A method, device and system for training a model for detecting fundus macular lesion areas

Through multiple training and verification of the fundus macular lesions regional detection model, combined with decision-making factors and non-maximum inhibition algorithm, the problem of difficult detection of fundus macular lesions in the existing technology is solved, and efficient and accurate fundus macular lesions regional detection is achieved. .

CN114119588BActive Publication Date: 2024-12-03BEIJING DAHENG PUXIN MEDICAL TECH CO LTD
View PDF 1 Cites 0 Cited by

Patent Information

Application Number
CN202111457655.8
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2021-12-02
Publication Date
2024-12-03
Estimated Expiration
2041-12-02

AI Technical Summary

Technical Problem

In the prior art, there are fewer algorithms for detecting macular lesions in fundus, and fewer fundus data sets are disclosed for detecting macular lesions, making it difficult to train detection models and unable to achieve efficient detection.

Method used

By acquiring multiple fundus image data, establishing a training data set, and through multiple training and verification of model parameters, adjusting the model using decision factors, combining non-maximum suppression algorithms, the detection accuracy and efficiency are improved.

Benefits of technology

It realizes rapid and accurate detection of macular lesions in the fundus, reduces the missed detection rate and false detection rate, and improves the accuracy and generalization ability of the detection model.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN114119588B_ABST
    Figure CN114119588B_ABST
Patent Text Reader

Abstract

Some embodiments of the present application provide a method, device and system for training a fundus macular lesion area detection model, the method comprising: obtaining multiple fundus image data, wherein the fundus image data comprises: original data and annotated area data of each fundus image; obtaining a training data set from the multiple fundus image data; repeating the following training process according to the training data set: inputting the original data of the fundus image in the training data set into the fundus macular lesion area detection model under training to obtain a predicted lesion area; confirming the accuracy of the detection result according to the predicted lesion area, and adjusting the value of the trained parameter according to the accuracy of the detection result. The embodiments of the present application can quickly and accurately detect the fundus macular lesion area and display it to doctors, and have certain application and promotion value.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the field of medical detection technology, and in particular to a method, device and system for training a fundus macular lesion area detection model. Background Art

[0002] The macula is an important area of ​​the retina. It is avascular and located at the posterior pole of the eye. It is mainly related to visual functions such as fine vision and color vision. If there is a lesion in the macula, vision loss and dark shadows in front of the eyes will usually occur. Therefore, it is necessary to have regular fundus examinations to prevent lesions in the macula.

[0003] At present, the inventors of this application have found in their research that there are relatively few research algorithms focusing on macular area lesion detection and there are relatively few publicly available fundus data sets for macular area lesion detection, which makes it difficult to train detection models. As a result, it is currently impossible to accurately, quickly, and batch detect fundus data.

[0004] Therefore, how to provide an efficient technical solution for detecting fundus macular lesion areas has become a technical problem that needs to be solved urgently. Summary of the invention

[0005] The purpose of some embodiments of the present application is to provide a method, device and system for training a fundus macular lesion area detection model. Through the technical solutions of some embodiments of the present application, the fundus macular lesion area detection model is trained, verified and tested multiple times using multiple fundus image data, the model parameters are modified, and the final fundus macular lesion area detection model is obtained, thereby realizing rapid and accurate detection of fundus macular data with high detection efficiency.

[0006] In a first aspect, some embodiments of the present application provide a method for training a fundus macular lesion area detection model, comprising: acquiring multiple fundus image data, wherein the fundus image data includes: original data and annotated area data of each fundus image; obtaining a training data set from the multiple fundus image data; repeating the following training process according to the training data set: inputting the original data of the fundus image in the training data set into the fundus macular lesion area detection model under training to obtain a predicted lesion area; confirming the accuracy of the detection result according to the predicted lesion area, and adjusting the value of the trained parameter according to the accuracy of the detection result, wherein the accuracy is characterized by a decision factor, and if it is confirmed that the predicted lesion area includes the annotated lesion area for the fundus image or it is confirmed that the annotated lesion area of ​​the fundus image includes the predicted lesion area, the value of the decision factor is set to a target value, and the target value is used to characterize that the detection result is accurate, or if it is confirmed that the predicted lesion area and the annotated lesion area for the fundus image partially overlap or have no overlap, the accuracy of the detection result is determined according to the calculated value of the decision factor.

