised reconstructed spectrum.

[0029] The real-time imaging noise reduction module combines an asynchronous communication mechanism with the F2P deep learning model to achieve real-time imaging noise reduction for Raman microscopy.

[0030] The present invention also provides an electronic device, comprising:

[0031] One or more processors;

[0032] A storage device for storing one or more programs that, when executed by one or more processors, cause the one or more processors to perform any of the methods described above.

[0033] The present invention also provides a computer-readable storage medium having a computer program stored thereon that, when executed by a processor, implements any of the methods described above.

[0034] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements any of the methods described above.

[0035] The present invention has the following beneficial effects:

[0036] (1) Real-time performance with near-zero physical lag: Compared with the problem of several seconds of lag caused by caching multiple frames of data in existing technologies, the present invention uses an asynchronous processing architecture to cover the forward inference time in the inherent mechanical displacement gap, achieving near-zero latency at the single-point scanning level, eliminating the time lag of traditional offline processing, and realizing the instant acquisition and display of data.

[0037] (2) No truth reference required and cross-device generalization capability: Unlike traditional deep learning models that rely on physical masks and clean reference data, the F2P algorithm of this invention adopts a mask self-supervised learning strategy, combines the Transformer architecture, and utilizes the statistical inconsistency of noise, requiring only noisy experimental data for pre-training. This lightweight pre-trained model can be directly used for real-time inference, eliminating the dependence on clean experimental data and significantly reducing the difficulty of automated deployment. The invention will be further described in detail below with reference to the accompanying drawings and embodiments, but the invention is not limited to the embodiments. Attached Figure Description

[0038] Figure 1 This is a flowchart illustrating the steps of a hardware-software co-processing Raman microscope real-time spectral signal processing method according to an embodiment of the present invention.

[0039] Figure 2 This is a schematic flowchart of a hardware-software co-processing Raman microscope real-time spectral signal processing method according to an embodiment of the present invention;

[0040] Figure 3 The flowchart of the F2P deep learning algorithm proposed in this embodiment of the invention is shown below; where (a) is the mask self-supervised pre-training process, and (b) is the single-frame online ultra-fast inference process after deploying the pre-trained model.

[0041] Figure 4 This invention provides a performance comparison between the F2P deep learning algorithm proposed in this embodiment and existing open-source algorithms.

[0042] Figure 5 This is a schematic diagram of a hardware-software co-processing Raman microscope real-time spectral signal processing system according to an embodiment of the present invention;

[0043] Figure 6 This is a schematic diagram of the hardware structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0044] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this invention, and not all of them. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0045] See Figure 1 and Figure 2 As shown in the figure, an embodiment of the present invention provides a real-time spectral signal processing method for Raman microscopy using a combination of hardware and software methods, comprising the following steps:

[0046] S101, establish an asynchronous communication mechanism between the instrument and the computing end. After the instrument collects the spectrum at position t, during the process of moving from position t to position t+1, the spectrum collected at position t will be sent from the instrument to the computing end.

[0047] S102 deploys a pre-trained deep learning model F2P on the computing end, performs single-frame forward inference based on noisy spectral data sent from the instrument end, and outputs a denoised reconstructed spectrum.

[0048] S103 combines an asynchronous communication mechanism with the F2P deep learning model to achieve real-time imaging noise reduction in Raman microscopy.

[0049] Specifically, S101 designs an asynchronous processing architecture that integrates end-to-cloud collaboration. Utilizing the hardware action gap (i.e., the time required for the stage to move from position t to t+1) during point scanning imaging using a Raman microscope, an asynchronous communication mechanism is established between the instrument and the computing end (such as a cloud server or local workstation). The system operation is driven by the following strict time constraint equation:

[0050] ;

[0051] in, For the time it takes for data to travel to and from the cloud, This refers to the model's ultra-fast forward inference time in the cloud. This refers to the mechanical motion time for the displacement stage to move to the next pixel. The time for spectral acquisition (exposure) of the next pixel. When the device is... The location has completed spectral acquisition, at this time The location's spectrum has been denoised and returned to the device.

