An unsupervised corneal contour extraction method, system and device

Abstract

The application discloses a kind of unsupervised corneal profile extraction method, system and equipment, it is related to image processing technical field, method includes: each sample corneal video in each frame sample corneal image is carried out SLIC superpixel segmentation, to obtain multiple pixel groups and corresponding grouping label;If sample corneal image is the first frame in the first sample corneal video, sample corneal image and multiple pixel groups corresponding grouping label are input into initial corneal profile extraction model and are trained;If sample corneal image is not the first frame in the first sample corneal video, sample corneal image and multiple pixel groups corresponding grouping label are input into corneal profile extraction model trained by last frame sample corneal image and are trained;When the loss function value corresponding to corneal profile extraction model meets preset condition, mark as training completion, and output corresponding corneal profile pixel point.The application improves corneal profile extraction precision.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of image processing, in particular to an unsupervised corneal contour extraction method, system and device. BACKGROUND

[0002] The biomechanical properties of the cornea have been proven to play an important role in maintaining its structure, the development of ectasia, the screening and measurement of glaucoma, the evaluation of refractive surgery, etc. The corneal biomechanical visualizer is a device currently used for in vivo corneal biomechanical evaluation. The stress deformation video of the cornea captured by the device is the most direct reflection of the biomechanical properties of the cornea. Precise extraction of the corneal contour from these dynamic videos is a prerequisite for accurate calculation and evaluation of corneal biomechanical characteristics, and is also an important basis for further establishing a standardized corneal biomechanical model.

[0003] Existing corneal contour extraction is mostly based on traditional edge detection methods. Magdalena et al. proposed a method for detecting the outer edge of the cornea based on corvisST series images, with a minimum average error of 0.16%. Wang Yi et al. proposed an automatic extraction algorithm for the contour of a swept-frequency optical coherence tomography corneal image. The algorithm divides the image into high and low signal-to-noise ratio regions for processing, and uses the peak point positioning method combined with the actual contour information of the cornea as a weighting factor to achieve corneal contour positioning, obtaining a corneal thickness that is closer to the actual thickness, with an average extraction accuracy improved by 4.9%. Ji et al. proposed a method based on edge detection and 5th order polynomial approximation. Kopr-owski et al. proposed a method for detecting the outer contour of the cornea based on Otsu segmentation and Canny operator, which achieved good results with a correct rate of 90%. However, the above methods have the following defects: 1) a large amount of labeled data is required, and the labeling process is lengthy and prone to human error. 2) When the image quality is not high or mixed with serious noise, the extraction accuracy of the corneal contour cannot meet the clinical requirements. SUMMARY

[0004] The purpose of the present application is to provide an unsupervised corneal contour extraction method, system and device to improve the extraction accuracy of the corneal contour.

[0005] To achieve the above purpose, the present application provides the following solutions:

[0006] In a first aspect, the present application provides an unsupervised corneal contour extraction method, comprising:

[0007] obtaining a set of sample corneal videos; the set of sample corneal videos includes a plurality of sample corneal videos arranged in sequence;

[0008] SLIC superpixel segmentation is performed on each frame of sample corneal images in each sample corneal video to obtain a plurality of pixel point groups and corresponding grouping labels;

[0009] If the sample corneal image is the first frame in the first sample corneal video, the sample corneal image and the grouping labels corresponding to the plurality of pixel point groups are input into an initial corneal contour extraction model for training; the initial corneal contour extraction model comprises a preset corneal region extraction module and a preset corneal contour extraction module; the preset corneal region extraction module is a convolutional neural network.

[0010] If the sample corneal image is not the first frame in the first sample corneal video, the sample corneal image and the grouping labels corresponding to the plurality of pixel point groups are input into a corneal contour extraction model trained by the previous frame of sample corneal image for training.

[0011] When the loss function value corresponding to the corneal contour extraction model meets a preset condition, the corneal contour extraction model is marked as trained, and the corneal contour pixel points corresponding to the sample corneal image are output; the loss function value is used to represent the loss value between the predicted clustering labels of the pixel point groups output by the corneal contour extraction model and the grouping labels.