[0007] Since some embodiments of the present application redefine the decision factors, that is, when it is confirmed that the predicted lesion area includes the marked lesion area marked on the image or when it is confirmed that the marked lesion area of ​​the fundus image includes the predicted lesion area, the decision factor is directly assigned a value. The decision factor corresponding to the predicted lesion area that is not an inclusion relationship is obtained by conventional calculation methods. This can effectively reduce the missed detection of lesions. This is because when the predicted lesion area includes the standard lesion area and there is an inclusion relationship, the IOU value of the decision factor calculated directly by the prior art method may be very low. At this time, the relevant technology will regard such a predicted lesion area as a negative sample, and the embodiments of the present application will determine such a detection result as an accurate detection. Therefore, the embodiments of the present application can effectively reduce the missed detection of lesions.

[0008] In some embodiments, the predicted lesion region is confirmed to include the annotated lesion region or the annotated lesion region is confirmed to include the predicted lesion region based on the relationship between the position information of the predicted lesion region and the position information of the annotated lesion region.

[0009] The embodiment of the present application confirms that the predicted lesion area includes the marked lesion area or the marked lesion area includes the predicted lesion area based on the positional relationship between the predicted lesion area and the marked lesion area. The algorithm is simple and can effectively overcome the problem that other methods cannot accurately determine whether two areas are in an inclusion relationship due to the different sizes of lesion areas marked by different experts, thereby effectively improving the speed of training the model.

[0010] In some embodiments, the confirmation that the predicted lesion area includes the annotated lesion area for the fundus image or the confirmation that the annotated lesion area of ​​the fundus image includes the predicted lesion area includes: if the distance between the center position of the predicted lesion area and the center position of the annotated lesion area is less than or equal to a set threshold, and it is confirmed that the boundary of the predicted lesion area is outside or overlaps with the boundary of the annotated lesion area, then the predicted lesion area is confirmed to include the annotated lesion area, or if the distance between the center position of the predicted lesion area and the center position of the annotated lesion area is less than or equal to a set threshold, and it is confirmed that the boundary of the annotated lesion area is outside or overlaps with the boundary of the predicted lesion area, then the annotated lesion area is confirmed to include the predicted lesion area.

[0011] The embodiment of the present application doubly determines the relationship between the two by predicting the distance between the center position of the lesion area and the center position of the marked lesion area and the boundary position of the two areas, further ensuring the accuracy of adjusting the model parameters in training, and ultimately greatly reducing the probability of false detection or wrong detection of the predicted lesion area on the input image to be detected.

[0012] In some embodiments, the confirmation that the predicted lesion area includes the annotated lesion area for the fundus image or the confirmation that the annotated lesion area of ​​the fundus image includes the predicted lesion area includes: including the boundary of the circumscribed polygon corresponding to the predicted lesion area according to the boundary of the circumscribed polygon corresponding to the annotated lesion area, or including the boundary of the circumscribed polygon corresponding to the annotated lesion area according to the boundary of the circumscribed polygon corresponding to the predicted lesion area.

[0013] The embodiment of the present application compares the boundary of the circumscribed polygon corresponding to the predicted lesion area with the boundary of the circumscribed polygon corresponding to the annotated lesion area to confirm the relationship between the predicted lesion area and the annotated lesion area. Since the results of the predicted lesion area and the annotated lesion area are irregular shapes, the use of the circumscribed polygon can more simply, quickly and accurately determine the positional relationship between the two.

[0014] In some embodiments, the boundary of the circumscribed polygon is characterized by the coordinates of the fixed points included in the boundary.

[0015] The embodiment of the present application can intuitively and quickly locate the positional relationship between the predicted lesion area and the marked lesion area's circumscribed polygon by comparing the coordinate information.