[0052] Specifically, the instrument (client) controls the Raman microscope to perform point scanning. An asynchronous HTTP request module is integrated into the control software. Once the detector (e.g., CCD) has read a frame of spectral data, the software immediately instructs the stage to move to the next scanning point without waiting for processing results, and simultaneously packages and sends the current spectral data to the server in a background thread. The computing end (backend) is deployed on a high-performance computing unit (cloud or local workstation). An inference service is built using a Python web framework (e.g., Flask / FastAPI) and a pre-loaded deep learning model. This service listens on a specific port; once it receives spectral data, it immediately feeds it into the model for forward inference and returns the denoised data in JSON or binary stream format.

[0053] Specifically, the asynchronous processing flow is as follows: Figure 2 As shown in (a), the definition is... Index of the currently scanned pixel:

[0054] Step S201: The instrument is in position Complete the exposure and acquisition to obtain the raw spectrum. The collection time is .

[0055] Step S202: The instrument's main thread immediately commands the displacement stage orientation position. The movement takes a time of [time missing]. .

[0056] Step S203: While the instrument is moving, the background thread on the instrument side will... Send to the processing end. The processing time at the processing end is... .

[0057] Step S204: Round-trip transmission time s As long as the conditions are met. This means that the data transmission and processing time is less than or equal to the sum of the instrument movement and the next exposure time, allowing users to see the real-time imaging and noise reduction effects on the software. In this embodiment, the entire closed-loop response time is controlled to around 40ms, and the real-time processing effect is demonstrated using graphene material as an example. Figure 2 As shown in (b).

[0058] Specifically, in S102, a mask self-supervised learning denoising algorithm (Fast2Peak, or F2P for short) is proposed. This addresses the challenges of extremely fast inference within the aforementioned time constraints. To meet the requirements, this algorithm is divided into two stages: offline pre-training and online high-speed inference. (1) Pre-training based on a large amount of noisy data: The input noisy spectral data is randomly masked; a neural network model is constructed and trained with the original noisy spectrum itself as the reconstruction target. The model predicts and reconstructs the masked data points by utilizing the context information of the unmasked data. Since random noise is usually not correlated in the spectral sequence, while real Raman signals have structural correlation, the model eventually learns to reconstruct a clean spectral structure. (2) Online inference deployed in the cloud: The model pre-trained by the self-supervised masking strategy is directly deployed in the cloud to perform high-speed forward inference on the single frame of noisy spectrum asynchronously transmitted from the instrument. Under the condition that a clean reference spectrum is not required, the single frame high-fidelity signal reconstruction at the millisecond level is achieved.

[0059] Specifically, in this embodiment of the invention, the mask self-supervised pre-training is as follows: Figure 3 As shown in (a), the input is a batch of noisy spectra y. According to a preset ratio (e.g., 15%), wavenumber points in the spectrum are randomly selected, and their intensity values ​​are replaced with random noise to generate the masked spectrum y. m . y m Input neural network (Transformer architecture, see...) Figure 3 (c) Train the array to output the reconstructed spectrum f(y). m f(y) m The loss (mean squared error, MSE) at the mask position is calculated and optimized using the original y. During this process, the model spontaneously learns the structural continuity between the Raman broadband background and the narrowband peak. Online ultra-fast forward inference is performed as follows: Figure 3 As shown in (b), the pre-trained model completed in step S1 is weighted and deployed to the computing terminal. During real-time point scanning acquisition, the asynchronously transmitted single-frame noisy spectrum is directly input into the pre-trained model. The model does not require any backpropagation calculations and only requires milliseconds of forward inference time. This allows the output of a reconstructed high signal-to-noise ratio spectrum. Benchmark tests show that the F2P algorithm outperforms all currently available open-source algorithms; see the comparison results below. Figure 4 As shown. Details of the Transformer architecture are as follows: Figure 3As shown in (c), the input spectrum is segmented into serialized tokens by the Patchify operation, and then superimposed with positional encoding to form input embeddings, which are then fed into the backbone network. The backbone network consists of 8 stacked Transformer encoders, each of which integrates layer normalization, multi-head self-attention, and a feedforward network. After feature extraction by the backbone network, the output vector is processed by the GELU activation function and a nonlinear mapping layer, and finally reassembled by Unpatchify to restore the high-dimensional features to the original spectral signal.

