Signal reconstruction method and device based on neural network, equipment and storage medium

Abstract

The invention relates to the technical field of artificial intelligence, and discloses a signal reconstruction method and device based on a neural network, computer equipment and a storage medium, and the method comprises the steps: obtaining an interference signal collected by a linear wave number SD-OCT system when the distance between a sample arm reflector and a zero optical path difference position is z depth; carrying out direct current removal processing on the interference signal, and carrying out Fourier transform on the interference signal after the direct current removal processing; extracting the interference signal after Fourier transform to obtain a first airspace signal; determining a second airspace signal based on the first airspace signal and a preset amplification coefficient; and performing model training based on the interference signal, the second spatial domain signal and an initial convolutional neural network to obtain a trained convolutional neural network, the convolutional neural network being used for reconstructing the interference signal. According to the invention, end-to-end learning is carried out on an SD-OCT reconstruction task in a network training mode, so that a convolutional neural network capable of realizing high-quality SD-OCT signal reconstruction is obtained.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular, to a signal reconstruction method, apparatus, device, and storage medium based on a neural network. Background Technique

[0002] Optical Coherence Tomography (OCT) has the characteristics of high resolution, fast imaging, non-contact, and non-invasiveness, and has been widely used in medical fields such as ophthalmology, dermatology, oncology, etc., and has developed into an important optical imaging technology. OCT is mainly based on the theory of low-coherence light interference and consists of a light source, a Michelson interferometer optical path, a detector, etc. During the imaging process of the OCT system, the light emitted by the light source is first split into two beams by a fiber optic coupler and then reaches the sample and the reference flat mirror through the sample arm and the reference arm respectively. Then, the light in the sample arm is backscattered by the sample and returns to the fiber optic coupler together with the light reflected back by the flat mirror in the reference arm. When their optical path difference is less than the coherence length of the light source, an interference signal will be generated. OCT technology has developed from the time domain to the frequency domain. Compared with time-domain OCT technology, spectral-domain OCT (SD-OCT) has obvious advantages in terms of imaging speed, signal-to-noise ratio, and sensitivity. However, with the continuous improvement of application requirements, the imaging performance requirements for SD-OCT are getting higher and higher, and higher resolution and deeper imaging depth are required during imaging. The signal formation process of the SD-OCT system is affected by imaging factors such as non-linear k-sampling, spectrometer pixel integration effect, and dispersion. Non-linear k-sampling means that after the frequency-domain interference signal is received by the spectrometer and subjected to spectral splitting processing, the light of each wavelength corresponds to different pixel points of the spectrometer respectively, and then forms an approximately linear arrangement distribution in the wavelength space.

[0003] However, due to the non - linear relationship between the wave number and the wavelength, the uniform distribution in the wavelength space is non - uniform in the wave number space. When performing a Fourier transform on the frequency - domain interference signal to reconstruct an image, the non - uniform distribution of the spectrum in the wave number space will cause the chirp phenomenon in the signal, resulting in a decrease in the axial resolution of the system and the broadening of the point - spread function. Chromatic dispersion refers to the fact that when different optical elements are used or the optical path lengths are different between the sample arm and the reference arm, chromatic dispersion mismatch will occur. In addition, the sample itself will also introduce chromatic dispersion because when light propagates in the sample at different depths, the path lengths are different, resulting in different degrees of chromatic dispersion for the signals at different depths. This chromatic dispersion mismatch and sample chromatic dispersion will cause the interference signal output by the system to be broadened and distorted after being converted into the spatial - domain signal, resulting in poor imaging quality. The pixel integration effect means that the detector pixels in the spectrometer have a certain width. Therefore, each pixel will not only detect the light corresponding to its wavelength but also may detect the light of adjacent wavelengths, resulting in the leakage of the spectral signal and the decrease of the signal amplitude with depth. Therefore, when SD - OCT is affected by various imaging factors, its imaging performance is restricted by the characteristic of the sensitivity decreasing with depth. Summary of the Invention

[0004] Based on this, in view of the technical problem of the poor imaging effect of the interference signal in the prior art, a signal reconstruction method, device, equipment and storage medium based on a neural network are proposed.

