Lacrimal river height detection method, device and equipment and storage medium

Abstract

The invention discloses a lacrimal river height detection method, device and equipment and a storage medium, and relates to the technical field of medical image.The method comprises the steps that a to-be-detected eye image is acquired, and the to-be-detected eye image is preprocessed through a preset gray level co-occurrence matrix enhancement algorithm; the preprocessed to-be-detected eye image is input into a preset lacrimal river segmentation model, a lacrimal river segmentation result is obtained, the preset lacrimal river segmentation model is constructed based on a Unet network, and a preset encoder and a preset decoder in the preset lacrimal river segmentation model each comprise a double-layer convolution layer replacing an original convolution layer; and determining a plurality of measurement key points based on the lacrimal river segmentation result, and determining a lacrimal river height value according to the height information corresponding to each measurement key point. According to the method, the image is preprocessed by adopting the preset gray level co-occurrence matrix enhancement algorithm, and the tear river segmentation model is constructed based on the improved Unet network, so that the accuracy and reliability of tear river height detection can be effectively improved.

Description

Technical Field

[0001] The present application relates to the field of medical imaging technology, and in particular to a tear meniscus height detection method, device, equipment and storage medium. Background Art

[0002] The tear meniscus height is the height of the plane formed by the tears, upper and lower eyelids and cornea, which can reflect the amount of tear gland secretion. At present, the detection of tear meniscus height usually relies on manual operation and traditional image processing methods, which mainly determine the tear meniscus height through manual measurement and edge detection-based algorithms.

[0003] However, manual operation takes a lot of time, and the detection results are affected by subjective factors. Traditional image processing methods are sensitive to lighting and image quality. Therefore, when using the existing tear meniscus height detection method to process images of different qualities, the reliability of the detected tear meniscus height is low. Summary of the invention

[0004] The main purpose of this application is to provide a tear meniscus height detection method, device, equipment and storage medium, aiming to solve the technical problem that the reliability of the tear meniscus height detected when processing images of different qualities using the existing tear meniscus height detection method is low.

[0005] To achieve the above objectives, the present application proposes a tear meniscus height detection method, the tear meniscus height detection method comprising:

[0006] Acquire an eye image to be detected, and preprocess the eye image to be detected using a preset gray-level co-occurrence matrix enhancement algorithm;

[0007] Input the preprocessed eye image to be detected into a preset tear river segmentation model to obtain a tear river segmentation result, wherein the preset tear river segmentation model is constructed based on a Unet network, and the preset tear river segmentation module includes a preset encoder and a preset decoder, and both the preset encoder and the preset decoder include a double-layer convolution layer that replaces the original convolution layer;

[0008] A plurality of measurement key points are determined based on the tear manifold segmentation result, and the tear manifold height value of the eye image to be detected is determined according to the height information corresponding to each of the measurement key points.

[0009] In one embodiment, the step of obtaining an eye image to be detected and preprocessing the eye image to be detected using a preset gray level co-occurrence matrix enhancement algorithm includes:

[0010] Acquire an eye image to be detected, wherein the eye image to be detected includes a corneal area and a tear meniscus area;

[0011] Performing image enhancement on the eye image to be detected by using a preset Placido ring reflection enhancement algorithm to obtain the eye image to be detected with corneal region projection ring enhancement;

[0012] The eye image to be detected with the projection ring enhanced in the corneal region is enhanced by using a preset gray-level co-occurrence matrix enhancement algorithm to obtain the eye image to be detected with the texture enhanced in the tear meniscus region.

[0013] In one embodiment, before the step of inputting the preprocessed eye image to be detected into a preset tear manifold segmentation model to obtain the tear manifold segmentation result, the step includes:

[0014] Initialize the tear river segmentation model to be trained;

[0015] Collecting a number of eye images, and generating an initial training set when the annotation of each of the eye images is completed;

[0016] Performing image preprocessing on each training sample in the initial training set to obtain a model training set;

[0017] The tear river segmentation model to be trained is trained using the model training set to obtain a preset tear river segmentation model.

[0018] In one embodiment, the step of initializing the tear river segmentation model to be trained includes:

[0019] The model architecture of the tear river segmentation model is initialized using the Unet network;

[0020] The model parameters of the tear river segmentation model are defined to obtain the tear river segmentation model to be trained, wherein the model parameters include: the number of input channels, the number of output channels, the number of basic channels, and the number of downsampling times.

[0021] In one embodiment, the step of using the model training set to train the tear manifold segmentation model to be trained to obtain a preset tear manifold segmentation model includes:

[0022] Based on a cross-validation method, the tear river segmentation model to be trained is trained using the model training set;

[0023] Based on the training results, hyperparameter tuning is used to determine the target optimal model parameters to obtain the preset tear river segmentation model.