[0012] Optionally, SLIC superpixel segmentation is performed on each frame of sample corneal images in each sample corneal video to obtain a plurality of pixel point groups and corresponding grouping labels, specifically comprising:

[0013] The sample corneal image is converted to a CIELAB color space, and the number of clustering centers and the distance metric are determined;

[0014] Based on the distance metric, the pixel points in the sample corneal image are assigned to a plurality of clustering centers to obtain a plurality of pixel point groups;

[0015] Each pixel point in each pixel point group is assigned the same label to determine the grouping label corresponding to each pixel point group.

[0016] Optionally, the preset corneal region extraction module comprises, in sequence, a fully convolutional network module, a first convolutional block, a first normalization layer, an Argmax classification layer, and an output layer; the fully convolutional network module comprises a plurality of convolutional units connected in sequence.

[0017] Optionally, the convolutional unit comprises, in sequence, a second convolutional block, a second normalization layer, a first ReLU activation layer, a third convolutional block, a third normalization layer, and a second ReLU activation layer.

[0018] Optionally, the sample corneal image and the grouping label corresponding to each pixel group are input into an initial corneal contour extraction model for training, specifically including:

[0019] The sample corneal image is input into a preset corneal region extraction module to obtain a corneal image detail feature mapping;

[0020] Based on the corneal image detail feature mapping, a predicted clustering label of each pixel in the sample corneal image is determined;

[0021] Based on the predicted clustering label corresponding to each pixel in a pixel group, the predicted clustering label with the largest number is selected as a group predicted clustering label;

[0022] Based on the group predicted clustering label and the grouping label, a loss function value is calculated;

[0023] If the loss function value meets a preset condition, the model training is stopped;

[0024] If the loss function value does not meet the preset condition, the grouping label corresponding to the pixel group is updated to the group predicted clustering label;

[0025] The preset corneal region extraction module is updated in neural network parameter optimization by using the updated grouping label of the pixel group;

[0026] The plurality of grouping label regions output by the preset corneal region extraction module after the neural network parameter optimization are input into a preset corneal contour extraction module for corneal contour extraction to obtain a predicted corneal contour pixel point.

[0027] Optionally, the preset corneal contour extraction module is configured to:

[0028] Calculate a shape factor corresponding to each grouping label region;

[0029] Based on a preset corneal shape factor, a grouping label region with the smallest difference between the shape factor and the preset corneal shape factor is selected from the plurality of grouping label regions and marked as a to-be-processed corneal region;

[0030] Based on the preset corneal shape factor, a ratio of the circumference to the area of the cornea is calculated;

[0031] Based on the ratio of the circumference to the area of the cornea, an actual corneal region is selected from the to-be-processed corneal region;

[0032] A boundary of the actual corneal region is extracted to obtain the predicted corneal contour pixel point.

[0033] In a second aspect, the present application provides an unsupervised corneal contour extraction system, comprising:

[0034] a sample cornea video set, wherein the sample cornea video set comprises a plurality of sample cornea videos arranged in sequence;

[0035] a SLIC superpixel segmentation module, configured to perform SLIC superpixel segmentation on each frame of sample cornea image in each of the sample cornea videos to obtain a plurality of pixel point groups and corresponding grouping labels;

[0036] a first model training module, configured to input the sample cornea image and the grouping labels corresponding to the plurality of pixel point groups into an initial cornea contour extraction model for training when the sample cornea image is the first frame in the first sample cornea video, wherein the initial cornea contour extraction model comprises a preset cornea region extraction module and a preset cornea contour extraction module, and the preset cornea region extraction module is a convolutional neural network;

[0037] a second model training module, configured to input the sample cornea image and the grouping labels corresponding to the plurality of pixel point groups into a cornea contour extraction model trained based on a previous frame of sample cornea image for training when the sample cornea image is not the first frame in the first sample cornea video;

[0038] a model training stopping module, configured to mark the cornea contour extraction model as trained when a loss function value corresponding to the cornea contour extraction model meets a preset condition, and output cornea contour pixel points corresponding to the sample cornea image, wherein the loss function value is used to represent a loss value between predicted clustering labels of pixel point groups output by the cornea contour extraction model and the grouping labels.

