Signal modulation method, OCT image axial super-resolution reconstruction method and system

Abstract

The present invention provides a signal modulation method, an OCT image axial super-resolution reconstruction method and system, which belong to the general field of image processing technology. The present invention designs a signal modulation strategy that reshapes multiple axial pixel lines of an OCT image into one-dimensional signal features. At the same time, the signal modulation method also has the ability to extract two-dimensional features. The one-dimensional signal obtained after signal modulation is input into a long short-term memory network for feature reconstruction. This method can very accurately enhance the model's ability to capture the front and back associations of pixels in the axial direction, and improve the accuracy of the model's reconstruction of pixel signal differences between consecutive pixels; the present invention proposes an OCT image axial super-resolution reconstruction method based on the pixel timing characteristics of the image, which improves the image objective indicators through three signal modulation processes, improves the image quality degradation problem caused by the weakening of adjacent signal differences, better restores the pixel intensity curve, and improves the image reconstruction effect.

Description

Technical Field

[0001] The present invention relates to the general field of image processing technology, and in particular to a signal modulation method, an OCT image axial super-resolution reconstruction method and a system. Background Art

[0002] The statements in this section merely provide background art related to the present invention and do not necessarily constitute prior art.

[0003] Optical coherence tomography (OCT)-based devices play an important role in various fields, such as transparent and smooth material defect detection, drug quality inspection, and guided laser welding. One of the technical advantages of OCT is the ability to perform high-resolution imaging in the depth direction, so OCT devices often have extremely high requirements for axial resolution. Image axial super-resolution is one of the important ways to upgrade low-axial resolution images to high-axial resolution images.

[0004] Although many studies have found that the reduction in axial resolution of OCT equipment will lead to the weakening of the signal difference between adjacent pixels in the axial direction of the image, the current deep learning model can only learn image structural information and statistical information, and cannot establish information connections between adjacent pixels, and has a weak ability to learn the temporal connections between image pixels. For example, methods based on spatial or frequency domain information can obtain very high objective indicators, but the model cannot learn the temporal information between pixels. The use of statistical information can improve the pixel distribution problem to a certain extent, but it still cannot improve the inefficiency of learning the temporal information before and after pixels; moreover, the existing memory network has not made targeted improvements for OCT images, resulting in a large pixel signal deviation between the axial super-resolution results of OCT images and the real images of the existing methods. Summary of the invention

[0005] In order to address the deficiencies in the prior art, the present invention provides a signal modulation method, an OCT image axial super-resolution reconstruction method and system, which reshape multiple axial pixel lines of the OCT image into one-dimensional signal features, and input the obtained one-dimensional signal into a long short-term memory network for feature reconstruction, which can accurately enhance the model's ability to capture the front and back correlation of pixels in the axial direction, and improve the accuracy of the model's reconstruction of pixel signal differences between consecutive pixels.

[0006] In order to achieve the above object, the present invention adopts the following technical solution:

[0007] In a first aspect, the present invention provides a signal modulation method.

[0008] A signal modulation method, comprising the following process:

[0009] The phase image and amplitude image corresponding to the OCT image are converted into real values, the real values ​​of the phase image and the real values ​​of the amplitude image are reshaped into complex numbers after complex convolution, and then the modulated two-dimensional feature map is obtained through inverse transformation;

[0010] The OCT image is processed by the first reshaping, the first linear mapping and the second linear mapping in sequence. The result after the second linear mapping is copied as the first signal and the second signal respectively. The transpose of the second signal is multiplied by the first signal, and then processed by the second reshaping and the third linear mapping to obtain the third signal. The third signal is residually connected with the result after the first reshaping to obtain the modulated one-dimensional signal feature.

[0011] As a further limitation of the first aspect of the present invention, the OCT image is subjected to fast Fourier transform to obtain the corresponding phase image and amplitude image.

[0012] As a further limitation of the first aspect of the present invention, the real value of the phase image and the real value of the amplitude image are reshaped into complex numbers after being processed by two layers of complex convolution layers, including:

[0013] ;

[0014] in, represents the input image, represents complex convolution, represents the magnitude image, Represented as a phase image, Represented as real part convolution, Represented as imaginary convolution, Expressed as an imaginary unit, it satisfies ;

[0015] After reshaping into complex numbers, inverse Fourier transform is performed to obtain the modulated two-dimensional feature map.

[0016] As a further limitation of the first aspect of the present invention, the size of the OCT image is (B, 1, H, W), and the size of the two-dimensional feature map is (B, 32, H, W), wherein B represents the batch, H represents the image height, and W represents the image width.

[0017] As a further limitation of the first aspect of the present invention, the one-dimensional signal feature includes:

[0018] ;

[0019] Among them, L=LP2(LP1(R1(Input))), Input represents the input OCT image, R1 and R2 represent the results of the first size reshaping processing and the second size reshaping processing respectively, LP1, LP2 and LP3 are the results of the first linear mapping processing, the second linear mapping processing and the third linear mapping processing respectively, and Out represents the output result of the one-dimensional signal feature.

[0020] As a further limitation of the first aspect of the present invention, the size of the input OCT image is (B, 1, H, W), and the size of the one-dimensional signal feature is (B, 1×W, H / 3).

[0021] In a second aspect, the present invention provides a signal modulation system.

[0022] A signal modulation system, comprising:

[0023] The two-dimensional modulation unit is configured to: convert the phase image and the amplitude image corresponding to the OCT image into real values, and reshape the real values ​​of the phase image and the amplitude image into complex numbers after complex convolution processing to obtain a modulated two-dimensional feature map;

[0024] The one-dimensional modulation unit is configured to: subject the OCT image to a first resize reshaping process, a first linear mapping process, and a second linear mapping process in sequence, copy the result after the second linear mapping process as the first signal and the second signal respectively, multiply the transpose of the second signal by the first signal, and then subject the third signal to a second resize reshaping process and a third linear mapping process to obtain a third signal, and perform a residual connection between the third signal and the result after the first resize reshaping process to obtain a modulated one-dimensional signal feature.

[0025] In a third aspect, the present invention provides a method for axial super-resolution reconstruction of OCT images.

[0026] An axial super-resolution reconstruction method for an OCT image includes the following steps:

[0027] Taking the OCT image as input, using the signal modulation method described in the first aspect of the present invention, to obtain a first two-dimensional feature map and a first one-dimensional signal feature;

[0028] Taking the first two-dimensional feature map as input, adopting the signal modulation method described in the first aspect of the present invention to obtain a second two-dimensional feature map and a second one-dimensional signal feature;

[0029] Taking the second two-dimensional feature map as input, adopting the signal modulation method described in the first aspect of the present invention, to obtain a third two-dimensional feature map and a third one-dimensional signal feature;

[0030] The first one-dimensional signal feature, the second one-dimensional signal feature and the third one-dimensional signal feature are all input into the long short-term memory network. The output of the long short-term memory network is resized and processed by the first convolution to obtain the restored two-dimensional feature. The third two-dimensional feature map is processed by the second convolution and residually connected with the restored two-dimensional feature to obtain the reconstruction result.