[0024] In one embodiment, the preset tear meniscus segmentation model further includes an output layer, and the step of inputting the preprocessed eye image to be detected into the preset tear meniscus segmentation model to obtain the tear meniscus segmentation result includes:

[0025] Down-sampling the pre-processed eye image to be detected by using a preset encoder to obtain a down-sampled feature map;

[0026] Performing upsampling processing on the downsampled feature map through a preset decoder to obtain an upsampled feature map, and concatenating the upsampled feature map with the corresponding downsampled feature map to obtain a target feature map;

[0027] The target feature map is converted through the output layer, and the tear river segmentation result is obtained according to the conversion result.

[0028] In one embodiment, the step of determining a plurality of measurement key points based on the tear manifold segmentation result, and determining the tear manifold height value of the eye image to be detected according to the height information corresponding to each of the measurement key points, includes:

[0029] Locate the segmentation edge line of the tear meniscus region in the eye image to be detected, and perform sampling from the center position of the segmentation edge line to the two end positions based on a preset interval to obtain a number of measurement key points;

[0030] The measured height value of each of the measurement key points is obtained, and a linear difference algorithm is used to determine the tear meniscus height value of the eye image to be detected according to each of the measured height values.

[0031] In addition, to achieve the above purpose, the present application also proposes a tear meniscus height detection device, the device comprising:

[0032] An image acquisition module is used to acquire an eye image to be detected and pre-process the eye image to be detected using a preset gray-level co-occurrence matrix enhancement algorithm;

[0033] An image segmentation module, used for inputting the preprocessed eye image to be detected into a preset tear river segmentation model to obtain a tear river segmentation result, wherein the preset tear river segmentation model is constructed based on a Unet network, and the preset tear river segmentation module includes a preset encoder and a preset decoder, and both the preset encoder and the preset decoder include a double-layer convolution layer that replaces the original convolution layer;

[0034] The post-processing module is used to determine a number of measurement key points based on the tear manifold segmentation result, and determine the tear manifold height value of the eye image to be detected according to the height information corresponding to each of the measurement key points.

[0035] In addition, to achieve the above-mentioned purpose, the present application also proposes a tear meniscus height detection device, which includes: a memory, a processor, and a tear meniscus height detection program stored in the memory and executable on the processor, and the tear meniscus height detection program is configured to implement the steps of the tear meniscus height detection method described above.

[0036] In addition, to achieve the above-mentioned purpose, the present application also proposes a storage medium, on which a tear meniscus height detection program is stored. When the tear meniscus height detection program is executed by a processor, the steps of the tear meniscus height detection method described above are implemented.

[0037] The present application discloses a tear meniscus height detection method, which comprises the following steps: obtaining an eye image to be detected, and preprocessing the eye image to be detected by using a preset gray-level co-occurrence matrix enhancement algorithm; inputting the preprocessed eye image to be detected into a preset tear meniscus segmentation model to obtain a tear meniscus segmentation result, wherein the preset tear meniscus segmentation model is constructed based on a Unet network, and the preset tear meniscus segmentation module comprises a preset encoder and a preset decoder, and both the preset encoder and the preset decoder comprise a double-layer convolution layer that replaces an original convolution layer; determining a number of measurement key points based on the tear meniscus segmentation result, and determining the tear meniscus height value of the eye image to be detected according to the height information corresponding to each of the measurement key points. Since the preset gray-level co-occurrence matrix enhancement algorithm is used for image preprocessing in this application, the texture features of the image can be effectively enhanced and the accuracy of subsequent model segmentation can be improved; the double-layer convolution layer is used in the preset tear river segmentation model constructed based on the Unet network, which can deeply extract image features and obtain more accurate tear river segmentation results, so as to finally determine the tear river height value based on the height information corresponding to the measurement key points determined by the segmentation results, thereby realizing the automation of tear river height detection, ensuring the reliability of the detection results while ensuring the detection efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0039] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0040] Figure 1 This is a flow chart of the first embodiment of the tear river height detection method of the present application;

[0041] Figure 2 This is a flow chart of the second embodiment of the tear river height detection method of the present application;

[0042] Figure 3 This is a flow chart of the third embodiment of the tear river height detection method of the present application;

[0043] Figure 4 Schematic diagram of eye images to be detected at different stages in the tear meniscus height detection method of the present application;

[0044] Figure 5 This is a schematic diagram of the module structure of the tear river height detection device of this application;

[0045] Figure 6 This is a schematic diagram of the equipment structure of the hardware operating environment involved in the tear river height detection method of this application.

[0046] The realization of the purpose, functional features and advantages of this application will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0047] It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0048] In order to better understand the technical solution of the present application, a detailed description will be given below in conjunction with the accompanying drawings and specific implementation methods.