[0016] In some embodiments, the confirming that the predicted lesion area and the annotated lesion area for the fundus image partially overlap or do not overlap includes: if the distance between the center position of the predicted lesion area and the center position of the annotated lesion area is greater than a set threshold and it is confirmed that the boundary of the predicted lesion area and the boundary of the annotated lesion area intersect or do not overlap, or the distance between the center position of the predicted lesion area and the center position of the annotated lesion area is greater than a set threshold and it is confirmed that the boundary of the annotated lesion area is outside the boundary of the predicted lesion area or does not overlap, then it is confirmed that the predicted lesion area and the annotated lesion area for the fundus image partially overlap or do not overlap; or, if the distance between the center position of the predicted lesion area and the center position of the annotated lesion area is less than or equal to the set threshold and it is confirmed that the boundary of the predicted lesion area and the boundary of the annotated lesion area intersect or do not overlap, then it is confirmed that the predicted lesion area and the annotated lesion area for the fundus image partially overlap or do not overlap.

[0017] The embodiment of the present application predicts the distance and boundary relationship between the center position of the lesion area and the center position of the marked lesion area, and infers that there is a partial overlap or no overlap between the two areas. This can quickly determine that the predicted lesion area does not include the marked lesion area, providing a basis for adjusting the model parameters.

[0018] In some embodiments, the target value is set in a range of 0.9 to 1.

[0019] The embodiment of the present application directly sets the decision factor of the predicted lesion area including the marked lesion area to a number close to or equal to the maximum value 1 to characterize the accuracy of the prediction result, thereby improving the error caused by the low decision factor obtained according to the conventional calculation method.

[0020] In some embodiments, the multiple fundus image data are divided into the training data set, the verification data set and the test data set; wherein, after repeating the following training process according to the training data set, the method further includes: repeating the following verification process according to the verification data set: inputting the original data of the fundus image included in the verification data set into the fundus macular lesion area detection model under verification to obtain the lesion area to be verified, wherein the fundus macular lesion area detection model under verification corresponding to the first execution of the verification process is the fundus macular lesion area detection model to be verified obtained through the training process; confirming the accuracy of the detection result according to the lesion area to be verified, and adjusting the value of the parameter of the fundus macular lesion area detection model to be verified according to the accuracy of the detection result.

[0021] The embodiment of the present application divides the data set into a training set and a validation set, verifies the trained model through the validation data set and adjusts the model parameters, thereby further improving the prediction accuracy of the obtained model for the lesion area.

[0022] In some embodiments, after repeating the following verification process according to the verification data set, the method further includes: repeating the following test process according to the test data set: inputting the original data of the fundus image in the test data set into the fundus macular lesion area detection model under test to obtain an image to be tested containing at least one detection area, wherein the fundus macular lesion area detection model under test corresponding to the first execution of the test process is the fundus macular lesion area detection model to be tested obtained through the verification process; confirming the image of the detection area to be tested according to at least one confidence level corresponding to the at least one detection area, wherein one confidence level corresponds to one detection area; confirming the accuracy of the detection result according to the image of the detection area to be tested, and adjusting the value of the parameter of the fundus macular lesion area detection model to be tested according to the accuracy of the detection result.

[0023] The embodiment of the present application tests the verified model through a test data set and then confirms the accuracy of the test result in combination with the confidence level, which can better evaluate the generalization ability of the model and further improve the prediction accuracy of the obtained model for the lesion area.

[0024] In some embodiments, the image of the detection area to be tested is confirmed based on at least one confidence level corresponding to the at least one detection area, including: confirming the image of the detection area to be tested based on a non-maximum suppression algorithm and the confidence level, wherein the image of the detection area to be tested includes a fundus macular area and one or more lesion type areas.

[0025] The embodiment of the present application obtains the image of the detection area to be tested based on the non-maximum suppression algorithm and confidence level, so as to ensure that a more reliable test result is selected, thereby reducing the probability of false detection in subsequent model applications and improving the detection accuracy.