[0031] As a further limitation of the third aspect of the present invention, the first convolution process is a 3×3 convolution.

[0032] As a further limitation of the third aspect of the present invention, the second convolution processing is: first performing a 1×1 convolution processing, and then performing a 3×3 convolution processing.

[0033] As a further limitation of the third aspect of the present invention, the first one-dimensional signal feature, the second one-dimensional signal feature and the third one-dimensional signal feature are input into the long short-term memory network after an activation function.

[0034] In a fourth aspect, the present invention provides an OCT image axial super-resolution reconstruction system.

[0035] An OCT image axial super-resolution reconstruction system, comprising:

[0036] The first modulation unit is configured to: take the OCT image as input, adopt the signal modulation method according to the first aspect of the present invention, and obtain a first two-dimensional feature map and a first one-dimensional signal feature;

[0037] The second modulation unit is configured to: take the first two-dimensional feature map as input, adopt the signal modulation method according to the first aspect of the present invention, and obtain a second two-dimensional feature map and a second one-dimensional signal feature;

[0038] The third modulation unit is configured to: take the second two-dimensional feature map as input, adopt the signal modulation method according to the first aspect of the present invention, and obtain a third two-dimensional feature map and a third one-dimensional signal feature;

[0039] The reconstruction result generating unit is configured as follows: the first one-dimensional signal feature, the second one-dimensional signal feature and the third one-dimensional signal feature are all input into the long short-term memory network, the output of the long short-term memory network is resized and processed by the first convolution to obtain the restored two-dimensional feature, and the third two-dimensional feature map is processed by the second convolution and then residually connected with the restored two-dimensional feature to obtain the reconstruction result.

[0040] In a fifth aspect, the present invention provides a method for training an axial super-resolution reconstruction model of OCT images.

[0041] A training method for an axial super-resolution reconstruction model of an OCT image includes the following steps:

[0042] The original RAW data collected by the OCT device is obtained, and a high axial resolution image is obtained without any restriction on the original RAW data. A Gaussian window is used to perform spectral cropping on the A line of the original RAW data to obtain a low axial resolution image, and a training set consisting of high axial resolution images and low axial resolution images is constructed;

[0043] The low axial resolution images in the training set are used as the input of the reconstruction model, and the high axial resolution images in the training set are used as the output of the reconstruction model. The reconstruction model is trained, including:

[0044] The low axial resolution image is processed using the signal modulation method described in the first aspect of the present invention to obtain a first two-dimensional feature map and a first one-dimensional signal feature;

[0045] Taking the first two-dimensional feature map as input, adopting the signal modulation method described in the first aspect of the present invention to obtain a second two-dimensional feature map and a second one-dimensional signal feature;

[0046] Taking the second two-dimensional feature map as input, adopting the signal modulation method described in the first aspect of the present invention, to obtain a third two-dimensional feature map and a third one-dimensional signal feature;

[0047] The first one-dimensional signal feature, the second one-dimensional signal feature and the third one-dimensional signal feature are all input into the long short-term memory network. The output of the long short-term memory network is resized and processed by the first convolution to obtain the restored two-dimensional feature. The third two-dimensional feature map is processed by the second convolution and residually connected with the restored two-dimensional feature to obtain the reconstruction result.

[0048] As a further limitation of the fifth aspect of the present invention, the first convolution processing is a 3×3 convolution, and the second convolution processing is: first performing a 1×1 convolution processing, and then performing a 3×3 convolution processing.

[0049] As a further limitation of the fifth aspect of the present invention, the first one-dimensional signal feature, the second one-dimensional signal feature and the third one-dimensional signal feature are input into the long short-term memory network after an activation function.

[0050] In a sixth aspect, the present invention provides a training system for an OCT image axial super-resolution reconstruction model.

[0051] A training system for an axial super-resolution reconstruction model of an OCT image, comprising:

[0052] The training set generation unit is configured to: obtain the original RAW data collected by the OCT device, obtain a high axial resolution image without any restriction on the original RAW data, use a Gaussian window to perform spectral cropping on the A line of the original RAW data to obtain a low axial resolution image, and construct a training set consisting of the high axial resolution image and the low axial resolution image;

[0053] The reconstruction model training unit is configured as follows:

[0054] The low axial resolution images in the training set are used as the input of the reconstruction model, and the high axial resolution images in the training set are used as the output of the reconstruction model. The reconstruction model is trained, including:

[0055] The low axial resolution image is processed using the signal modulation method described in the first aspect of the present invention to obtain a first two-dimensional feature map and a first one-dimensional signal feature;

[0056] Taking the first two-dimensional feature map as input, adopting the signal modulation method described in the first aspect of the present invention to obtain a second two-dimensional feature map and a second one-dimensional signal feature;

[0057] Taking the second two-dimensional feature map as input, adopting the signal modulation method described in the first aspect of the present invention, to obtain a third two-dimensional feature map and a third one-dimensional signal feature;

[0058] The first one-dimensional signal feature, the second one-dimensional signal feature and the third one-dimensional signal feature are all input into the long short-term memory network. The output of the long short-term memory network is resized and processed by the first convolution to obtain the restored two-dimensional feature. The third two-dimensional feature map is processed by the second convolution and residually connected with the restored two-dimensional feature to obtain the reconstruction result.

[0059] In a seventh aspect, the present invention provides a computer device, comprising: a processor and a computer-readable storage medium;

[0060] a processor adapted to execute a computer program;

[0061] A computer-readable storage medium having a computer program stored therein, wherein when the computer program is executed by the processor, the signal modulation method as described in the first aspect of the present invention is implemented; or, the OCT image axial super-resolution reconstruction method as described in the third aspect of the present invention is implemented; or, the OCT image axial super-resolution reconstruction model training method as described in the fifth aspect of the present invention is implemented.

[0062] In an eighth aspect, the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, wherein the computer program is suitable for being loaded by a processor and executing the signal modulation method as described in the first aspect of the present invention; or, the computer program is suitable for being loaded by a processor and executing the OCT image axial super-resolution reconstruction method as described in the third aspect of the present invention; or, the computer program is suitable for being loaded by a processor and executing the OCT image axial super-resolution reconstruction model training method as described in the fifth aspect of the present invention.