[0049] The present application embodiment provides a tear river height detection method, referring to Figure 1 , Figure 1 This is a flow chart of the first embodiment of the tear meniscus height detection method of the present application. In this embodiment, the method includes: Steps S10 to S30:

[0050] Step S10: obtaining an eye image to be detected, and preprocessing the eye image to be detected using a preset gray-level co-occurrence matrix enhancement algorithm.

[0051] It should be noted that the method of this embodiment can be used in scenarios that assist in eye image analysis during medical examinations, or in other scenarios that require analysis of the tear meniscus area of ​​the eye. The executor of the method of this embodiment can be a computing service device with data processing, network communication, and program running functions, such as an eye detector, a dry eye comprehensive detection and analysis instrument, or other electronic devices that are similar or identical to the tear meniscus height detection method of this application. Here, the tear meniscus height detection device (hereinafter referred to as "detection device") is used as an example to illustrate the various embodiments of this application.

[0052] It should be understood that the eye image to be detected may be an eye surface image of a user who needs to undergo eye detection, and the eye image to be detected may include a corneal area and a tear meniscus area of ​​the user's eye.

[0053] It is understandable that the preset gray-level co-occurrence matrix enhancement algorithm can be an image enhancement algorithm based on the gray-level co-occurrence matrix (GLCM), which can extract texture information by analyzing the grayscale relationship between a pixel in an image and its neighboring pixels. Specifically, a threshold can be preset according to the texture feature to be enhanced, and then the elements in the GLCM can be adjusted according to the preset threshold, so that the grayscale value of the eye image can be reconstructed using the adjusted GLCM, thereby enhancing specific texture features.

[0054] In a specific implementation, when receiving the eye image to be detected, the detection device can first perform preprocessing operations such as normalization, image denoising, and image enhancement on the eye image to be detected, thereby improving the image quality; then, a preset GLCM enhancement algorithm can be used to extract the texture features of the tear canal area in the eye image to be detected, and these texture features can be enhanced to improve the contrast and clarity of the image.

[0055] Furthermore, considering that a Placido disk is usually provided in the detection device to obtain corneal information in eye detection, the Placido disk is composed of a series of equidistant concentric rings, so that the corneal area in the eye image to be detected can present a Placido ring texture. Step S10 specifically includes: Steps S101 to S103:

[0056] Step S101: Acquire an eye image to be detected, wherein the eye image to be detected includes a corneal area and a tear meniscus area.

[0057] Step S102: performing image enhancement on the eye image to be detected by using a preset Placido ring reflection enhancement algorithm to obtain the eye image to be detected with corneal region projection ring enhancement.

[0058] It can be understood that the preset Placido ring reflection enhancement algorithm can be an enhancement method set based on the Placido ring texture directional field characteristics, such as contrast enhancement, edge detection, ring positioning and highlighting operations.

[0059] In a specific implementation, first, the eye image to be detected can be subjected to preprocessing operations such as normalization, image denoising, and basic image enhancement. Then, the main texture direction of the projection ring in the corneal area of ​​the eye image to be detected can be determined. Based on the obtained main texture direction, the eye image to be detected can be enhanced by, for example, a high-pass filter.

[0060] Step S103: performing image enhancement on the eye image to be detected with the projection ring enhanced in the corneal region using a preset gray-level co-occurrence matrix enhancement algorithm to obtain the eye image to be detected with the texture enhanced in the tear meniscus region.

[0061] It should be understood that after texture enhancement is performed on the corneal area, texture enhancement can also be performed on the tear meniscus area. Specifically, on the image after Placido ring reflection enhancement, the preset GLCM enhancement algorithm is used to extract the texture features of the tear meniscus area in the eye image to be detected, and image enhancement is performed based on the texture features of the tear meniscus area, thereby further improving the image contrast and completing the preprocessing of the eye image to be detected.

[0062] Step S20: Input the preprocessed eye image to be detected into a preset tear river segmentation model to obtain a tear river segmentation result. The preset tear river segmentation model is constructed based on a Unet network. The preset tear river segmentation module includes a preset encoder and a preset decoder. Both the preset encoder and the preset decoder include a double-layer convolution layer that replaces the original convolution layer.

[0063] It should be understood that the preset tear river segmentation model can be based on an adaptive Unet network as the infrastructure. The Unet network includes an encoder and a decoder. This embodiment improves the original Unet network by using a double-layer convolution layer (DoubleConv layer) to replace the original single-layer convolution layer in the encoder and decoder, which can better capture the subtle features in the image.