[0060] See Figure 5 As shown, this embodiment of the invention provides a hardware-software co-processing Raman microscopy real-time spectral signal processing system, which is similar to... Figure 1 The illustrated method embodiments correspond to various electronic devices and can be specifically applied to them. The system includes:

[0061] The asynchronous communication mechanism establishment module 501 establishes an asynchronous communication mechanism between the instrument and the computing end. After the instrument collects the spectrum at position t, it moves from position t to position t+1. During this process, the spectrum collected at position t is sent from the instrument to the computing end.

[0062] The self-supervised model training module 502 deploys a pre-trained deep learning model F2P on the computing end, performs single-frame forward inference based on the noisy spectral data sent by the instrument, and outputs a denoised reconstructed spectrum.

[0063] The real-time imaging noise reduction module 503 combines an asynchronous communication mechanism with the deep learning model F2P to achieve real-time imaging noise reduction for Raman microscopy.

[0064] See Figure 6 The diagram shown is a schematic representation of the hardware structure of an electronic device provided in an embodiment of the present invention. Figure 6 As shown, the electronic device of this embodiment includes a processor 601 and a memory 602; wherein the memory 602 is used to store computer execution instructions; and the processor 601 is used to execute the computer execution instructions stored in the memory to implement the various steps performed by the electronic device in the above embodiment. For details, please refer to the relevant descriptions in the foregoing method embodiments.

[0065] Alternatively, the memory 602 can be either standalone or integrated with the processor 601.

[0066] When the memory 602 is set up independently, the electronic device also includes a bus 603 for connecting the memory 602 and the processor 601.

[0067] This invention also provides a computer storage medium storing computer execution instructions, which, when executed by a processor, implement the method described above.

[0068] This invention also provides a computer program product, including a computer program that, when executed by a processor, implements the above-described method.

[0069] In the embodiments provided by this invention, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be indirect coupling or communication connection through some interfaces, devices, or modules, and may be electrical, mechanical, or other forms.

[0070] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to implement the solution of this embodiment according to actual needs.

[0071] Furthermore, the functional modules in the various embodiments of this invention can be integrated into one processing unit, or each module can exist physically separately, or two or more modules can be integrated into one unit. The unit integrating the above modules can be implemented in hardware or in the form of hardware plus software functional units.

[0072] The integrated modules described above, implemented as software functional modules, can be stored in a computer-readable storage medium. These software functional modules, stored in a storage medium, include several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) or processor to execute some steps of the methods of the various embodiments of this application.

[0073] It should be understood that the aforementioned processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), etc. A general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in this invention can be directly manifested as being executed by a hardware processor, or executed by a combination of hardware and software modules within the processor.

[0074] The memory may include high-speed RAM, and may also include non-volatile storage (NVM), such as at least one disk storage device, and may also be a USB flash drive, external hard drive, read-only memory, disk or optical disc, etc.

[0075] The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc. For ease of illustration, the buses shown in the accompanying drawings are not limited to a single bus or a single type of bus.

[0076] The aforementioned storage medium can be implemented from any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. The storage medium can be any available medium accessible to general-purpose or special-purpose computers.

[0077] An exemplary storage medium is coupled to a processor, enabling the processor to read information from and write information to the storage medium. Alternatively, the storage medium can be an integral part of the processor. Both the processor and the storage medium can reside in an Application Specific Integrated Circuit (ASIC). Alternatively, the processor and storage medium can exist as discrete components in an electronic device or host device.

[0078] Those skilled in the art will understand that all or part of the steps of the above-described method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When executed, the program performs the steps of the above-described method embodiments; and the aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disks, or optical disks.

[0079] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for real-time spectral signal processing in Raman microscopy with a combination of hardware and software, characterized in that, Includes the following steps: An asynchronous communication mechanism is established between the instrument and the computing end. After the instrument collects the spectrum at position t, it moves from position t to position t+1. During this process, the spectrum collected at position t is sent from the instrument to the computing end. A pre-trained deep learning model F2P is deployed on the computing side, and single-frame forward inference is performed based on the noisy spectral data sent by the instrument to output the denoised reconstructed spectrum. By combining asynchronous communication mechanism with deep learning model F2P, real-time imaging noise reduction of Raman microscopy is achieved.