[0063] In a ninth aspect, the present invention provides a computer program product, comprising a computer program, which, when executed by a processor, implements the signal modulation method as described in the first aspect of the present invention; or, implements the OCT image axial super-resolution reconstruction method as described in the third aspect of the present invention; or, implements the OCT image axial super-resolution reconstruction model training method as described in the fifth aspect of the present invention.

[0064] Compared with the prior art, the present invention has the following beneficial effects:

[0065] 1. In view of the characteristic of reduced axial resolution of OCT images, the present invention designs a signal modulation strategy to reshape multiple axial pixel lines of OCT images into one-dimensional signal features. At the same time, the signal modulation method also has the ability to extract two-dimensional features. The one-dimensional signal obtained after signal modulation is input into a long short-term memory network for feature reconstruction. This method can very accurately enhance the model's ability to capture the correlation between pixels in the axial direction and improve the accuracy of the model's reconstruction of pixel signal differences between consecutive pixels.

[0066] 2. According to the pixel timing characteristics of the image, the present invention proposes an axial super-resolution reconstruction method for OCT images. Through three signal modulation processes, the objective indicators of the image are improved, the image quality degradation problem caused by the weakening of adjacent signal differences is improved, the pixel intensity curve is better restored, and the image reconstruction effect is improved.

[0067] Advantages of additional aspects of the present invention will be given in part in the following description, and in part will become obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] The accompanying drawings in the specification, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0069] Figure 1 A schematic diagram of the principle of the signal modulation method provided in Embodiment 1 of the present invention;

[0070] Figure 2 A schematic diagram of the principle of a signal modulation system provided in Embodiment 2 of the present invention;

[0071] Figure 3 A schematic diagram of the process of the OCT image axial super-resolution reconstruction method provided in Example 3 of the present invention;

[0072] Figure 4 A schematic diagram of the principle of the method for axial super-resolution reconstruction of OCT images provided in Example 3 of the present invention;

[0073] Figure 5A schematic diagram of the principle of an OCT image axial super-resolution reconstruction system provided in Example 4 of the present invention;

[0074] Figure 6 A schematic flow chart of a method for training an axial super-resolution reconstruction model for OCT images provided in Example 5 of the present invention;

[0075] Figure 7 A schematic diagram of the principle of a training method for an axial super-resolution reconstruction model of OCT images provided in Example 5 of the present invention;

[0076] Figure 8 A schematic diagram of the principle of a training system for an axial super-resolution reconstruction model of OCT images provided in Example 6 of the present invention;

[0077] Fig. 9 A schematic diagram of a computer device provided for Example 7 of the present invention. DETAILED DESCRIPTION

[0078] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0079] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present invention belongs.

[0080] In the absence of conflict, the embodiments of the present invention and the features of the embodiments may be combined with each other.

[0081] Embodiment 1:

[0082] This implementation proposes a signal modulation method, including the following process:

[0083] S1.1: The phase image and amplitude image corresponding to the OCT image are converted into real values. The real values ​​of the phase image and the real values ​​of the amplitude image are reshaped into complex numbers after complex convolution processing to obtain a modulated two-dimensional feature map;

[0084] S1.2: The OCT image is sequentially processed by the first resizing process, the first linear mapping process, and the second linear mapping process, and the result after the second linear mapping process is copied as the first signal (i.e. Figure 1 L1 in the Figure 1 The transpose of the second signal L2 is multiplied by the first signal L1, and then processed by the second size reshaping and the third linear mapping to obtain the third signal L3. The third signal is residually connected with the result L0 after the first size reshaping process to obtain the modulated one-dimensional signal feature.

[0085] More specifically, Figure 1 As shown, the original OCT image (B, 1, H, W) is input into the signal modulation network and then processed in parallel by two branches: resizing and Fourier transform.

[0086] In one of the paths, corresponding to step S1.1, the input image is transformed by fast Fourier transform to obtain a phase image I P and the amplitude image I A And converted into real values, the extracted phase information and amplitude information are enhanced by complex convolution. The two are extracted through two layers of 3×3 complex convolution layers to obtain the output result size of (B, 32, H, W).

[0087] During the complex convolution process, I A and I P Respectively with the convolution kernel K A and K P Multiply and then subtract, and I A and I P Respectively with the convolution kernel K P and K A After multiplication, add them together, and finally reshape the two results into complex numbers as the input of the next signal modulator. The specific operation process is as follows:

[0088] (1);

[0089] Among them, I represents the input image, K represents the complex convolution, and I A It is represented as the amplitude image, Ip is represented as the phase image, Ka is represented as the real part convolution, Kp is represented as the imaginary part convolution, j is represented as the imaginary unit, satisfying ;

[0090] After reshaping into complex numbers, inverse Fourier transform is performed to obtain the modulated two-dimensional feature map.

[0091] In another path, the idea of ​​encoding is used to modulate all A lines into continuous encoding, and the idea of ​​attention is used to extract the key information of continuous signal features, which corresponds to the process of step S1.2. Specifically, the input OCT image size is (B, 1, H, W), where B represents batch, H represents image height, and W represents image width. The input image is processed by resizing 1, and the size of its one-dimensional signal result is (B, 1×W, H). This one-dimensional signal result is obtained after linear mapping layer 1, and its size is (B, 1×W, H / 2). After linear mapping layer 2, it is copied to obtain L1 (i.e., the first signal) and L2 (i.e., the second signal), and their sizes are both (B, 1×W, 1, H / 2). Then the transpose of L2 is multiplied with L1 and reshaped by size 2 and linear mapping layer 3 to obtain the result third signal L3, whose size is (B, 1×W, H). The result is residually connected with L0 to obtain a one-dimensional signal feature output with a size of (B, 1×W, H / 3). The process is described as:

[0092] L=LP2(LP1(R1(Input)))(2);

[0093] (3);

[0094] Where, Input represents the input image, L represents the output result of formula (2), R1 and R2 represent resize 1 (i.e., the first resize processing) and resize 2 (i.e., the second resize processing), respectively, LP1, LP2 and LP3 represent linear mapping layer 1 (i.e., the first resize processing), linear mapping layer 2 (i.e., the second resize processing) and linear mapping layer 3 (i.e., the third resize processing), respectively, and Out represents the output result.