[0005] In a first aspect, a signal reconstruction method based on a neural network is provided. The method includes:

[0006] Obtain the interference signal when the distance of the sample - arm mirror from the zero optical path difference is z depth, collected by a linear - wave - number SD - OCT system;

[0007] Perform a direct - current removal process on the interference signal, and perform a Fourier transform on the interference signal after the direct - current removal process;

[0008] Extract the interference signal after the Fourier transform to obtain a first spatial - domain signal;

[0009] Determine a second spatial - domain signal based on the first spatial - domain signal and a preset amplification factor;

[0010] Based on the interference signal, the second spatial - domain signal and an initial convolutional neural network, perform model training to obtain a trained convolutional neural network, where the convolutional neural network is used to reconstruct the interference signal.

[0011] In a second aspect, a signal reconstruction device based on a neural network is provided. The device includes:

[0012] An acquisition module, configured to acquire an interference signal when the distance between the sample arm mirror and the zero optical path difference is z depth, which is acquired by a linear wavenumber SD-OCT system;

[0013] A processing module, configured to perform DC removal processing on the interference signal and perform Fourier transform on the interference signal after DC removal processing;

[0014] An extraction module, configured to extract the interference signal after Fourier transform to obtain a first spatial domain signal;

[0015] A determination module, configured to determine a second spatial domain signal based on the first spatial domain signal and a preset amplification factor;

[0016] A training module, configured to perform model training based on the interference signal, the second spatial domain signal, and an initial convolutional neural network to obtain a trained convolutional neural network, where the convolutional neural network is used to reconstruct the interference signal.

[0017] In a third aspect, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of the above-mentioned neural network-based signal reconstruction method are implemented.

[0018] In a fourth aspect, a computer-readable storage medium is provided. The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the above-mentioned neural network-based signal reconstruction method are implemented.

[0019] The neural network-based signal reconstruction method proposed by the present invention obtains an interference signal when the distance between the sample arm mirror and the zero optical path difference is z depth, which is acquired by a linear wavenumber SD-OCT system. Then, DC removal processing is performed on the interference signal, and Fourier transform is performed on the interference signal after DC removal processing. Then, the interference signal after Fourier transform is extracted to obtain a first spatial domain signal. Then, based on the first spatial domain signal and a preset amplification factor, a second spatial domain signal is determined. Finally, model training is performed based on the interference signal, the second spatial domain signal, and an initial convolutional neural network to obtain a trained convolutional neural network, where the convolutional neural network is used to reconstruct the interference signal. The present invention performs end-to-end learning on the SD-OCT reconstruction task through network training, so as to obtain a convolutional neural network that can achieve high-quality SD-OCT signal reconstruction. After the model is trained, by directly inputting the original OCT spectral interference signal data, an accurate reconstruction result can be output. The reconstruction method based on the convolutional neural network does not need to change the OCT hardware system, and is of great significance in improving imaging performance and reducing costs. Description of the Drawings

[0020] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the accompanying drawings required in the description of the embodiments or the prior art. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other accompanying drawings can be obtained based on these drawings.

[0021] Among them:

[0022] Figure 1 It is an application environment diagram of a signal reconstruction method based on a neural network in an embodiment;

[0023] Figure 2 It is a flowchart of a signal reconstruction method based on a neural network in an embodiment;

[0024] Figure 3 It is a flowchart of a signal reconstruction method based on a neural network in an embodiment;

[0025] Figure 4 It is a flowchart of a signal reconstruction method based on a neural network in an embodiment;

[0026] Figure 5 It is a schematic diagram of the first reconstruction result of a signal reconstruction method based on a neural network in an embodiment;

[0027] Figure 6 It is a schematic diagram of the second reconstruction result of a signal reconstruction method based on a neural network in an embodiment;

[0028] Figure 7 It is a schematic diagram of the third reconstruction result of a signal reconstruction method based on a neural network in an embodiment;

[0029] Figure 8 It is a structural block diagram of a signal reconstruction device based on a neural network in an embodiment;

[0030] Figure 9 It is a structural block diagram of a computer device in an embodiment;

[0031] Figure 10 It is a structural block diagram of a computer device in another embodiment. Detailed implementation manners

[0032] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the technical field to which this application belongs; the terms used in the specification of this application are only for the purpose of describing specific embodiments and are not intended to limit this application; the terms "including" and "having" and any variations thereof in the specification and claims of this application and the above drawings are intended to cover non-exclusive inclusion. The terms "first", "second", etc. in the specification and claims of this application or the above drawings are used to distinguish different objects and not to describe a specific order.