[0039] In a third aspect, the present application provides an electronic device comprising a memory and a processor, wherein the memory is configured to store a computer program, and the processor is configured to run the computer program to enable the electronic device to perform the unsupervised cornea contour extraction method.

[0040] According to the embodiments of the present application, the following technical effects are achieved:

[0041] The application discloses an unsupervised corneal contour extraction method, system and device, adopts an unsupervised algorithm to perform corneal contour extraction on corneal images in a to-be-processed corneal video, and obtains corresponding corneal contour pixel points. Wherein, the SLIC superpixel method is adopted to segment sample corneal images, and a plurality of pixel point groups and corresponding grouping labels are obtained, without labeling the sample corneal images in advance, so that manpower is greatly saved and convenience is improved. In the model training process, if the sample corneal image is the first frame in the first sample corneal video, the sample corneal image and the corresponding grouping label are directly input into an initial corneal contour extraction model for training; if the sample corneal image is not the first frame in the first sample corneal video, the corneal contour extraction model trained by the last sample corneal image is used for training, so that the corneal contour extraction based on the model sharing mechanism is realized, the accuracy of semantic segmentation is improved, the robustness of pixel segmentation is enhanced, and the training speed of the model is accelerated. In addition, the initial corneal contour extraction model comprises a preset corneal region extraction module and a preset corneal contour extraction module, the corneal contour extraction module is separately set to enhance the attention to the corneal shape, effectively screen the corneal pixel region, and improve the corneal contour extraction precision. BRIEF DESCRIPTION OF DRAWINGS

[0042] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments. Obviously, the drawings in the following description only constitute some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor.

[0043] Figure 1 It is a structural schematic diagram of the unsupervised corneal contour extraction method of the present application.

[0044] Figure 2 It is a flowchart of the present application.

[0045] Figure 3 It is a structural schematic diagram of the full convolution network module of the present application.

[0046] Figure 4 It is a structural schematic diagram of the unsupervised corneal contour extraction system of the present application. DETAILED DESCRIPTION

[0047] The technical solutions in the embodiments of the present application will be described clearly and completely in combination with the drawings in the embodiments of the present application. Obviously, the described embodiments only constitute some embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the protection scope of the present application.

[0048] The present application provides an unsupervised corneal contour extraction method, system and device, based on a model sharing mechanism, which realizes the extraction of corneal contour in corneal stress deformation video, without manual annotation of corneal contour, and even in the case of low video quality or serious noise mixing, the corneal contour can still be extracted completely and accurately.

[0049] In order to make the above-mentioned purposes, characteristics and advantages of the present application more obvious and easy to understand, the present application will be further described in detail below in combination with the drawings and specific embodiments.

[0050] Embodiment one

[0051] As shown in Figure 1 and Figure 2 The present application provides an unsupervised corneal contour extraction method, comprising:

[0052] Step 100, obtaining a sample corneal video set; the sample corneal video set comprises a plurality of sample corneal videos arranged in sequence, and each sample corneal video set is a corneal dynamic deformation video stream.

[0053] Step 200, performing SLIC superpixel segmentation on each frame of sample corneal image in each sample corneal video to obtain a plurality of pixel point groups and corresponding grouping labels. That is, the pre-grouping of pixels is performed according to the spatial continuity of pixels in the corneal stress deformation video.

[0054] Step 200, specifically comprising:

[0055] 1) converting the sample corneal image to CIELAB color space and determining the number of clustering centers and distance measure. Specifically, for N sample corneal video sets, the distance measure is constructed as follows:

[0056] Wherein m is a fixed constant, which is 10 here; S represents the expected side length of superpixel, S=sqrt(N / K), N is the number of all pixel points in the sample corneal image; K is the number of pre-grouping, that is, the number of clustering centers; d c is the color distance in CIELAB color space, d s represents the spatial distance, i and j represent two pixel points in the sample corneal image.

[0057] 2) Based on the distance measure, the pixel points in the sample corneal image are assigned to a plurality of clustering centers to obtain a plurality of pixel point groups.

[0058] 3) assigning the same label to each pixel in each of the pixel groups to determine a group label corresponding to each of the pixel groups. Specifically, by distance metric D', the same label is assigned to spatially continuous pixels with similar color or texture features, and different labels are assigned to adjacent pixels that are not similar, to obtain K pixel index sets Wherein, the assigned label can be defined by relevant personnel, and generally includes corneal region, non-corneal region, etc.