[0095] Embodiment 2:

[0096] like Figure 2 As shown, this implementation provides a signal modulation system, including:

[0097] The two-dimensional modulation unit is configured to: convert the phase image and the amplitude image corresponding to the OCT image into real values, and reshape the real values ​​of the phase image and the amplitude image into complex numbers after complex convolution processing to obtain a modulated two-dimensional feature map; the specific working process of this unit is the same as the process of step S1.1 in Example 1, which will not be repeated here;

[0098] The one-dimensional modulation unit is configured to: subject the OCT image to a first reshaping process, a first linear mapping process, and a second linear mapping process in sequence, copy the result after the second linear mapping process as the first signal and the second signal respectively, multiply the transpose of the second signal by the first signal, and then subject the OCT image to a second reshaping process and a third linear mapping process to obtain a third signal, and perform a residual connection between the third signal and the result after the first reshaping process to obtain a modulated one-dimensional signal feature; the specific working process of this unit is the same as the process of step S1.2 in Example 1, and will not be repeated here.

[0099] It is understandable that the above-mentioned units can be separately or completely combined into one or several other units to constitute, or one (some) of the units can be further divided into multiple functionally smaller units to constitute, which can achieve the same operation without affecting the realization of the technical effects of the embodiments of the present application. The above-mentioned units are divided based on logical functions. In practical applications, the function of one unit can also be implemented by multiple units, or the functions of multiple units can be implemented by one unit. In other embodiments of the present application, the signal modulation system may also include other units. In practical applications, these functions can also be implemented with the assistance of other units, and can be implemented by the collaboration of multiple units.

[0100] According to another embodiment of the present application, the system described in this embodiment can be constructed and the signal modulation method of the embodiment of the present application can be implemented by running a computer program (including program code) capable of executing the steps involved in the corresponding method described in Example 1 on a general computing device such as a computer including processing elements and storage elements such as a central processing unit (CPU), random access memory (RAM), and read-only memory (ROM). The computer program can be recorded on, for example, a computer-readable recording medium, and loaded into the above-mentioned computing device through the computer-readable recording medium and run therein.

[0101] Embodiment 3:

[0102] This implementation proposes an axial super-resolution reconstruction method for OCT images, such as Figure 3 and Figure 4 As shown, the following process is included:

[0103] S3.1: Using the OCT image as input, the signal modulation method described in Embodiment 1 of the present invention is used to obtain a first two-dimensional feature map S1 and a first one-dimensional signal feature F1;

[0104] S3.2: Taking the first two-dimensional feature map as input, adopting the signal modulation method described in Embodiment 1 of the present invention to obtain a second two-dimensional feature map S2 and a second one-dimensional signal feature F2;

[0105] S3.3: using the second two-dimensional feature map as input, and adopting the signal modulation method described in Embodiment 1 of the present invention, obtaining a third two-dimensional feature map S3 and a third one-dimensional signal feature F3;

[0106] S3.4: The first one-dimensional signal feature, the second one-dimensional signal feature and the third one-dimensional signal feature are all input into the long short-term memory network. The output S4 of the long short-term memory network is resized to obtain feature F5. Feature F5 is processed by the first convolution (i.e., 3×3 convolution) to obtain the restored two-dimensional feature. The third two-dimensional feature map is processed by the second convolution (i.e., 1×1 convolution and 3×3 convolution) to obtain feature F4, which is residually connected with the restored two-dimensional feature to obtain the reconstruction result.

[0107] More specifically, the input image first undergoes the first signal modulation process of step 3.1 to obtain a first one-dimensional signal feature and a first two-dimensional feature map, wherein the first one-dimensional signal feature will be input into the long short-term memory network downstream of the model, and the first two-dimensional feature map will be input into the second signal modulation process of step 3.2;

[0108] The result of the second signal modulation process is to generate a second one-dimensional signal feature and input it into the long short-term memory network, and the generated second two-dimensional feature map undergoes the third signal modulation process in step 3.3, and the result is to generate a third one-dimensional signal feature and input it into the long short-term memory network, and the third two-dimensional feature map undergoes 1×1 convolution and 3×3 convolution to obtain a deep image space feature map;

[0109] In step S3.4 of this implementation, the first one-dimensional signal feature, the second one-dimensional signal feature and the third one-dimensional signal feature are all activated by the Sigmoid activation function (i.e. Figure 5 The S in the circle in the figure is activated and input into the long short-term memory network (an additional channel merger is added to the long short-term memory network to integrate the one-dimensional signal outputs from different depths, and the final result is S4). The long short-term memory network extracts the feature information between the pixels before and after the one-dimensional signal, and establishes the feature structure of the associated pixels. The result is restored to a two-dimensional feature through size reshaping and 3×3 convolution, and then residually connected with the deep image space feature map to finally obtain the reconstruction result.

[0110] Embodiment 4:

[0111] like Figure 5 As shown, this implementation provides an OCT image axial super-resolution reconstruction system, including:

[0112] The first modulation unit is configured to: take the OCT image as input, adopt the signal modulation method described in Embodiment 1 of the present invention, and obtain a first two-dimensional feature map and a first one-dimensional signal feature; the specific working method of this unit is the same as the process of step S3.1 in Embodiment 1, and will not be repeated here;

[0113] The second modulation unit is configured to: take the first two-dimensional feature map as input, adopt the signal modulation method described in Embodiment 1 of the present invention, obtain a second two-dimensional feature map and a second one-dimensional signal feature; the specific working method of this unit is the same as the process of step S3.2 in Embodiment 1, and will not be repeated here;

[0114] The third modulation unit is configured to: take the second two-dimensional feature map as input, adopt the signal modulation method described in Embodiment 1 of the present invention, obtain a third two-dimensional feature map and a third one-dimensional signal feature; the specific working method of this unit is the same as the process of step S3.3 in Embodiment 1, and will not be repeated here;

[0115] The reconstruction result generating unit is configured as follows: the first one-dimensional signal feature, the second one-dimensional signal feature and the third one-dimensional signal feature are all input into the long short-term memory network; the output of the long short-term memory network is resized and subjected to the first convolution process to obtain the restored two-dimensional feature; the third two-dimensional feature map is subjected to the second convolution process and then residually connected with the restored two-dimensional feature to obtain the reconstruction result; the specific working method of this unit is the same as the process of step S3.4 in Example 1, and will not be repeated here.

[0116] Embodiment 5:

[0117] This implementation provides a training method for an axial super-resolution reconstruction model of an OCT image. Figure 6 As shown, the following process is included:

[0118] S5.1: Obtain the original RAW data collected by the OCT device, obtain a high axial resolution image without any restriction on the original RAW data, use a Gaussian window to perform spectral cropping on the A line of the original RAW data to obtain a low axial resolution image, and construct a training set consisting of high axial resolution images and low axial resolution images;

[0119] S5.2: Using the low axial resolution images in the training set as the input of the reconstruction model and the high axial resolution images in the training set as the output of the reconstruction model, the reconstruction model is trained, including:

[0120] The low axial resolution image is processed using the signal modulation method described in Embodiment 1 of the present invention to obtain a first two-dimensional feature map and a first one-dimensional signal feature;

[0121] Taking the first two-dimensional feature map as input, adopting the signal modulation method described in Embodiment 1 of the present invention to obtain a second two-dimensional feature map and a second one-dimensional signal feature;

[0122] Taking the second two-dimensional feature map as input, adopting the signal modulation method described in Embodiment 1 of the present invention, to obtain a third two-dimensional feature map and a third one-dimensional signal feature;

[0123] The first one-dimensional signal feature, the second one-dimensional signal feature and the third one-dimensional signal feature are all input into the long short-term memory network. The output of the long short-term memory network is resized and processed by the first convolution to obtain the restored two-dimensional feature. The third two-dimensional feature map is processed by the second convolution and residually connected with the restored two-dimensional feature to obtain the reconstruction result.