[0033] Reference herein to "an embodiment" means that a particular feature, structure, or characteristic described in connection with the embodiment can be included in at least one embodiment of the application. The phrase appears in various places in the specification and does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment mutually exclusive of other embodiments. Those skilled in the art will explicitly and implicitly understand that the embodiments described herein can be combined with other embodiments.

[0034] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts fall within the protection scope of the present invention.

[0035] The signal reconstruction method based on a neural network provided by the embodiments of the present invention can be applied, for example, in Figure 1In the application environment, the client 110 communicates with the server 120 through a network. The server 120 can receive and obtain, through the client 110, the interference signal when the distance of the sample arm mirror from the zero optical path difference is at a depth of z by using a linear wavenumber SD-OCT system. Then, the interference signal is subjected to a DC removal process, and the DC-removed interference signal is subjected to a Fourier transform. Then, the interference signal after the Fourier transform is extracted to obtain a first spatial domain signal. Then, based on the first spatial domain signal and a preset amplification factor, a second spatial domain signal is determined. Finally, based on the interference signal, the second spatial domain signal, and an initial convolutional neural network, model training is performed to obtain a trained convolutional neural network, where the convolutional neural network is used to reconstruct the interference signal. The present invention performs end-to-end learning on the SD-OCT reconstruction task through network training, thereby obtaining a convolutional neural network that can achieve high-quality SD-OCT signal reconstruction. After the model is trained, by directly inputting the original OCT spectral interference signal data, an accurate reconstruction result can be output. This reconstruction method based on a convolutional neural network does not require any changes to the OCT hardware system and is of great significance in improving imaging performance and reducing costs. Among them, the client 110 can be, but is not limited to, various personal computers, laptop computers, smart phones, tablet computers, and portable wearable devices. The server 120 can be implemented by an independent server or a server cluster composed of multiple servers. The present invention will be described in detail below through specific embodiments.

[0036] Please refer to Figure 2 as shown in Figure 2 a flowchart of a signal reconstruction method based on a neural network provided by an embodiment of the present invention, including the following steps:

[0037] Step S101: Obtain the interference signal when the distance of the sample arm mirror from the zero optical path difference is at a depth of z by using a linear wavenumber SD-OCT system;

[0038] For example, data acquisition is performed by using a linear wavenumber SD-OCT system, which is equipped with a high-speed linear array camera and a linear wavenumber spectrometer, and the mirror reflection interference signal data at different depth positions is acquired.

[0039] Step S102: Perform a DC removal process on the interference signal, and perform a Fourier transform on the DC-removed interference signal;

[0040] First, a DC removal process is performed on the acquired interference signal. Subsequently, it is converted into a spatial domain signal through a Fourier transform, and a logarithmic transform process is further performed to enhance the visualization effect of the signal. The processed signal is as Figure 3As shown, it can be observed that as the optical path difference increases, the peak intensity of the specular reflection signal gradually decreases, decreasing by 17.96 dB at approximately 4.7 mm. This phenomenon indicates that although the system uses a linear wavenumber spectrometer, making the acquired interference spectrum linearly distributed in the wavenumber space, due to factors such as pixel integration effect and dispersion, the signal sensitivity still has the problem of too fast attenuation rate and still needs to be further optimized. Neural network technology can be used to slow down the signal attenuation and achieve reconstruction to improve the overall imaging performance of the system.

[0041] Step S103: Extract the interferometric signal after Fourier transform to obtain the first spatial domain signal;

[0042] Specifically, use the linear wavenumber SD-OCT system to collect data of 731 specular reflection signals at different depth positions, perform DC removal processing on the acquired interferometric signal, and take the first half after Fourier transform as the first spatial domain signal because the second half is a mirror image of the first half and does not contain additional useful information.

[0043] Step S104: Determine the second spatial domain signal based on the first spatial domain signal and a preset amplification factor;

[0044] Specifically, multiply the effective signal region (the first spatial domain signal) after Fourier transform of the interferometric signal at the z depth by the corresponding coefficient for amplification, amplifying it to the size corresponding to the signal closer to zero optical path difference (less affected by various imaging factors and more ideal), thereby obtaining the second spatial domain signal.

[0045] Step S105: Perform model training based on the interferometric signal, the second spatial domain signal, and the initial convolutional neural network to obtain a trained convolutional neural network, where the convolutional neural network is used to reconstruct the interferometric signal.