[0064] It should be noted that the design of the DoubleConv layer imitates the residual block in the residual network (ResNet). Through two convolution operations, more complex feature representations can be learned while avoiding the problem of gradient disappearance or explosion. In the DoubleConv layer, two convolution operations can be performed on the model input in succession, and each convolution is followed by batch normalization (BatchNorm) and ReLU activation function, thereby improving the linear expression ability of the network, which is conducive to improving the model performance of the preset tear river segmentation model constructed.

[0065] Step S30: determining a number of key measurement points based on the tear manifold segmentation result, and determining the tear manifold height value of the eye image to be detected according to the height information corresponding to each of the key measurement points.

[0066] It should be understood that the tear river segmentation result can be an eye image to be detected with a tear river region segmentation edge line. The detection device can select several key measurement points in the tear river region obtained by segmentation, and then obtain the height values ​​measured at each key measurement point, and obtain the tear river height value of the eye image to be detected by averaging the height values.

[0067] This embodiment uses a preset gray-level co-occurrence matrix enhancement algorithm for image preprocessing, which can effectively enhance the texture features of the image and improve the accuracy of subsequent model segmentation; a double-layer convolution layer is used in the preset tear river segmentation model constructed based on the Unet network, which can deeply extract image features and obtain more accurate tear river segmentation results, thereby finally determining the tear river height value based on the height information corresponding to the measurement key points determined by the segmentation results, realizing the automation of tear river height detection, and ensuring the reliability of the detection results while ensuring the detection efficiency.

[0068] Based on the first embodiment of the present application, in the second embodiment of the present application, the same or similar contents as those in the above-mentioned embodiment 1 can be referred to the above introduction, and will not be repeated in the following. Figure 2 , Figure 2 This is a flow chart of the second embodiment of the tear meniscus height detection method of the present application.

[0069] In this embodiment, in order to specifically illustrate the construction process of the preset tear river segmentation model, before step S20, the following steps are also included: Steps S01 to S04:

[0070] Step S01: Initialize the tear river segmentation model to be trained.

[0071] It should be understood that the model architecture of the tear river segmentation model can be first initialized using the Unet network; then the model parameters of the tear river segmentation model are defined to obtain the tear river segmentation model to be trained, and the model parameters include: the number of input channels, the number of output channels, the number of basic channels, and the number of downsampling times.

[0072] It should also be noted that in this embodiment, an adaptive U-net network can be selected as the basic architecture of the tear river segmentation model. Image segmentation based on the adaptive U-net can improve segmentation accuracy and robustness. Compared with the traditional U-net, the adaptive U-net can better adapt to different image features and improve the processing ability of complex images by dynamically adjusting the convolution kernel and pooling layer parameters.

[0073] Specifically, when the model is initialized, the number of input channels (in_channels), the number of output channels (out_channels), the number of basic channels (base_c), and the number of downsampling times (depth) are used as model parameters. By adjusting base_c, the number of convolution kernels in each layer can be controlled, and by adjusting depth, the depth of Unet can be controlled to adapt to input images of different sizes, and the most suitable model depth can be selected according to the operating environment.

[0074] Step S02: Collect a number of eye images, and generate an initial training set when the annotation of each eye image is completed.

[0075] It should be understood that the eye image can be directly obtained from an open source eye image dataset on the Internet, or can be collected in advance by a detection device.

[0076] It is understandable that after obtaining a large number of eye images, the eye images can be first screened to distinguish between closed eyes and open eyes images, and the open eyes images are used as subsequent training samples. Experts mark the tear meniscus area on each training sample to ensure the accuracy of the obtained training set.

[0077] Step S03: performing image preprocessing on each training sample in the initial training set to obtain a model training set.

[0078] It should be understood that first, basic image preprocessing operations such as normalization, image denoising, and basic image enhancement can be performed on each training sample, and then a preset Placido ring reflection enhancement algorithm is used to focus on enhancing the texture features of the Placido ring reflection image in the corneal area. Finally, a preset GLCM enhancement algorithm is used to extract the texture features of the tear mantle area in the eye image to be detected and enhance it. Thus, through image preprocessing and enhancement technology, the model's adaptability to images of different qualities and its robustness to complex scenes are improved.

[0079] Step S04: using the model training set to train the tear river segmentation model to be trained to obtain a preset tear river segmentation model.

[0080] It should be understood that the tear river segmentation model to be trained can be trained using the model training set based on a cross-validation method; based on the training results, a hyperparameter tuning method is used to determine the target optimal model parameters to obtain a preset tear river segmentation model.

[0081] It can be understood that the cross-validation method evaluates the stability and generalization ability of the model by dividing the data set into multiple subsets and training and validating the model on these subsets. In this embodiment, K-fold cross-validation can be selected to train and test the model.