[0124] In step S5.1 of the present implementation method, specifically, the OCT device images the object to obtain raw RAW data. When the raw RAW data is subjected to an image reconstruction algorithm, two types of processing are performed to produce image data pairs of high and low axial resolution images. A high axial resolution image is obtained without any restriction on the raw RAW data, and a Gaussian window is used to perform spectral cropping on the A line of the raw RAW data to obtain a low axial resolution image to form a training set.

[0125] More specifically, Figure 7 As shown, the OCT system collects raw image data, reconstructs the raw data without any restrictions to obtain a high axial resolution B-scan image, uses a Gaussian window to truncate the spectrum when reconstructing the image from the raw data, and obtains a low axial resolution B-scan image, and inputs all B-scan images into a temporal pixel signal restoration network to achieve axial super-resolution of the OCT image, sets hyperparameters and trains the neural network model, and the OCT device loads the trained model, collects raw data and reconstructs the image, then starts the algorithm to perform axial super-resolution reconstruction of the image.

[0126] Embodiment 6:

[0127] This implementation provides a training system for an axial super-resolution reconstruction model of OCT images. Figure 8 As shown, including:

[0128] The training set generation unit is configured to: obtain the original RAW data collected by the OCT device, obtain a high axial resolution image without any restriction on the original RAW data, use a Gaussian window to perform spectral cropping on the A line of the original RAW data to obtain a low axial resolution image, and construct a training set consisting of the high axial resolution image and the low axial resolution image;

[0129] The reconstruction model training unit is configured as follows:

[0130] The low axial resolution images in the training set are used as the input of the reconstruction model, and the high axial resolution images in the training set are used as the output of the reconstruction model. The reconstruction model is trained, including:

[0131] The low axial resolution image is processed using the signal modulation method described in Embodiment 1 of the present invention to obtain a first two-dimensional feature map and a first one-dimensional signal feature;

[0132] Taking the first two-dimensional feature map as input, adopting the signal modulation method described in Embodiment 1 of the present invention to obtain a second two-dimensional feature map and a second one-dimensional signal feature;

[0133] Taking the second two-dimensional feature map as input, adopting the signal modulation method described in Embodiment 1 of the present invention, to obtain a third two-dimensional feature map and a third one-dimensional signal feature;

[0134] The first one-dimensional signal feature, the second one-dimensional signal feature and the third one-dimensional signal feature are all input into the long short-term memory network. The output of the long short-term memory network is resized and processed by the first convolution to obtain the restored two-dimensional feature. The third two-dimensional feature map is processed by the second convolution and residually connected with the restored two-dimensional feature to obtain the reconstruction result.

[0135] Embodiment 7:

[0136] like Fig. 9 As shown, this implementation provides an electronic device, which includes a processor 1001, a communication interface 1002, and a computer-readable storage medium 1003. The processor 1001, the communication interface 1002, and the computer-readable storage medium 1003 may be connected via a bus or other means.

[0137] Among them, the communication interface 1002 is used to receive and send data, the computer-readable storage medium 1003 can be stored in the memory of the electronic device, the computer-readable storage medium 1003 is used to store a computer program, the computer program includes program instructions, and the processor 1001 is used to execute the program instructions stored in the computer-readable storage medium 1003.

[0138] The processor 1001 (or CPU (Central Processing Unit)) is the computing core and control core of the electronic device, which is suitable for implementing one or more instructions, and specifically suitable for loading and executing one or more instructions to implement corresponding method processes or corresponding functions.

[0139] The processor 1001 is configured to execute the following process:

[0140] The phase image and the amplitude image corresponding to the OCT image are converted into real values, and the real values ​​of the phase image and the amplitude image are reshaped into complex numbers after complex convolution processing to obtain a modulated two-dimensional feature map; the specific working process is the same as the process of step S1.1 in Example 1, which will not be repeated here;

[0141] The OCT image is sequentially subjected to a first resize reshaping process, a first linear mapping process, and a second linear mapping process, and the result after the second linear mapping process is copied as the first signal and the second signal respectively, the transpose of the second signal is multiplied by the first signal, and then subjected to a second resize reshaping process and a third linear mapping process to obtain a third signal, and the third signal is residually connected with the result after the first resize reshaping process to obtain a modulated one-dimensional signal feature; the specific working process is the same as the process of step S1.2 in Example 1, and will not be repeated here;

[0142] Alternatively, implement the following process:

[0143] Taking the OCT image as input, the signal modulation method described in Example 1 of the present invention is adopted to obtain a first two-dimensional feature map and a first one-dimensional signal feature; the specific working method is the same as the process of step S3.1 in Example 1, which will not be repeated here;

[0144] Taking the first two-dimensional feature map as input, the signal modulation method described in Example 1 of the present invention is adopted to obtain a second two-dimensional feature map and a second one-dimensional signal feature; the specific working method is the same as the process of step S3.2 in Example 1, which will not be repeated here;

[0145] Taking the second two-dimensional feature map as input, the signal modulation method described in Example 1 of the present invention is adopted to obtain a third two-dimensional feature map and a third one-dimensional signal feature; the specific working method is the same as the process of step S3.3 in Example 1, which will not be repeated here;

[0146] The first one-dimensional signal feature, the second one-dimensional signal feature and the third one-dimensional signal feature are all input into the long short-term memory network. The output of the long short-term memory network is resized and processed by the first convolution to obtain the restored two-dimensional feature. The third two-dimensional feature map is processed by the second convolution and then residually connected with the restored two-dimensional feature to obtain a reconstruction result. The specific working method is the same as the process of step S3.4 in Example 1, which will not be repeated here.