[0046] In one embodiment, the step of determining the second spatial domain signal based on the first spatial domain signal and a preset amplification factor includes: Step 201: Multiply the first spatial domain signal by the corresponding coefficient for amplification to obtain the second spatial domain signal;

[0047] Step 202: Determine the target output based on the real part and the imaginary part in the imaginary number of the second spatial domain signal.

[0048] Specifically, take the real part in the imaginary number of the second spatial domain signal as the first 1024 data, and the imaginary part as the last 1024 data. The total output is 2048 data, which is used as the target output for network training.

[0049] In one embodiment, the step of performing model training based on the interferometric signal, the second spatial domain signal, and the initial convolutional neural network to obtain a trained convolutional neural network, where the convolutional neural network is used to reconstruct the interferometric signal includes:

[0050] Step 301: Input the interference signal into the initial convolutional neural network to obtain reconstructed data;

[0051] Step 302: Calculate the loss function based on the reconstructed data and the target output;

[0052] Calculate the error between the reconstructed data and the target output to obtain the loss function. For example, the MAE loss function is adopted.

[0053] Step 303: Update the parameters of the initial convolutional neural network based on the loss function to obtain a trained convolutional neural network.

[0054] As an example, to construct the training set and validation set of the convolutional neural network, first, 696 mirror reflection signals are selected from 731 different mirror positions. From these individual signals, the signals are selected for pairwise combination. At the same time, the input and output are multiplied by coefficients for corresponding scaling (the scaling coefficients range from 0.2 to 1, with an interval of 0.2, a total of 5 coefficients) to expand the amount of training data. In this way, a training set containing 42,930 A-lines is constructed. These data cover the combination situations of different reflectivities and mirror positions, providing rich samples for the training of the neural network.

[0055] Similarly, to verify the generalization ability of the model and prevent overfitting, the mirror reflection signals at the remaining 35 positions are selected and combined in the ways of single, pairwise combination, and triple combination. Similarly, they are multiplied by coefficients for corresponding scaling processing (the scaling coefficients range from 0.1 to 0.9, with an interval of 0.2, a total of 5 coefficients). Finally, a validation set containing 8,950 A-lines is obtained. The setting of the validation set prevents overfitting during the model training process. The setting of the validation set can not only effectively monitor the performance of the model during the training process but also ensure that the model still has good reconstruction ability when facing unseen data. Through this designed method for constructing the data set, the neural network can better learn and identify the combined features of different reflectivities and mirror reflection signals, thereby significantly improving its performance and robustness in the OCT signal reconstruction task.

[0056] In one embodiment, the initial convolutional neural network includes successive convolutional modules, a flattening layer, and a fully connected layer. The convolutional module includes a convolutional layer, an activation layer process, and a normalization layer connected in sequence.

[0057] In one embodiment, the initial convolutional neural network includes a first convolutional layer, a first activation layer process, a first normalization layer, a second convolutional layer, a second activation layer process, a second normalization layer, a third convolutional layer, a third activation layer process, a third normalization layer, a fourth convolutional layer, a fourth activation layer process, a fourth normalization layer, a flattening layer, and a fully connected layer connected in sequence.

[0058] As an example, a one-dimensional convolutional neural network (1D CNN) is designed as the core architecture, and its structure consists of multiple key components, including a convolutional layer, a batch normalization layer (BN), a rectified linear unit activation layer (ReLU), and a fully connected layer. The convolutional kernel size of the convolutional layer is set to 3, and the stride is 1, which is used to extract local features from the input signal; the batch normalization layer is used to accelerate the training process and improve the stability of the model; the ReLU activation function introduces non-linearity and enhances the expressive ability of the network; during the training process, the number of training epochs of all models is uniformly set to 400 to ensure that the model can fully learn the data features. The Adam optimizer is selected, and its initial learning rate is set to 0.001. Combined with the cosine annealing learning rate decay strategy, the learning rate is dynamically adjusted to enable the model to converge quickly in the initial stage of training and fine-tune in the later stage to avoid falling into local optima. To further improve the training efficiency, the batch size is set to 64 to make full use of hardware resources and accelerate the training process. In the selection of the loss function, MAE is used as the training loss function. The final fully connected layer maps the extracted features to the target output space to complete the signal reconstruction task. As Figure 4 shown.