[0059] Step 300, if the sample corneal image is the first frame in the first sample corneal video, input the sample corneal image and the group labels corresponding to the plurality of pixel groups into an initial corneal contour extraction model for training; the initial corneal contour extraction model includes a preset corneal region extraction module and a preset corneal contour extraction module; the preset corneal region extraction module is a convolutional neural network.

[0060] Step 400, if the sample corneal image is not the first frame in the first sample corneal video, input the sample corneal image and the group labels corresponding to the plurality of pixel groups into the corneal contour extraction model trained by the last sample corneal image for training. Specifically, the last sample corneal image can be an image frame in the same sample corneal video, or the last image frame in the last sample corneal video.

[0061] By loading the model parameters of the previous frame to initialize the convolutional network of the adjacent frame image, the network is trained once to obtain an ideal corneal image semantic segmentation effect, fully utilizes the known training model, and only needs to be iterated a few times to obtain a complete and accurate corneal region, thereby effectively saving the training time of the model. In addition, initializing the network with the effective model parameters of the previous frame can also effectively avoid the interference of random noise data in the current frame, and has good noise resistance.

[0062] Wherein, the preset corneal region extraction module can be an initially set convolutional neural network, or a convolutional neural network pre-trained for a corneal region. As shown in the figure, Figure 2 The preset corneal region extraction module includes a fully convolutional network module, a first convolutional block (1x1), a first batch normalization layer (BatchNorm), an Argmax classification layer and an output layer (Output) connected in sequence. The fully convolutional network module includes a plurality of convolutional units connected in sequence, and the convolutional unit includes a second convolutional block (3x3), a second batch normalization layer (BatchNorm), a first ReLU activation layer, a third convolutional block (1x1), a third batch normalization layer (BatchNorm) and a second ReLU activation layer connected in sequence. In this embodiment, as shown in the figure, Figure 3As shown, the full convolution network module includes six convolution units connected in sequence.

[0063] In one specific embodiment, the sample corneal image and the grouping label corresponding to the pixel group are input into an initial corneal contour extraction model for training, specifically including:

[0064] 1) The sample corneal image is input into a preset corneal region extraction module to obtain a corneal image detail feature mapping Wherein, The nth sample corneal image in the mth sample corneal video set is represented.

[0065] 2) Based on the corneal image detail feature mapping, the predicted clustering label of each pixel in the sample corneal image is determined; specifically, the dimension with the maximum value is selected to obtain the clustering label of each pixel That is That is, each pixel is assigned to the nearest point.

[0066] 3) Based on the predicted clustering label corresponding to each pixel in a pixel group, the predicted clustering label with the largest number is selected as the group predicted clustering label.

[0067] 4) Based on the group predicted clustering label and the grouping label, the loss function value is calculated.

[0068] 5) If the loss function value meets the preset condition, the model training is stopped; if the loss function value does not meet the preset condition, the grouping label corresponding to the pixel group is updated to the group predicted clustering label.

[0069] 6) The preset corneal region extraction module is updated in neural network parameter optimization by using the updated grouping label of the pixel group; that is, the network parameters W and b are updated by using the predicted pixel label.

[0070] 7) The multiple grouping label regions output by the preset corneal region extraction module after neural network parameter optimization are input into a preset corneal contour extraction module for corneal contour extraction to obtain predicted corneal contour pixels, realizing unsupervised semantic segmentation of the corneal pixel region.

[0071] The preset corneal contour extraction module is used to: 1) calculate the shape factor corresponding to each grouping label region; it should be noted that the present application adopts an unsupervised algorithm, and the preset corneal region extraction module can output multiple grouping label regions, and the shape factors calculated by different grouping label regions are different. Specifically, the shape factor is the ratio of the perimeter to the area of the grouping label region.

[0072] 2) based on the preset corneal shape factor, from the plurality of grouping label regions, a grouping label region with the minimum difference between the shape factor and the preset corneal shape factor is screened out as a corneal region to be processed.