[0147] Alternatively, implement the following process:

[0148] The original RAW data collected by the OCT device is obtained, and a high axial resolution image is obtained without any restriction on the original RAW data. A Gaussian window is used to perform spectral cropping on the A line of the original RAW data to obtain a low axial resolution image, and a training set consisting of high axial resolution images and low axial resolution images is constructed;

[0149] Conduct training of the reconstruction model, including:

[0150] The low axial resolution images in the training set are used as the input of the reconstruction model, and the high axial resolution images in the training set are used as the output of the reconstruction model. The reconstruction model is trained, including:

[0151] The low axial resolution image is processed using the signal modulation method described in Embodiment 1 of the present invention to obtain a first two-dimensional feature map and a first one-dimensional signal feature;

[0152] Taking the first two-dimensional feature map as input, adopting the signal modulation method described in Embodiment 1 of the present invention to obtain a second two-dimensional feature map and a second one-dimensional signal feature;

[0153] Taking the second two-dimensional feature map as input, adopting the signal modulation method described in Embodiment 1 of the present invention, to obtain a third two-dimensional feature map and a third one-dimensional signal feature;

[0154] The first one-dimensional signal feature, the second one-dimensional signal feature and the third one-dimensional signal feature are all input into the long short-term memory network. The output of the long short-term memory network is resized and processed by the first convolution to obtain the restored two-dimensional feature. The third two-dimensional feature map is processed by the second convolution and residually connected with the restored two-dimensional feature to obtain the reconstruction result.

[0155] Embodiment 8:

[0156] This implementation provides a computer-readable storage medium (Memory), which is a memory device in an electronic device for storing programs and data. It is understandable that the computer-readable storage medium here can include both built-in storage media in the electronic device and, of course, extended storage media supported by the electronic device. The computer-readable storage medium provides a storage space that stores the processing system of the electronic device.

[0157] In addition, the storage space also stores one or more instructions suitable for being loaded and executed by the processor, and these instructions may be one or more computer programs (including program codes). It should be noted that the computer-readable storage medium here may be a high-speed RAM memory, or a non-volatile memory (non-volatile memory), such as at least one disk storage; optionally, it may also be at least one computer-readable storage medium located away from the aforementioned processor.

[0158] In one embodiment, the computer-readable storage medium stores one or more instructions; the processor loads and executes the one or more instructions stored in the computer-readable storage medium to implement the following process:

[0159] The phase image and the amplitude image corresponding to the OCT image are converted into real values, and the real values ​​of the phase image and the amplitude image are reshaped into complex numbers after complex convolution processing to obtain a modulated two-dimensional feature map; the specific working process is the same as the process of step S1.1 in Example 1, which will not be repeated here;

[0160] The OCT image is sequentially subjected to a first resize reshaping process, a first linear mapping process, and a second linear mapping process, and the result after the second linear mapping process is copied as the first signal and the second signal respectively, the transpose of the second signal is multiplied by the first signal, and then subjected to a second resize reshaping process and a third linear mapping process to obtain a third signal, and the third signal is residually connected with the result after the first resize reshaping process to obtain a modulated one-dimensional signal feature; the specific working process is the same as the process of step S1.2 in Example 1, and will not be repeated here;

[0161] Alternatively, implement the following process:

[0162] Taking the OCT image as input, the signal modulation method described in Example 1 of the present invention is adopted to obtain a first two-dimensional feature map and a first one-dimensional signal feature; the specific working method is the same as the process of step S3.1 in Example 1, which will not be repeated here;

[0163] Taking the first two-dimensional feature map as input, the signal modulation method described in Example 1 of the present invention is adopted to obtain a second two-dimensional feature map and a second one-dimensional signal feature; the specific working method is the same as the process of step S3.2 in Example 1, which will not be repeated here;

[0164] Taking the second two-dimensional feature map as input, the signal modulation method described in Example 1 of the present invention is adopted to obtain a third two-dimensional feature map and a third one-dimensional signal feature; the specific working method is the same as the process of step S3.3 in Example 1, which will not be repeated here;

[0165] The first one-dimensional signal feature, the second one-dimensional signal feature and the third one-dimensional signal feature are all input into the long short-term memory network. The output of the long short-term memory network is resized and processed by the first convolution to obtain the restored two-dimensional feature. The third two-dimensional feature map is processed by the second convolution and then residually connected with the restored two-dimensional feature to obtain a reconstruction result. The specific working method is the same as the process of step S3.4 in Example 1, which will not be repeated here.

[0166] Alternatively, implement the following process:

[0167] The original RAW data collected by the OCT device is obtained, and a high axial resolution image is obtained without any restriction on the original RAW data. A Gaussian window is used to perform spectral cropping on the A line of the original RAW data to obtain a low axial resolution image, and a training set consisting of high axial resolution images and low axial resolution images is constructed;

[0168] Conduct training of the reconstruction model, including:

[0169] The low axial resolution images in the training set are used as the input of the reconstruction model, and the high axial resolution images in the training set are used as the output of the reconstruction model. The reconstruction model is trained, including:

[0170] The low axial resolution image is processed using the signal modulation method described in Embodiment 1 of the present invention to obtain a first two-dimensional feature map and a first one-dimensional signal feature;

[0171] Taking the first two-dimensional feature map as input, adopting the signal modulation method described in Embodiment 1 of the present invention to obtain a second two-dimensional feature map and a second one-dimensional signal feature;

[0172] Taking the second two-dimensional feature map as input, adopting the signal modulation method described in Embodiment 1 of the present invention, to obtain a third two-dimensional feature map and a third one-dimensional signal feature;

[0173] The first one-dimensional signal feature, the second one-dimensional signal feature and the third one-dimensional signal feature are all input into the long short-term memory network. The output of the long short-term memory network is resized and processed by the first convolution to obtain the restored two-dimensional feature. The third two-dimensional feature map is processed by the second convolution and residually connected with the restored two-dimensional feature to obtain the reconstruction result.

[0174] Embodiment 9:

[0175] The present implementation provides a computer program product or a computer program, which includes computer instructions stored in a computer-readable storage medium. A processor of an electronic device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the electronic device performs the following process:

[0176] The phase image and the amplitude image corresponding to the OCT image are converted into real values, and the real values ​​of the phase image and the amplitude image are reshaped into complex numbers after complex convolution processing to obtain a modulated two-dimensional feature map; the specific working process is the same as the process of step S1.1 in Example 1, which will not be repeated here;

[0177] The OCT image is sequentially subjected to a first resize reshaping process, a first linear mapping process, and a second linear mapping process, and the result after the second linear mapping process is copied as the first signal and the second signal respectively, the transpose of the second signal is multiplied by the first signal, and then subjected to a second resize reshaping process and a third linear mapping process to obtain a third signal, and the third signal is residually connected with the result after the first resize reshaping process to obtain a modulated one-dimensional signal feature; the specific working process is the same as the process of step S1.2 in Example 1, and will not be repeated here;