[0059] Refer to Figure 5 , Figure 5 which shows the reconstruction results of the convolutional neural network of the present invention. Among them, (A) is after direct demodulation by the original SD-OCT system. (B) is the reconstruction result of the convolutional neural network. It can be found that the reconstruction results of the convolutional neural network show high accuracy in the reconstruction of specular reflection signals at different depths and effectively slow down the signal attenuation rate. The maximum drop value of the signal reconstructed by the convolutional neural network is only 4.88 dB, which is significantly better than the maximum drop value of the directly demodulated signal of the system (17.96 dB), with an improvement of 13.08 dB. This result indicates that the convolutional neural network of the present invention can not only reconstruct signals more accurately but also effectively slow down signal attenuation, thus significantly improving the imaging quality of the system.

[0060] Refer to Figure 6 and Figure 7 , Figure 6 and Figure 7 respectively show the generalization results of the present network in the reconstruction of tape and skin signals. Among them, A is the source image directly demodulated by the system acquisition; B is the image reconstructed by the network. It can be seen that the present network can generalize to tape and skin, achieve accurate reconstruction, and effectively enhance the signal intensity of the sample.

[0061] Please refer to Figure 8 shown. In one embodiment, a signal reconstruction device based on a neural network is provided, and the device includes:

[0062] An acquisition module 10 for acquiring an interference signal when the sample arm mirror is at a z-depth from the zero optical path difference;

[0063] A processing module 20 for performing DC removal processing on the interference signal and performing Fourier transform on the interference signal after DC removal processing;

[0064] An extraction module 30 for extracting the interference signal after Fourier transform to obtain a first spatial domain signal;

[0065] A determination module 40 for determining a second spatial domain signal based on the first spatial domain signal and a preset amplification factor;

[0066] A training module 50 for performing model training based on the interference signal, the second spatial domain signal, and an initial convolutional neural network to obtain a trained convolutional neural network, where the convolutional neural network is used to reconstruct the interference signal.

[0067] In one embodiment, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as Figure 9 shown. The computer device includes a processor, a memory, a network interface, and a database connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile and / or volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external client through a network connection. When the computer program is executed by the processor, it realizes the functions or steps on the server side of a signal reconstruction method based on a neural network.

[0068] In one embodiment, a computer device is provided. The computer device may be a client, and its internal structure diagram may be as Figure 10 shown. The computer device includes a processor, a memory, a network interface, a display screen, and an input device connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external server through a network connection. When the computer program is executed by the processor, it realizes the functions or steps on the client side of a signal reconstruction method based on a neural network.

[0069] In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the following steps are implemented:

[0070] Obtain an interference signal when the distance of the sample arm mirror from the zero optical path difference is at a depth of z by using a linear wavenumber SD-OCT system;

[0071] Perform DC removal processing on the interference signal, and perform Fourier transform on the interference signal after DC removal processing;

[0072] Extract the interference signal after Fourier transform to obtain a first spatial domain signal;

[0073] Determine a second spatial domain signal based on the first spatial domain signal and a preset amplification factor;

[0074] Perform model training based on the interference signal, the second spatial domain signal, and an initial convolutional neural network to obtain a trained convolutional neural network, where the convolutional neural network is used to reconstruct the interference signal.

[0075] The present invention performs end-to-end learning on the SD-OCT reconstruction task through network training, so as to obtain a convolutional neural network that can achieve high-quality SD-OCT signal reconstruction. After the model is trained, directly input the original OCT spectral interference signal data, and an accurate reconstruction result can be output. This reconstruction method based on the convolutional neural network does not require changing the OCT hardware system, and is of great significance in improving imaging performance and reducing costs.

[0076] In one embodiment, a computer-readable storage medium is provided. The computer-readable storage medium stores a computer program. When the computer program is executed by a processor, the following steps are implemented:

[0077] Obtain an interference signal when the distance of the sample arm mirror from the zero optical path difference is at a depth of z by using a linear wavenumber SD-OCT system;

[0078] Perform DC removal processing on the interference signal, and perform Fourier transform on the interference signal after DC removal processing;

[0079] Extract the interference signal after Fourier transform to obtain a first spatial domain signal;

[0080] Determine a second spatial domain signal based on the first spatial domain signal and a preset amplification factor;

[0081] Perform model training based on the interference signal, the second spatial domain signal, and an initial convolutional neural network to obtain a trained convolutional neural network, where the convolutional neural network is used to reconstruct the interference signal.