[0082] Specifically, the model training set can be first randomly divided into K subsets to ensure that the distribution of each subset is roughly the same. Then, in each fold, one of the subsets is used as the test set, and the remaining K-1 subsets are combined into the training set, the model is trained on the training set, and the model performance is evaluated on the test set, and the performance indicators of the model (such as accuracy, recall rate, F1 score, and IoU, etc.) are recorded.

[0083] It can be understood that the hyperparameter tuning method is to systematically search the hyperparameter space of the model to find the best parameter combination to improve the performance of the model on the validation set.

[0084] Specifically, you can first determine the hyperparameters that need to be tuned, such as learning rate, batch size, optimizer type, number of convolution layers, convolution kernel size, etc.; then you can choose an appropriate hyperparameter tuning strategy, such as grid search, random search, and Bayesian optimization; then perform hyperparameter tuning measurements, evaluate each set of hyperparameter combinations, and record the performance indicators of the model.

[0085] It should be understood that the optimal hyperparameter combination can be determined based on the model performance indicators recorded in the above cross-validation method and hyperparameter tuning method; and then the optimal hyperparameter combination is used to perform final model training on the entire model training set to obtain the preset tear river segmentation model.

[0086] This embodiment initializes the tear river segmentation model to be trained; collects a number of eye images, and generates an initial training set when the annotation of each eye image is completed; performs image preprocessing on each training sample in the initial training set to obtain a model training set; uses the model training set to train the tear river segmentation model to be trained to obtain a preset tear river segmentation model. Specifically, the adaptive Unet is used to dynamically adjust the convolution kernel and pooling layer parameters to adapt to different image features, achieve high-precision tear river area segmentation, and significantly improve the accuracy and robustness of detection.

[0087] Based on the first and second embodiments of the present application, in the third embodiment of the present application, the same or similar contents as those of the first and second embodiments can be referred to the above description, and will not be described in detail later. Figure 3 , Figure 3 This is a flow chart of the third embodiment of the tear meniscus height detection method of the present application.

[0088] In this embodiment, in order to specifically illustrate the processing process of the preset tear meniscus segmentation model on the eye image to be detected, the preset tear meniscus segmentation model also includes an output layer, and the step S20 specifically includes steps S201 to S203:

[0089] Step S201: down-sampling the pre-processed eye image to be detected by using a preset encoder to obtain a down-sampled feature map.

[0090] It should be understood that the preset encoder is composed of multiple convolutional layers and pooling layers, each of which is composed of a double convolutional layer (DoubleConv layer). After each convolutional layer, the size of the feature map of the image is halved, and the number of feature channels increases, which helps to extract features of different scales.

[0091] It should be noted that after each convolutional layer of the encoder, Max Pooling can be used for downsampling to reduce the spatial size of the feature map while increasing the number of feature channels.

[0092] It should also be noted that at the end of the encoder, a layer with the largest number of feature channels is usually set, called the bottleneck layer, which captures the high-level feature representation of the input image. At this time, the feature map obtained has the smallest size and the largest depth.

[0093] In the specific implementation, a series of DoubleConv layers in the preset encoder are used to extract features of the eye image to be detected, and Max Pooling is used for downsampling to obtain downsampled feature maps of different layers. Finally, a downsampled feature map with the largest size and depth is obtained at the end of the encoder.

[0094] Step S202: up-sample the down-sampled feature map through a preset decoder to obtain an up-sampled feature map, and concatenate the up-sampled feature map with the corresponding down-sampled feature map to obtain a target feature map.

[0095] It should be understood that the preset decoder gradually restores the spatial size of the feature map while reducing the number of feature channels. Each upsampling step usually includes a transposed convolution layer to increase the size of the feature map.

[0096] It can be understood that after each upsampling, the decoder will splice the features of the corresponding layer with the encoder, that is, the upsampled feature map of the same size from the encoder will be spliced ​​with the corresponding downsampled feature map of the decoder to obtain the target feature map, so that the high-resolution features in the encoder can be combined with the upsampled features in the decoder to restore the spatial information and improve the segmentation accuracy.

[0097] Step S203: converting the target feature map through the output layer, and obtaining the tear river segmentation result according to the conversion result.

[0098] It is understood that at the end of the decoder, i.e., in the output layer, a 1x1 convolutional layer can be used to convert the feature map to the desired number of output channels. For tear river segmentation, the number of output channels is usually 1, generating a binary segmentation map.

[0099] Further, in order to optimize the segmentation result output by the model so as to facilitate the accuracy of the tear river height calculated subsequently, step S30 specifically includes: steps S301 to S302:

[0100] Step S301: Locate the segmentation edge line of the tear meniscus region in the eye image to be detected, and perform sampling from the center position of the segmentation edge line to both ends based on a preset interval to obtain a number of measurement key points.

[0101] Step S302: obtaining the measured height value of each of the measurement key points, and using a linear difference algorithm to determine the tear meniscus height value of the eye image to be detected according to each of the measured height values.