[0073] 3) based on the preset corneal shape factor Fs, the ratio of the circumference to the area of the cornea is calculated; specifically, the corneal shape to be extracted is constrained according to the preset corneal shape factor, and the calculation formula is:

[0074]

[0075] wherein L is the circumference of the region surrounded by the contour, S is the area of the region surrounded by the contour, and the ratio of the two is the constraint on the corneal shape to be extracted in the application.

[0076] 4) based on the ratio of the circumference to the area of the cornea, an actual corneal region is screened out from the corneal region to be processed. Since the preset corneal shape factor is a fixed value, the circumference and the area of the specific shape of the cornea are also a fixed ratio, and based on the fixed ratio, the actual corneal region is screened out from the grouping label region output by the preset corneal region extraction module.

[0077] 5) boundary extraction is performed on the actual corneal region to obtain the predicted corneal contour pixel point.

[0078] The preset corneal shape factor is set, the attention to the specific shape of the cornea in the training process is enhanced, the corneal pixel region is effectively screened, and the corneal contour pixel point is extracted.

[0079] Step 500, when the loss function value corresponding to the corneal contour extraction model meets the preset condition, the corneal contour extraction model is marked as training completed, and the corneal contour pixel point corresponding to the sample corneal image is output; the loss function value is used to represent the loss value between the predicted clustering label of the pixel point group output by the corneal contour extraction model and the grouping label.

[0080] In summary, in the training of the corneal contour extraction model, the application first performs SLIC superpixel segmentation on each frame image in the sample corneal video, then extracts low-order and high-order features of the image according to the geometric shape characteristics of the cornea and texture similarity information, assigns different labels to each pixel region in each frame image in the corneal stress deformation video, and realizes unsupervised semantic segmentation of the corneal image. Unlike supervised methods, the application does not require a large amount of labeled corneal contour data, but can automatically and accurately extract the corneal contour according to the geometric shape, spatial position and texture feature information of the cornea. When training the neural network, the continuity of each type of pixel region between frames of the video is used to design a model sharing mechanism, fully utilize the known training model, initialize the convolutional network of the adjacent frame image using the model parameters trained in the previous frame, speed up the segmentation speed of each type of pixel region in the image, effectively avoid the interference of each type of noise data, and improve the robustness of the segmentation. Finally, a corneal shape factor is designed to enhance the attention to the specific shape of the cornea in the model training process, effectively filter the corneal pixel region, and improve the accuracy of the corneal contour extraction.

[0081] Embodiment Two

[0082] As shown in Figure 4 To achieve the technical solutions in Embodiment One and achieve the corresponding functions and technical effects, the present application provides an unsupervised corneal contour extraction system, which comprises:

[0083] A sample acquisition module is configured to acquire a sample corneal video set, wherein the sample corneal video set comprises a plurality of sample corneal videos arranged in sequence.

[0084] A SLIC superpixel segmentation module is configured to perform SLIC superpixel segmentation on each frame of sample corneal image in each of the sample corneal videos to obtain a plurality of pixel point groups and corresponding grouping labels.

[0085] A first model training module is configured to, when the sample corneal image is the first frame in the first sample corneal video, input the sample corneal image and the grouping labels corresponding to the plurality of pixel point groups into an initial corneal contour extraction model for training, wherein the initial corneal contour extraction model comprises a preset corneal region extraction module and a preset corneal contour extraction module, and the preset corneal region extraction module is a convolutional neural network.

[0086] A second model training module is configured to, when the sample corneal image is not the first frame in the first sample corneal video, input the sample corneal image and the grouping labels corresponding to the plurality of pixel point groups into the corneal contour extraction model trained by the previous frame of sample corneal image for training.

[0087] The model training stopping module is configured to mark the corneal contour extraction model as training completed and output the corneal contour pixel points corresponding to the sample corneal image when a loss function value corresponding to the corneal contour extraction model meets a preset condition, and the loss function value is used to represent a loss value between a predicted clustering label of a pixel point group output by the corneal contour extraction model and the grouping label.

[0088] Embodiment three

[0089] The embodiment provides an electronic device, including a memory and a processor, the memory is used for storing a computer program, and the processor runs the computer program to make the electronic device execute the unsupervised corneal contour extraction method of embodiment one.