[0082] The present invention conducts end-to-end learning on the SD-OCT reconstruction task through network training, thereby obtaining a convolutional neural network that can achieve high-quality SD-OCT signal reconstruction. After the model is trained, by directly inputting the original OCT spectral interference signal data, an accurate reconstruction result can be output. This reconstruction method based on the convolutional neural network does not require any changes to the OCT hardware system and is of great significance in improving imaging performance and reducing costs.

[0083] It should be noted that for the functions or steps that can be achieved by the above-mentioned computer-readable storage medium or computer device, reference can be made to the relevant descriptions on the server side and the client side in the foregoing method embodiments. To avoid repetition, they will not be described in detail here.

[0084] Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it may include the processes of the above method embodiments. Among them, any reference to a memory, storage, database, or other medium used in the various embodiments provided in the present application may include non-volatile and / or volatile memories. Non-volatile memories may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memories may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and Rambus dynamic RAM (RDRAM), etc.

[0085] Those skilled in the art can clearly understand that for the sake of convenience and brevity of description, only the above division of each functional unit and module is used as an example. In actual applications, the above functions can be allocated to different functional units and modules according to needs, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0086] The above-described embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the various embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A signal reconstruction method based on a neural network, characterized in that The neural network-based signal reconstruction method includes: Obtaining an interference signal when the distance of the sample arm mirror from the zero optical path difference is at a depth of z by using a linear wavenumber SD-OCT system; Performing a DC removal process on the interference signal, and performing a Fourier transform on the interference signal after the DC removal process; Extracting the interference signal after the Fourier transform to obtain a first spatial domain signal; Determining a second spatial domain signal based on the first spatial domain signal and a preset amplification factor; Performing model training based on the interference signal, the second spatial domain signal, and an initial convolutional neural network to obtain a trained convolutional neural network, where the convolutional neural network is used to reconstruct the interference signal.

2. The signal reconstruction method based on a neural network according to claim 1, wherein The step of determining the second spatial domain signal based on the first spatial domain signal and a preset amplification factor includes: Multiplying the first spatial domain signal by a corresponding coefficient for amplification to obtain a second spatial domain signal; Determining a target output based on the real part and the imaginary part in the imaginary number of the second spatial domain signal.

3. The signal reconstruction method based on neural network according to claim 2, wherein, The step of performing model training based on the interference signal, the second spatial domain signal, and an initial convolutional neural network to obtain a trained convolutional neural network, where the convolutional neural network is used to reconstruct the interference signal includes: Inputting the interference signal into the initial convolutional neural network to obtain reconstructed data; Calculating a loss function based on the reconstructed data and the target output; Updating the parameters of the initial convolutional neural network based on the loss function to obtain a trained convolutional neural network.

4. The signal reconstruction method based on a neural network according to claim 1, wherein The initial convolutional neural network includes sequentially connected convolutional modules, a flattening layer, and a fully connected layer, and the convolutional module includes sequentially connected a convolutional layer, an activation layer process, and a normalization layer.

5. The signal reconstruction method based on a neural network according to claim 4, wherein The initial convolutional neural network includes sequentially connected a first convolutional layer, a first activation layer process, a first normalization layer, a second convolutional layer, a second activation layer process, a second normalization layer, a third convolutional layer, a third activation layer process, a third normalization layer, a fourth convolutional layer, a fourth activation layer process, a fourth normalization layer, a flattening layer, and a fully connected layer.

6. A signal reconstruction device based on a neural network, characterized in that, The neural network-based signal reconstruction device includes: An acquisition module for acquiring an interference signal when the distance of the sample arm mirror from the zero optical path difference is at a depth of z by using a linear wavenumber SD-OCT system; A processing module for performing a DC removal process on the interference signal and performing a Fourier transform on the interference signal after the DC removal process; An extraction module for extracting the interference signal after the Fourier transform to obtain a first spatial domain signal; A determination module for determining a second spatial domain signal based on the first spatial domain signal and a preset amplification factor; A training module for performing model training based on the interference signal, the second spatial domain signal, and an initial convolutional neural network to obtain a trained convolutional neural network, where the convolutional neural network is used to reconstruct the interference signal.

7. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, the steps of the neural network-based signal reconstruction method according to any one of claims 1 to 5 are implemented.

8. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, the steps of the neural network-based signal reconstruction method according to any one of claims 1 to 5 are implemented.

Images