[0178] Alternatively, implement the following process:

[0179] Taking the OCT image as input, the signal modulation method described in Example 1 of the present invention is adopted to obtain a first two-dimensional feature map and a first one-dimensional signal feature; the specific working method is the same as the process of step S3.1 in Example 1, which will not be repeated here;

[0180] Taking the first two-dimensional feature map as input, the signal modulation method described in Example 1 of the present invention is adopted to obtain a second two-dimensional feature map and a second one-dimensional signal feature; the specific working method is the same as the process of step S3.2 in Example 1, which will not be repeated here;

[0181] Taking the second two-dimensional feature map as input, the signal modulation method described in Example 1 of the present invention is adopted to obtain a third two-dimensional feature map and a third one-dimensional signal feature; the specific working method is the same as the process of step S3.3 in Example 1, which will not be repeated here;

[0182] The first one-dimensional signal feature, the second one-dimensional signal feature and the third one-dimensional signal feature are all input into the long short-term memory network. The output of the long short-term memory network is resized and processed by the first convolution to obtain the restored two-dimensional feature. The third two-dimensional feature map is processed by the second convolution and then residually connected with the restored two-dimensional feature to obtain a reconstruction result. The specific working method is the same as the process of step S3.4 in Example 1, which will not be repeated here.

[0183] Alternatively, implement the following process:

[0184] The original RAW data collected by the OCT device is obtained, and a high axial resolution image is obtained without any restriction on the original RAW data. A Gaussian window is used to perform spectral cropping on the A line of the original RAW data to obtain a low axial resolution image, and a training set consisting of high axial resolution images and low axial resolution images is constructed;

[0185] Conduct training of the reconstruction model, including:

[0186] The low axial resolution images in the training set are used as the input of the reconstruction model, and the high axial resolution images in the training set are used as the output of the reconstruction model. The reconstruction model is trained, including:

[0187] The low axial resolution image is processed using the signal modulation method described in Embodiment 1 of the present invention to obtain a first two-dimensional feature map and a first one-dimensional signal feature;

[0188] Taking the first two-dimensional feature map as input, adopting the signal modulation method described in Embodiment 1 of the present invention to obtain a second two-dimensional feature map and a second one-dimensional signal feature;

[0189] Taking the second two-dimensional feature map as input, adopting the signal modulation method described in Embodiment 1 of the present invention, to obtain a third two-dimensional feature map and a third one-dimensional signal feature;

[0190] The first one-dimensional signal feature, the second one-dimensional signal feature and the third one-dimensional signal feature are all input into the long short-term memory network. The output of the long short-term memory network is resized and processed by the first convolution to obtain the restored two-dimensional feature. The third two-dimensional feature map is processed by the second convolution and residually connected with the restored two-dimensional feature to obtain the reconstruction result.

[0191] A person skilled in the art can appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed in this application can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. A person skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0192] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware or any combination thereof. When implemented using software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on the computer, the process or function according to the embodiment of the present application is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium or transmitted through a computer-readable storage medium. The computer instructions can be transmitted from a website site, a computer, a server or a data center to another website site, a computer, a server or a data center by wired (e.g., coaxial cable, optical fiber, digital line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium that a computer can access or a data processing device such as a server, a data center, etc. that includes one or more available media integrated. The available medium can be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium (e.g., a solid-state drive (SSD)), etc.

[0193] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A signal modulation method, characterized in that: The process includes: The phase image and amplitude image corresponding to the OCT image are converted into real values, and the real values ​​of the phase image and the amplitude image are reshaped into complex numbers after complex convolution processing to obtain a modulated two-dimensional feature map; The OCT image is processed by the first reshaping, the first linear mapping and the second linear mapping in sequence. The result after the second linear mapping is copied as the first signal and the second signal respectively. The transpose of the second signal is multiplied by the first signal, and then processed by the second reshaping and the third linear mapping to obtain the third signal. The third signal is residually connected with the result after the first reshaping to obtain the modulated one-dimensional signal feature.

2. The signal modulation method according to claim 1, characterized in that: The OCT image is transformed by fast Fourier transform to obtain the corresponding phase image and amplitude image.

3. The signal modulation method according to claim 1 or 2, characterized in that: The real-valued phase image and the real-valued magnitude image are reshaped into complex numbers after being processed by two layers of complex convolution layers, including: ； in, represents the input image, represents complex convolution, represents the magnitude image, Represented as a phase image, Represented as real part convolution, Represented as imaginary convolution, Expressed as an imaginary unit, it satisfies ; After reshaping into complex numbers, inverse Fourier transform is performed to obtain the modulated two-dimensional feature map.

4. The signal modulation method according to claim 1 or 2, characterized in that: The size of the OCT image is (B, 1, H, W), and the size of the two-dimensional feature map is (B, 32, H, W), where B represents the batch, H represents the image height, and W represents the image width.

5. The signal modulation method according to claim 1 or 2, characterized in that: The one-dimensional signal features include: ； Among them, L=LP2(LP1(R1 (Input))), Input represents the input OCT image, R1 and R2 represent the results of the first size reshaping processing and the second size reshaping processing respectively, LP1, LP2 and LP3 are the results of the first linear mapping processing, the second linear mapping processing and the third linear mapping processing respectively, and Out represents the output result of the one-dimensional signal feature.

6. The signal modulation method according to claim 5, characterized in that: The size of the input OCT image is (B, 1, H, W), and the size of the one-dimensional signal feature is (B, 1×W, H / 3); wherein B represents batch, H represents image height, and W represents image width.

7. A signal modulation system, characterized in that: include: The two-dimensional modulation unit is configured to: convert the phase image and the amplitude image corresponding to the OCT image into real values, and reshape the real values ​​of the phase image and the amplitude image into complex numbers after complex convolution processing to obtain a modulated two-dimensional feature map; The one-dimensional modulation unit is configured to: subject the OCT image to a first resize reshaping process, a first linear mapping process, and a second linear mapping process in sequence, copy the result after the second linear mapping process as the first signal and the second signal respectively, multiply the transpose of the second signal by the first signal, and then subject the third signal to a second resize reshaping process and a third linear mapping process to obtain a third signal, and perform a residual connection between the third signal and the result after the first resize reshaping process to obtain a modulated one-dimensional signal feature.