[0026] In the second aspect, an embodiment of the present application provides a method for detecting fundus macular lesion areas, comprising: obtaining a fundus macular detection result including at least one detection area and at least one confidence level corresponding to the at least one detection area based on a fundus image to be detected and a fundus macular lesion target detection model obtained by executing the corresponding method of any embodiment in the first aspect; confirming the fundus macular detection result based on a non-maximum suppression algorithm and the confidence level, wherein the fundus macular detection result includes a fundus macular area and one or more lesion type areas.

[0027] In a third aspect, an embodiment of the present application provides a device for training a fundus macular lesion area detection model, comprising: a data acquisition module, configured to acquire multiple fundus image data, wherein the fundus image data includes: original data and annotated area data of each fundus image; a training data set module, configured to obtain a training data set from the multiple fundus image data; a model training module, configured to repeat the following training process according to the training data set: inputting the original data of the fundus image in the training data set into the fundus macular lesion area detection model under training to obtain a predicted lesion area; confirming the accuracy of the detection result according to the predicted lesion area, and adjusting the value of the trained parameter according to the accuracy of the detection result, wherein the accuracy is characterized by a decision factor, and if it is confirmed that the predicted lesion area includes the annotated lesion area for the fundus image or it is confirmed that the annotated lesion area of ​​the fundus image includes the predicted lesion area, the value of the decision factor is set to a target value, and the target value is used to characterize that the detection result is accurate, or if it is confirmed that the predicted lesion area and the annotated lesion area for the fundus image partially overlap or have no overlap, the accuracy of the detection result is determined according to the calculated value of the decision factor.

[0028] In a fourth aspect, an embodiment of the present application provides a system, comprising one or more computers and one or more storage devices storing instructions, wherein when the instructions are executed by the one or more computers, the one or more computers perform the operations of the corresponding methods of any embodiment in the first aspect.

[0029] In a fifth aspect, an embodiment of the present application provides one or more computer storage media storing instructions, which, when executed by one or more computers, enable the one or more computers to perform the operations of the corresponding methods of any embodiment in the first aspect. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the embodiments of the present application will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present application and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0031] Figure 1 A system diagram of a method for detecting fundus macular lesion area using an embodiment of the present application provided in the present application;

[0032] Figure 2 A flow chart of a method for training a fundus macular lesion area detection model provided in an embodiment of the present application;

[0033] Figure 3 A schematic diagram of the annotation boxes of the predicted lesion area and the actual result provided in the embodiment of the present application;

[0034] Figure 4 A flow chart of a method for training a fundus macular lesion area detection model based on an artificial neural network model provided in an embodiment of the present application;

[0035] Figure 5 A schematic diagram of annotating fundus macular lesion data for training an artificial neural network model provided in an embodiment of the present application;

[0036] Figure 6 A schematic diagram of annotating fundus macular lesion-free data for training an artificial neural network model provided in an embodiment of the present application;

[0037] Figure 7 A schematic diagram of the positional relationship between the actual result and the predicted lesion area provided in the embodiment of the present application;

[0038] Figure 8 A schematic diagram of the annotation results of the test images in the test data set provided in the embodiment of the present application;

[0039] Fig. 9 A schematic diagram of the detection result of the image of the detection area to be tested provided in an embodiment of the present application;

[0040] Fig.10 A schematic diagram of the detection results of regions including multiple lesion types provided in an embodiment of the present application;

[0041] Fig.11 A schematic diagram of a lesion area to be tested after treatment of the same category provided in an embodiment of the present application;

[0042] Fig.12 A schematic diagram of the lesion area to be tested after being processed between different categories provided in the embodiment of the present application;

[0043] Fig.13 A block diagram of a device for training a fundus macular lesion area detection model provided in an embodiment of the present application. DETAILED DESCRIPTION

[0044] The technical solutions in the embodiments of the present application will be described below in conjunction with the drawings in the embodiments of the present application.