[0102] It should be understood that first, the eye image to be detected can be post-processed by morphological operations and filtering, so as to optimize the segmentation result output by the model by removing noise and artifacts.

[0103] In the specific implementation, for the segmented tear river area, sampling is performed at equal intervals from the center to two sections on the segmentation edge line to obtain a number of measurement key points; then the height value of each measurement key point is measured, and the height values ​​corresponding to each measurement key point are connected through a linear difference algorithm to form a continuous height curve of the tear river, and the continuous height curve can also be smoothed to intuitively obtain the tear river height value information of the eye image to be detected. Among them, the average height value of the tear river area can be obtained by adding the height values ​​corresponding to all measurement key points and dividing it by the number of measurement key points.

[0104] In addition, this can be combined with Figure 4 The whole process of the tear river height detection method of this application is explained. Figure 4 Schematic diagram of eye images to be detected at different stages in the tear meniscus height detection method of the present application.

[0105] exist Figure 4 In the present invention, the eye image A to be detected is subjected to image preprocessing and image enhancement to obtain the preprocessed eye image B to be detected, and then the preprocessed eye image to be detected is input into the preset tear river segmentation model to obtain the tear river area segmentation schematic diagram C, and finally, based on the tear river segmentation schematic diagram, a number of measurement key points are selected in the tear river area segmented in the eye image to be detected, and the height values ​​measured at each measurement key point are obtained respectively to obtain the tear river height detection diagram D, so as to obtain the tear river height value of the eye image to be detected by taking the average of each height value.

[0106] In this embodiment, the pre-processed eye image to be detected is downsampled by a preset encoder to obtain a downsampled feature map; the downsampled feature map is upsampled by a preset decoder to obtain an upsampled feature map, and the upsampled feature map is spliced ​​with the corresponding downsampled feature map to obtain a target feature map; the target feature map is converted by the output layer, and the tear river segmentation result is obtained according to the conversion result. In this embodiment, the entire preset tear river segmentation model utilizes the structural advantages of Unet, extracts features by the encoder and gradually restores the image size in the decoder, and uses a double-layer convolution layer to enhance the feature extraction capability, and finally realizes the accurate segmentation of the tear river area. And further, the segmentation edge line of the tear river area is located in the eye image to be detected, and sampling is performed from the center position of the segmentation edge line to the two end positions based on the preset interval to obtain a number of measurement key points; the measured height value of each of the measurement key points is obtained, and the tear river height value of the eye image to be detected is determined according to each of the measured height values ​​by using a linear difference algorithm. Thus, the segmentation result is optimized by using morphological operations and filtering technology to ensure the accuracy and reliability of the detection result.

[0107] It should be noted that the above examples are only used to understand the present application and do not constitute a limitation on the tear river height detection method of the present application. More simple transformations based on this technical concept are all within the scope of protection of the present application.

[0108] The present application also provides a tear river height detection device, referring to Figure 5 , Figure 5 This is a schematic diagram of the module structure of the tear meniscus height detection device of the present application, and the tear meniscus height detection device includes:

[0109] An image acquisition module 501 is used to acquire an eye image to be detected and pre-process the eye image to be detected using a preset gray level co-occurrence matrix enhancement algorithm;

[0110] An image segmentation module 502 is used to input the preprocessed eye image to be detected into a preset tear river segmentation model to obtain a tear river segmentation result, wherein the preset tear river segmentation model is constructed based on a Unet network, and the preset tear river segmentation module includes a preset encoder and a preset decoder, and both the preset encoder and the preset decoder include a double-layer convolution layer that replaces the original convolution layer;

[0111] The post-processing module 503 is used to determine a number of measurement key points based on the tear manifold segmentation result, and determine the tear manifold height value of the eye image to be detected according to the height information corresponding to each of the measurement key points.

[0112] In one embodiment, the image acquisition module 501 is also used to acquire an eye image to be detected, wherein the eye image to be detected includes a corneal area and a tear meniscus area; the eye image to be detected is enhanced by using a preset Placido ring reflection enhancement algorithm to obtain an eye image to be detected with a projection ring enhancement in the corneal area; the eye image to be detected with a projection ring enhancement in the corneal area is enhanced by using a preset grayscale co-occurrence matrix enhancement algorithm to obtain an eye image to be detected with a texture enhancement in the tear meniscus area.

[0113] In one embodiment, the image segmentation module 502 is also used to initialize the tear manifold segmentation model to be trained; collect a number of eye images, and generate an initial training set when the annotation of each eye image is completed; perform image preprocessing on each training sample in the initial training set to obtain a model training set; use the model training set to train the tear manifold segmentation model to be trained to obtain a preset tear manifold segmentation model.