[0090] Optionally, the electronic device can be a server.

[0091] In addition, the embodiment of the present application also provides a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to realize the unsupervised corneal contour extraction method of embodiment one.

[0092] Embodiments of the present application can be provided as methods, systems, or computer program products. Therefore, the present application can take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROMs, optical storage, etc.) containing computer-usable program code.

[0093] The present application is described with reference to flowcharts and / or block diagrams of the method, device (system), and computer program product according to the embodiments of the present application. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and the combination of flows and / or blocks in the flowcharts and / or block diagrams can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing apparatus to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing apparatus produce a device that implements the flow Figure 1 The function of one flow or multiple flows and / or blocks Figure 1 The function of one flow or multiple flows and / or blocks

[0094] These computer program instructions can also be stored in a computer readable memory capable of guiding a computer or other programmable data processing apparatus to work in a specific manner, so that the instructions stored in the computer readable memory produce a product including instruction means, which implements the flow Figure 1 The function of one flow or multiple flows and / or blocksFigure 1 the function(s) specified in the block or blocks.

[0095] These computer program instructions can also be loaded into computer or other programmable data processing devices, so that a series of operation steps are performed on the computer or other programmable data processing devices to generate computer-implemented processes, thus the instructions executed on the computer or other programmable data processing devices provide a process for implementing the flowchart Figure 1 the flowchart or flowcharts and / or a block Figure 1 the function(s) specified in the block or blocks.

[0096] The various embodiments described in the specification are progressive in nature, and each embodiment highlights the differences from other embodiments, and the same and similar parts between the various embodiments can be mutually referred to. For the system disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple, and the relevant parts can be referred to the method part.

[0097] The principles and implementation manners of the present application are described by using specific examples in the present application, and the above embodiment description is only used to help understand the method of the present application and its core idea; meanwhile, for the general skilled in the art, the specific implementation manners and application ranges will be changed according to the idea of the present application. In conclusion, the content of the specification should not be understood as the limitation of the present application.

Claims

1. An unsupervised corneal profile extraction method, characterized in that, The method comprises: acquiring a sample cornea video set; the sample cornea video set comprises a plurality of sample cornea videos arranged in sequence; for each frame of sample cornea image in each sample cornea video, SLIC superpixel segmentation is performed to obtain a plurality of pixel point groups and corresponding grouping labels; if the sample cornea image is the first frame in the first sample cornea video, the sample cornea image and the grouping labels corresponding to the plurality of pixel point groups are input into an initial cornea contour extraction model for training; the initial cornea contour extraction model comprises a preset cornea region extraction module and a preset cornea contour extraction module; the preset cornea region extraction module is a convolutional neural network; if the sample cornea image is not the first frame in the first sample cornea video, the sample cornea image and the grouping labels corresponding to the plurality of pixel point groups are input into the cornea contour extraction model trained by the previous frame of sample cornea image for training; when the loss function value corresponding to the cornea contour extraction model meets a preset condition, the cornea contour extraction model is marked as training completed, and the cornea contour pixel points corresponding to the sample cornea image are output; the loss function value is used to represent the loss value between the predicted clustering labels of the pixel point groups output by the cornea contour extraction model and the grouping labels; the preset cornea contour extraction module is used to: calculate the shape factor corresponding to each grouping label region; based on a preset cornea shape factor, filter out a grouping label region with the smallest difference between the shape factor and the preset cornea shape factor from the plurality of grouping label regions, and mark it as a cornea region to be processed; based on the preset cornea shape factor, calculate the ratio of the circumference to the area of the cornea; based on the ratio of the circumference to the area of the cornea, filter out an actual cornea region from the cornea region to be processed; and perform boundary extraction on the actual cornea region to obtain predicted cornea contour pixel points; the calculation formula of the preset cornea shape factor Fs is: wherein L is the circumference of the region surrounded by the contour, and S is the area of the region surrounded by the contour.