[0045] It should be noted that similar reference numerals and letters represent similar items in the following drawings, so once an item is defined in one drawing, it does not need to be further defined and explained in subsequent drawings. At the same time, in the description of this application, the terms "first", "second", etc. are only used to distinguish the description and cannot be understood as indicating or implying relative importance.

[0046] In the related art examples, on the one hand, there are few algorithms that focus on detecting fundus macular lesion areas. On the other hand, there are few publicly available fundus data sets for fundus macular lesion detection, and even fewer annotated data. The fundus lesion detection models trained with these data have low detection accuracy, and it is difficult to train models with higher accuracy. It can be seen from the above related technologies that it is currently impossible to achieve efficient detection of fundus macular lesion areas.

[0047] In view of this, some embodiments of the present application provide a method, device and system for training a fundus macular lesion area detection model. The method collects a large amount of fundus data and has multiple experts annotate it as a data set for training the fundus macular lesion area detection model. The training parameters are modified during the training process to improve the detection rate and accuracy of the fundus macular lesion area detection model.

[0048] The following is an exemplary introduction to the process of training a fundus macular lesion area detection model.

[0049] like Figure 1 As described, the figure provides a system diagram of the fundus macular lesion area detection method that can be applied to the embodiment of the present application, and the figure includes an image shooting terminal device 100 and a server detection terminal device 200, wherein the shooting terminal device 100 can be used to shoot a fundus image data 101, and send the shot fundus image data 101 to the server detection terminal device 200 for processing to obtain a predicted lesion area.

[0050] In addition, it should be noted that in some other embodiments of the present application, the shooting terminal device 100 has the functions of shooting and detecting to obtain the predicted lesion area, and the server detection terminal device may not be set. Among them, the image shooting terminal device can be a PC terminal or a mobile terminal.

[0051] Figure 1 The server detection terminal device 200 is provided with a fundus macular lesion area detection model. It is through this trained model that the server detection terminal device 200 can obtain the predicted lesion area. It should be noted that, unlike the method of training a fundus macular lesion area detection model in the related art, Figure 1 The fundus macular lesion area detection model set by the detection terminal device 200 is obtained by training through redefined decision factor parameters.

[0052] It is understandable that in order to enable the fundus macular lesion area detection model on the server to have the ability to predict the lesion area, the model needs to be trained first.

[0053] Combine the following Figure 2 The process of training a fundus macular lesion area detection model provided in some embodiments of the present application is exemplified.

[0054] like Figure 2 As shown, some embodiments of the present application provide a flow chart of a method for training a fundus macular lesion area detection model, the method comprising: S210, acquiring multiple fundus image data, wherein the fundus image data comprises: original data and annotated area data of each fundus image; S220, obtaining a training data set from the multiple fundus image data; S230, repeating the following training process according to the training data set: inputting the original data of the fundus image in the training data set into the fundus macular lesion area detection model under training to obtain a predicted lesion area; confirming the accuracy of the detection result according to the predicted lesion area, and adjusting the value of the trained parameter according to the accuracy of the detection result, wherein the accuracy is characterized by a decision factor, if it is confirmed that the predicted lesion area includes the annotated lesion area for the fundus image or it is confirmed that the annotated lesion area of ​​the fundus image includes the predicted lesion area, then the value of the decision factor is set to a target value, and the target value is used to characterize that the detection result is accurate, or if it is confirmed that the predicted lesion area and the annotated lesion area for the fundus image partially overlap or do not overlap, then the accuracy of the detection result is determined according to the calculated value of the decision factor.

[0055] In some embodiments of the present application, the raw data of the multiple fundus image data in S210 may include multiple images of fundus macular lesions and multiple images of fundus macular non-lesions, and the marked area data may include images of fundus macular lesion areas (i.e., marked lesion areas) marked on each image of fundus macular lesions and images of fundus macular areas (Macula, MAC for short) marked on each image of fundus macular non-lesions. In order to balance the two types of data, the embodiments of the present application use a certain ratio of mixed data with lesion data and data without lesions, wherein only MAC positions are marked on images of fundus macular non-lesions.