[0114] In one embodiment, the image segmentation module 502 is also used to initialize the model architecture of the tear river segmentation model using the Unet network; define the model parameters of the tear river segmentation model to obtain the tear river segmentation model to be trained, and the model parameters include: the number of input channels, the number of output channels, the number of basic channels, and the number of downsampling times.

[0115] In one embodiment, the image segmentation module 502 is also used to train the tear river segmentation model to be trained using the model training set based on a cross-validation method; based on the training results, a hyperparameter tuning method is used to determine the target optimal model parameters to obtain a preset tear river segmentation model.

[0116] In one embodiment, the image segmentation module 502 is also used to downsample the preprocessed eye image to be detected through a preset encoder to obtain a downsampled feature map; upsample the downsampled feature map through a preset decoder to obtain an upsampled feature map, and splice the upsampled feature map with the corresponding downsampled feature map to obtain a target feature map; convert the target feature map through the output layer, and obtain a tear river segmentation result based on the conversion result.

[0117] In one embodiment, the post-processing module 503 is used to locate the segmentation edge line of the tear meniscus area in the eye image to be detected, and to perform sampling from the center position of the segmentation edge line to the two end positions based on a preset interval to obtain a plurality of measurement key points; obtain the measurement height value of each of the measurement key points, and use a linear difference algorithm to determine the tear meniscus height value of the eye image to be detected according to each of the measurement height values.

[0118] This embodiment uses a preset gray-level co-occurrence matrix enhancement algorithm for image preprocessing, which can effectively enhance the texture features of the image and improve the accuracy of subsequent model segmentation; a double-layer convolution layer is used in the preset tear river segmentation model constructed based on the Unet network, which can deeply extract image features and obtain more accurate tear river segmentation results, thereby finally determining the tear river height value based on the height information corresponding to the measurement key points determined by the segmentation results, realizing the automation of tear river height detection, and ensuring the reliability of the detection results while ensuring the detection efficiency.

[0119] Other embodiments or specific implementations of the tear meniscus height detection device described in the present application can refer to the above-mentioned method embodiments and will not be repeated here.

[0120] The present application also provides a tear river height detection device, which includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the tear river height detection method in the above-mentioned embodiment one.

[0121] Reference below Figure 6 , Figure 6 This is a schematic diagram of the equipment structure of the hardware operating environment involved in the tear river height detection method of this application.

[0122] like Figure 6As shown, the tear river height detection device may include a processing device 1001 (such as a central processing unit, a graphics processor, etc.), which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM: Read Only Memory) 1002 or a program loaded from a storage device 1003 to a random access memory (RAM: Random Access Memory) 1004. In RAM1004, various programs and data required for the operation of the tear river height detection device are also stored. The processing device 1001, ROM1002 and RAM1004 are connected to each other through a bus 1005. An input / output (I / O) interface 1006 is also connected to the bus. Generally, the following systems can be connected to the I / O interface 1006: an input device 1007 including, for example, a touch screen, a touchpad, a keyboard, a mouse, an image sensor, a microphone, an accelerometer, a gyroscope, etc.; an output device 1008 including, for example, a liquid crystal display (LCD: Liquid Crystal Display), a speaker, a vibrator, etc.; a storage device 1003 including, for example, a magnetic tape, a hard disk, etc.; and a communication device 1009. The communication device 1009 can allow the tear river height detection device to communicate with other devices wirelessly or wired to exchange data. Although the tear river height detection device with various systems is shown in the figure, it should be understood that it is not required to implement or have all the systems shown. More or fewer systems can be implemented or have alternatively.

[0123] The tear river height detection device provided by the present application adopts the tear river height detection method in the above embodiment, which can solve the technical problem that the reliability of the tear river height detected by the existing tear river height detection method is low when processing images of different qualities. Compared with the prior art, the beneficial effects of the tear river height detection device provided by the present application are the same as the beneficial effects of the tear river height detection method provided by the above embodiment, which will not be elaborated here.

[0124] In addition, an embodiment of the present application further proposes a computer-readable storage medium, on which a tear meniscus height detection program is stored. When the tear meniscus height detection program is executed by a processor, the tear meniscus height detection method as described above is implemented.

[0125] The above descriptions are only some embodiments of the present application, and are not intended to limit the patent scope of the present application. All equivalent structural changes made using the contents of the present application specification and drawings under the technical concept of the present application, or direct / indirect applications in other related technical fields are included in the patent protection scope of the present application.