2. The real-time spectral signal processing method for Raman microscopy with hardware and software coordination according to claim 1, characterized in that, The asynchronous communication mechanism between the instrument and the computing end satisfies the timing constraints: ； in, For the time it takes for data to travel to and from the cloud, This refers to the model's ultra-fast forward inference time in the cloud. This refers to the mechanical motion time for the displacement stage to move to the next pixel. For the spectral acquisition time of the next pixel; when the device is in The location has completed spectral acquisition, at this time The location's spectrum has been denoised and returned to the device.

3. The real-time spectral signal processing method for Raman microscopy with hardware and software coordination according to claim 1, characterized in that, The process of deploying a pre-trained deep learning model F2P on the computing end, performing single-frame forward inference based on noisy spectral data sent from the instrument, and outputting a denoised reconstructed spectrum includes the following steps: Obtain a large number of noisy Raman spectral samples y; For each sample y, randomly select 10%–20% of the wavenumber points, replace their spectral intensity values ​​with Gaussian noise or zero values, and generate a masked spectrum ym; Input ym into a neural network based on the Transformer architecture and output the reconstructed spectrum f(ym); The mean square error (MSE) between f(ym) and the original y at the mask position is calculated as the loss function. The network parameters are optimized through backpropagation, so that the model learns the structural continuity between the broadband background and the narrow band characteristic peaks in the Raman spectrum. Freeze the parameters of the trained model and deploy it as a lightweight model that only supports forward inference for real-time online noise reduction.

4. The real-time spectral signal processing method for Raman microscopy with hardware and software coordination according to claim 1, characterized in that, The asynchronous communication mechanism is implemented through a background thread in the instrument control software. This thread immediately triggers data upload via HTTP or TCP / IP protocol after the spectral acquisition is completed, without blocking the main control thread's motion control commands to the displacement stage.

5. The method for real-time spectral signal processing of Raman microscopy with hardware and software coordination according to claim 1, characterized in that, The computing end is a server deployed in a high-performance computing unit, using a cloud server or local workstation, on which a RESTful inference service built based on Python's FastAPI or Flask framework runs, supporting concurrent reception of multi-channel spectral data streams.

6. The method for real-time spectral signal processing of Raman microscopy with hardware and software coordination according to claim 1, characterized in that, The method of combining asynchronous communication mechanism and deep learning model F2P to achieve real-time imaging noise reduction in Raman microscopy includes the following steps: During Raman microscopy point scanning imaging, the instrument acquires spectra at the i-th scanning point; After completing the spectral acquisition at the i-th scanning point, the displacement stage moves to the (i+1)-th scanning point. During the movement of the displacement stage, the noisy Raman spectral data of the i-th scanning point is simultaneously sent to the cloud computing server through an asynchronous communication mechanism. After receiving the noisy spectral data, the cloud computing server calls the pre-trained deep learning model F2P to perform single-frame forward inference, outputs the denoised reconstructed spectrum, and returns the denoising result to the instrument before the displacement stage completes its movement to the (i+1)th point and begins the next exposure. The instrument receives and displays the denoised reconstructed spectrum at point i, achieving real-time imaging denoising.

7. A hardware-software co-processing Raman microscope real-time spectral signal processing system, characterized in that, include: The asynchronous communication mechanism establishment module establishes an asynchronous communication mechanism between the instrument and the computing end. During the process of moving from position t to position t+1 after the instrument has collected the spectrum at position t, the spectrum collected at position t will be sent from the instrument to the computing end. The self-supervised model training module deploys a pre-trained deep learning model F2P on the computing end, performs single-frame forward inference based on noisy spectral data sent from the instrument, and outputs a denoised reconstructed spectrum. The real-time imaging noise reduction module combines an asynchronous communication mechanism with the F2P deep learning model to achieve real-time imaging noise reduction for Raman microscopy.

8. An electronic device, characterized in that, include: One or more processors; A storage device for storing one or more programs, which, when executed by one or more processors, cause the one or more processors to perform the method as described in any one of claims 1-6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the method as described in any one of claims 1-6.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the method as described in any one of claims 1-6.

Images