[0056] It should be noted that, in some embodiments of the present application, the annotated data in S210 is obtained by collecting a large amount of fundus data and annotating these data by multiple experts. Because different expert doctors have different opinions on the size, severity and category of the same fundus macular lesion to a certain extent. In order to reduce the differentiation of annotated data by different expert doctors, some embodiments of the present application distribute a large amount of fundus data to multiple expert doctors for cross-annotation, so as to obtain high-quality annotated data and improve the accuracy of the training model.

[0057] In some embodiments of the present application, the fundus macular lesion area detection model involved in the training in S210 can be any segmentation network model, such as any one of the artificial neural network Mask-RCNN model, Cascade-RCNN model and gcnet model, which can obtain the rectangular box result of target detection (i.e., output the predicted lesion area of ​​the rectangle) and the segmentation result. Since the shape of the macular lesion is irregular, in some embodiments of the present application, the detection result can be displayed as a polygonal segmentation result.

[0058] The following is an illustrative description of the process of confirming the predicted lesion area involved in S230, including marking the lesion area for the fundus image. In some embodiments of the present application, the positional relationship between the predicted lesion area and the marked lesion area can be directly confirmed based on the boundary position information of the predicted lesion area and the marked lesion area. Alternatively, the positional relationship between the predicted lesion area and the marked lesion area is jointly confirmed based on the center position of the predicted lesion area and the center position of the marked lesion area and the boundary position information of the predicted lesion area and the marked lesion area. In other embodiments of the present application, the positional relationship between the predicted lesion area and the marked lesion area can also be directly confirmed based on the fixed-point coordinate information of the circumscribed polygons of the predicted lesion area and the marked lesion area.

[0059] In order to accurately evaluate the predicted data and improve the accuracy of the trained model, in some embodiments of the present application, S230 can confirm that the predicted lesion area includes the annotated lesion area or confirm that the annotated lesion area includes the predicted lesion area through the relationship between the position information of the predicted lesion area and the position information of the annotated lesion area.

[0060] It should be noted that the predicted lesion area includes the marked lesion area, or confirming that the marked lesion area includes the predicted lesion area can mean that the size and position of the predicted lesion area and the marked lesion area are basically the same, or the predicted lesion area completely includes the marked lesion area, or the marked lesion area completely includes the predicted lesion area.

[0061] In some other embodiments of the present application, the confirmation method of S230 may include: if the distance between the center position of the predicted lesion area and the center position of the marked lesion area is less than or equal to a set threshold, and it is confirmed that the boundary of the predicted lesion area is outside or overlaps with the boundary of the marked lesion area, then confirming that the predicted lesion area includes the marked lesion area, or if the distance between the center position of the predicted lesion area and the center position of the marked lesion area is less than or equal to a set threshold, and it is confirmed that the boundary of the marked lesion area is outside or overlaps with the boundary of the predicted lesion area, then confirming that the marked lesion area includes the predicted lesion area.

[0062] In some other embodiments of the present application, the confirmation method of S230 may further include: including the boundary of the circumscribed polygon corresponding to the annotated lesion area according to the boundary of the circumscribed polygon corresponding to the predicted lesion area, or including the boundary of the circumscribed polygon corresponding to the annotated lesion area according to the boundary of the circumscribed polygon corresponding to the predicted lesion area. The boundary of the circumscribed polygon is characterized by the coordinates of the fixed point included in the boundary.