Claims

1. A tear river height detection method, characterized in that: The method comprises: Acquire an eye image to be detected, and preprocess the eye image to be detected using a preset gray-level co-occurrence matrix enhancement algorithm; Input the preprocessed eye image to be detected into a preset tear river segmentation model to obtain a tear river segmentation result, wherein the preset tear river segmentation model is constructed based on a Unet network, and the preset tear river segmentation module includes a preset encoder and a preset decoder, and both the preset encoder and the preset decoder include a double-layer convolution layer that replaces the original convolution layer; A plurality of measurement key points are determined based on the tear manifold segmentation result, and the tear manifold height value of the eye image to be detected is determined according to the height information corresponding to each of the measurement key points.

2. The method according to claim 1, characterized in that The step of obtaining an eye image to be detected and preprocessing the eye image to be detected using a preset gray level co-occurrence matrix enhancement algorithm comprises: Acquire an eye image to be detected, wherein the eye image to be detected includes a corneal area and a tear meniscus area; Performing image enhancement on the eye image to be detected by using a preset Placido ring reflection enhancement algorithm to obtain the eye image to be detected with corneal region projection ring enhancement; The eye image to be detected with the projection ring enhanced in the corneal region is enhanced by using a preset gray-level co-occurrence matrix enhancement algorithm to obtain the eye image to be detected with the texture enhanced in the tear meniscus region.

3. The method according to claim 1, characterized in that Before the step of inputting the preprocessed eye image to be detected into the preset tear manifold segmentation model to obtain the tear manifold segmentation result, the method includes: Initialize the tear river segmentation model to be trained; Collecting a number of eye images, and generating an initial training set when the annotation of each of the eye images is completed; Performing image preprocessing on each training sample in the initial training set to obtain a model training set; The tear river segmentation model to be trained is trained using the model training set to obtain a preset tear river segmentation model.

4. The method according to claim 3, characterized in that The step of initializing the tear river segmentation model to be trained includes: The model architecture of the tear river segmentation model is initialized using the Unet network; The model parameters of the tear river segmentation model are defined to obtain the tear river segmentation model to be trained, wherein the model parameters include: the number of input channels, the number of output channels, the number of basic channels, and the number of downsampling times.

5. The method according to claim 3, characterized in that The step of using the model training set to train the tear river segmentation model to be trained to obtain a preset tear river segmentation model includes: Based on a cross-validation method, the tear river segmentation model to be trained is trained using the model training set; Based on the training results, hyperparameter tuning is used to determine the target optimal model parameters to obtain the preset tear river segmentation model.

6. The method according to claim 1, characterized in that The preset tear river segmentation model also includes an output layer. The step of inputting the preprocessed eye image to be detected into the preset tear river segmentation model to obtain the tear river segmentation result includes: Down-sampling the pre-processed eye image to be detected by using a preset encoder to obtain a down-sampled feature map; Performing upsampling processing on the downsampled feature map through a preset decoder to obtain an upsampled feature map, and concatenating the upsampled feature map with the corresponding downsampled feature map to obtain a target feature map; The target feature map is converted through the output layer, and the tear river segmentation result is obtained according to the conversion result.

7. The method according to claim 1, characterized in that The step of determining a plurality of measurement key points based on the tear manifold segmentation result, and determining the tear manifold height value of the eye image to be detected according to the height information corresponding to each of the measurement key points, comprises: Locate the segmentation edge line of the tear meniscus region in the eye image to be detected, and perform sampling from the center position of the segmentation edge line to the two end positions based on a preset interval to obtain a number of measurement key points; The measured height value of each of the measurement key points is obtained, and a linear difference algorithm is used to determine the tear meniscus height value of the eye image to be detected according to each of the measured height values.

8. A tear river height detection device, characterized in that: The device comprises: An image acquisition module is used to acquire an eye image to be detected and pre-process the eye image to be detected using a preset gray-level co-occurrence matrix enhancement algorithm; An image segmentation module, used for inputting the preprocessed eye image to be detected into a preset tear river segmentation model to obtain a tear river segmentation result, wherein the preset tear river segmentation model is constructed based on a Unet network, and the preset tear river segmentation module includes a preset encoder and a preset decoder, and both the preset encoder and the preset decoder include a double-layer convolution layer that replaces the original convolution layer; The post-processing module is used to determine a number of measurement key points based on the tear manifold segmentation result, and determine the tear manifold height value of the eye image to be detected according to the height information corresponding to each of the measurement key points.

9. A tear river height detection device, characterized in that: The tear meniscus height detection device comprises: a memory, a processor, and a tear meniscus height detection program stored in the memory and executable on the processor. When the tear meniscus height detection program is executed by the processor, the steps of the tear meniscus height detection method as described in any one of claims 1 to 7 are implemented.

10. A storage medium, characterized in that: The storage medium stores a tear meniscus height detection program, and when the tear meniscus height detection program is executed by the processor, the steps of the tear meniscus height detection method according to any one of claims 1 to 7 are implemented.

Images