2. The unsupervised corneal topography extraction method of claim 1, wherein, For each frame of sample cornea image in each sample cornea video, SLIC superpixel segmentation is performed to obtain a plurality of pixel point groups and corresponding grouping labels, which specifically comprises: convert the sample cornea image to CIELAB color space, and determine the number of clustering centers and distance metrics; based on the distance metrics, assign the pixel points in the sample cornea image to a plurality of clustering centers to obtain a plurality of pixel point groups; assign the same label to each pixel point in each pixel point group to determine the grouping label corresponding to each pixel point group.

3. The unsupervised corneal topography extraction method of claim 1, wherein, The preset cornea region extraction module comprises a fully convolutional network module, a first convolutional block, a first normalization layer, an Argmax classification layer and an output layer connected in sequence; the fully convolutional network module comprises a plurality of convolutional units connected in sequence.

4. The unsupervised corneal topography extraction method of claim 3, wherein, The convolutional unit comprises a second convolutional block, a second normalization layer, a first ReLU activation layer, a third convolutional block, a third normalization layer and a second ReLU activation layer connected in sequence.

5. The unsupervised corneal topography extraction method of claim 2, wherein, The sample corneal image and the grouping labels corresponding to the pixel point groups are input into an initial corneal contour extraction model for training, specifically including: The sample corneal image is input into a preset corneal region extraction module to obtain a corneal image detail feature mapping; Based on the corneal image detail feature mapping, a predicted clustering label of each pixel in the sample corneal image is determined; Based on the predicted clustering labels corresponding to the pixels in a pixel point group, the predicted clustering label with the largest number is selected as a group predicted clustering label; Based on the group predicted clustering label and the grouping label, a loss function value is calculated; If the loss function value meets a preset condition, the model training is stopped; If the loss function value does not meet the preset condition, the grouping label corresponding to the pixel point group is updated to the group predicted clustering label; The preset corneal region extraction module is updated in neural network parameter optimization by using the updated grouping label of the pixel point group; The multiple grouping label regions output by the preset corneal region extraction module after the neural network parameter optimization are input into a preset corneal contour extraction module for corneal contour extraction to obtain a predicted corneal contour pixel point.

6. An unsupervised corneal profile extraction system characterized in that, The system comprises: A sample acquisition module configured to acquire a sample corneal video set; the sample corneal video set comprises a plurality of sample corneal videos arranged in sequence; An SLIC superpixel segmentation module configured to perform SLIC superpixel segmentation on each sample corneal image in each sample corneal video to obtain a plurality of pixel point groups and corresponding grouping labels; A first model training module configured to, when the sample corneal image is the first frame in the first sample corneal video, input the sample corneal image and the grouping labels corresponding to the pixel point groups into an initial corneal contour extraction model for training; the initial corneal contour extraction model comprises a preset corneal region extraction module and a preset corneal contour extraction module; the preset corneal region extraction module is a convolutional neural network; A second model training module configured to, when the sample corneal image is not the first frame in the first sample corneal video, input the sample corneal image and the grouping labels corresponding to the pixel point groups into a corneal contour extraction model trained based on a previous sample corneal image for training; A model training stop module configured to, when a loss function value corresponding to the corneal contour extraction model meets a preset condition, mark the corneal contour extraction model as trained and output a corneal contour pixel point corresponding to the sample corneal image; the loss function value is used to represent a loss value between a predicted clustering label of a pixel point group output by the corneal contour extraction model and the grouping label; The preset corneal contour extraction module is configured to: Calculate a shape factor corresponding to each grouping label region; Based on a preset corneal shape factor, filter a grouping label region with the smallest difference between the shape factor and the preset corneal shape factor from the multiple grouping label regions and mark the grouping label region as a corneal region to be processed; Based on the preset corneal shape factor, calculate a ratio of the circumference to the area of the cornea; Filtering an actual cornea region from the to-be-processed cornea region based on a ratio of a perimeter to an area of the cornea, and performing boundary extraction on the actual cornea region to obtain a predicted cornea contour pixel point. A preset cornea shape factor Fs has a calculation formula as follows: Wherein, L is a perimeter of a region surrounded by the contour, and S is an area of the region surrounded by the contour.

7. An electronic device, comprising: The electronic device comprises a memory and a processor. The memory is configured to store a computer program. The processor is configured to execute the computer program to enable the electronic device to perform the unsupervised cornea contour extraction method in any one of claims 1-5.

Images