8. A method for axial super-resolution reconstruction of OCT images, characterized in that: The process includes: Taking the OCT image as input, adopting the signal modulation method according to any one of claims 1 to 6 to obtain a first two-dimensional feature map and a first one-dimensional signal feature; Taking the first two-dimensional feature map as input, adopting the signal modulation method according to any one of claims 1 to 6 to obtain a second two-dimensional feature map and a second one-dimensional signal feature; Taking the second two-dimensional feature map as input, adopting the signal modulation method according to any one of claims 1 to 6 to obtain a third two-dimensional feature map and a third one-dimensional signal feature; The first one-dimensional signal feature, the second one-dimensional signal feature and the third one-dimensional signal feature are all input into the long short-term memory network. The output of the long short-term memory network is resized and processed by the first convolution to obtain the restored two-dimensional feature. The third two-dimensional feature map is processed by the second convolution and residually connected with the restored two-dimensional feature to obtain the reconstruction result.

9. The method for axial super-resolution reconstruction of OCT images according to claim 8, characterized in that: The first convolution process is a 3×3 convolution.

10. The method for axial super-resolution reconstruction of OCT images according to claim 8, characterized in that: The second convolution process is: first perform a 1×1 convolution process, and then perform a 3×3 convolution process.

11. The method for axial super-resolution reconstruction of OCT images according to claim 8, characterized in that: The first one-dimensional signal feature, the second one-dimensional signal feature and the third one-dimensional signal feature are input into the long short-term memory network after the activation function.

12. An OCT image axial super-resolution reconstruction system, characterized in that: include: The first modulation unit is configured to: take the OCT image as input, adopt the signal modulation method according to any one of claims 1 to 6, and obtain a first two-dimensional feature map and a first one-dimensional signal feature; The second modulation unit is configured to: take the first two-dimensional feature map as input, adopt the signal modulation method according to any one of claims 1 to 6, to obtain a second two-dimensional feature map and a second one-dimensional signal feature; The third modulation unit is configured to: take the second two-dimensional feature map as input, adopt the signal modulation method according to any one of claims 1 to 6, to obtain a third two-dimensional feature map and a third one-dimensional signal feature; The reconstruction result generating unit is configured as follows: the first one-dimensional signal feature, the second one-dimensional signal feature and the third one-dimensional signal feature are all input into the long short-term memory network, the output of the long short-term memory network is resized and processed by the first convolution to obtain the restored two-dimensional feature, and the third two-dimensional feature map is processed by the second convolution and then residually connected with the restored two-dimensional feature to obtain the reconstruction result.

13. A training method for an axial super-resolution reconstruction model of an OCT image, characterized in that: The process includes: The original RAW data collected by the OCT device is obtained, and a high axial resolution image is obtained without any restriction on the original RAW data. A Gaussian window is used to perform spectral cropping on the A line of the original RAW data to obtain a low axial resolution image, and a training set consisting of high axial resolution images and low axial resolution images is constructed; The low axial resolution images in the training set are used as the input of the reconstruction model, and the high axial resolution images in the training set are used as the output of the reconstruction model. The reconstruction model is trained, including: Processing the low axial resolution image using the signal modulation method according to any one of claims 1 to 6 to obtain a first two-dimensional feature map and a first one-dimensional signal feature; Taking the first two-dimensional feature map as input, adopting the signal modulation method according to any one of claims 1 to 6 to obtain a second two-dimensional feature map and a second one-dimensional signal feature; Taking the second two-dimensional feature map as input, adopting the signal modulation method according to any one of claims 1 to 6 to obtain a third two-dimensional feature map and a third one-dimensional signal feature; The first one-dimensional signal feature, the second one-dimensional signal feature and the third one-dimensional signal feature are all input into the long short-term memory network. The output of the long short-term memory network is resized and processed by the first convolution to obtain the restored two-dimensional feature. The third two-dimensional feature map is processed by the second convolution and residually connected with the restored two-dimensional feature to obtain the reconstruction result.

14. The method for training an axial super-resolution reconstruction model of an OCT image according to claim 13, characterized in that: The first convolution process is a 3×3 convolution, and the second convolution process is: first perform a 1×1 convolution process, and then perform a 3×3 convolution process.

15. The method for training an axial super-resolution reconstruction model of an OCT image according to claim 13, characterized in that: The first one-dimensional signal feature, the second one-dimensional signal feature and the third one-dimensional signal feature are input into the long short-term memory network after the activation function.

16. A training system for an axial super-resolution reconstruction model of OCT images, characterized in that: include: The training set generation unit is configured to: obtain the original RAW data collected by the OCT device, obtain a high axial resolution image without any restriction on the original RAW data, use a Gaussian window to perform spectral cropping on the A line of the original RAW data to obtain a low axial resolution image, and construct a training set consisting of the high axial resolution image and the low axial resolution image; The reconstruction model training unit is configured as follows: The low axial resolution images in the training set are used as the input of the reconstruction model, and the high axial resolution images in the training set are used as the output of the reconstruction model. The reconstruction model is trained, including: Processing the low axial resolution image using the signal modulation method according to any one of claims 1 to 6 to obtain a first two-dimensional feature map and a first one-dimensional signal feature; Taking the first two-dimensional feature map as input, adopting the signal modulation method according to any one of claims 1 to 6 to obtain a second two-dimensional feature map and a second one-dimensional signal feature; Taking the second two-dimensional feature map as input, adopting the signal modulation method according to any one of claims 1 to 6 to obtain a third two-dimensional feature map and a third one-dimensional signal feature; The first one-dimensional signal feature, the second one-dimensional signal feature and the third one-dimensional signal feature are all input into the long short-term memory network. The output of the long short-term memory network is resized and processed by the first convolution to obtain the restored two-dimensional feature. The third two-dimensional feature map is processed by the second convolution and residually connected with the restored two-dimensional feature to obtain the reconstruction result.

17. A computer device, characterized in that: include: a processor and a computer readable storage medium; a processor adapted to execute a computer program; A computer-readable storage medium having a computer program stored therein, wherein when the computer program is executed by the processor, the signal modulation method as described in any one of claims 1 to 6 is implemented; or, the OCT image axial super-resolution reconstruction method as described in any one of claims 8 to 11 is implemented; or, the OCT image axial super-resolution reconstruction model training method as described in any one of claims 13 to 15 is implemented.

18. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and the computer program is suitable for being loaded by a processor and executing the signal modulation method as described in any one of claims 1 to 6; or, the computer program is suitable for being loaded by a processor and executing the OCT image axial super-resolution reconstruction method as described in any one of claims 8 to 11; or, the computer program is suitable for being loaded by a processor and executing the OCT image axial super-resolution reconstruction model training method as described in any one of claims 13 to 15.

19. A computer program product, characterized in that The computer program product includes a computer program, which, when executed by a processor, implements the signal modulation method as described in any one of claims 1 to 6; or, implements the OCT image axial super-resolution reconstruction method as described in any one of claims 8 to 11; or, implements the OCT image axial super-resolution reconstruction model training method as described in any one of claims 13 to 15.

Images