[0063] For example, in some embodiments of the present application, for the same fundus lesion, the sizes of the areas marked by different expert doctors are different, but the positions of the lesion areas are consistent. By predicting that the distance between the center position of the lesion area and the center position of the marked lesion area is less than or equal to a set threshold and confirming that the boundary of the predicted lesion area is outside the boundary of the marked lesion area, it is determined that the predicted lesion area includes the marked lesion area (that is, the true result, Ground Truth, GT for short), otherwise the two do not belong to an inclusive relationship (that is, the predicted lesion area and the marked lesion area may partially overlap or have no overlap). In addition, the display results of the predicted lesion area and the marked lesion area can be in the form of a circumscribed polygon, and the results all contain the position coordinate information of the marked box. For example, by judging the relationship between the position coordinate information, it is determined that the predicted lesion area includes the true result. For example, if Figure 3 As shown, the coordinate information of the diagonal corners of the rectangular box circumscribed to the predicted lesion area (solid line irregular area) are bbox1(x1, y1) and bbox1(x2, y2), and the coordinate information of the diagonal corners of the rectangular box circumscribed to the true result (dashed line irregular area) are bbox2(s1, t1) and bbox2(s2, t2). According to the information in the figure, the circumscribed rectangular box of the predicted lesion area includes the circumscribed rectangular box of the marked lesion area, and it is determined that the predicted lesion area includes the true result. It can be understood that the collected diagonal coordinate information is not limited to that shown in the figure, and at least two coordinate information of other positions can also be collected to confirm the positional relationship between the predicted lesion area and the true result.

[0064] In some embodiments of the present application, the method for confirming in S230 that the predicted lesion area and the annotated lesion area for the fundus image partially overlap or do not overlap exemplarily includes: if the distance between the center position of the predicted lesion area and the center position of the annotated lesion area is greater than a set threshold and it is confirmed that the boundary of the predicted lesion area and the boundary of the annotated lesion area intersect or do not overlap, or the distance between the center position of the predicted lesion area and the center position of the annotated lesion area is greater than a set threshold and it is confirmed that the boundary of the annotated lesion area is outside the boundary of the predicted lesion area or does not overlap, then it is confirmed that the predicted lesion area and the annotated lesion area for the fundus image partially overlap or do not overlap; or, if the distance between the center position of the predicted lesion area and the center position of the annotated lesion area is less than or equal to the set threshold and it is confirmed that the boundary of the predicted lesion area and the boundary of the annotated lesion area intersect or do not overlap, then it is confirmed that the predicted lesion area and the annotated lesion area for the fundus image partially overlap or do not overlap.

[0065] In some embodiments of the present application, the target value involved in S230 can be set to any value between 0.9 and 1. For example, in some embodiments, the target value can be directly set to 1, or the target value can be set to 0.9, etc.

[0066] In some embodiments of the present application, when it is confirmed that the predicted lesion area includes the annotated lesion area, it can be easily known that the prediction result (i.e., the predicted lesion area) output by the model in training is confirmed to be consistent with the true value (i.e., the annotated lesion area). In order to avoid this situation being judged as a conclusion of detection error, the embodiment of the present application directly sets the decision factor to a larger value (i.e., directly assigning the determined target value to the decision factor corresponding to this situation). It can be understood that the higher the target value, the higher the accuracy. For example, the target value can be set to 1, indicating that the prediction accuracy of the predicted lesion area is high. When it is confirmed that the predicted lesion area partially overlaps or does not overlap with the annotated lesion area, the accuracy of the detection result is determined according to the value of the calculated decision factor, that is, the larger the value, the higher the accuracy. As an example, the decision factor can be represented by Intersection over Union (IOU for short). The value of IOU is 0 to 1, and IOU=0 means that the predicted lesion area does not overlap with the annotated lesion area, and the predicted lesion area result is wrong. IOU=1 means that the predicted lesion area overlaps with the annotated lesion area, and the predicted lesion area result is highly accurate. It can be seen that the larger the value, the more overlap there is between the predicted lesion area and the annotated lesion area, which in turn reflects that the detection effect of the model is better.

[0067] In order to ensure the detection effect of the model in training and improve the detection accuracy of the final model. In some embodiments of the present application, the multiple fundus image data are divided into the training data set, the verification data set and the test data set; wherein, after S230, the method further includes: S240 (not shown in the figure), repeating the following verification process according to the verification data set: inputting the original data of the fundus image included in the verification data set into the fundus macular lesion area detection model in the verification to obtain the lesion area to be verified, wherein the fundus macular lesion area detection model in the verification corresponding to the first execution of the verification process is the fundus macular lesion area detection model to be verified obtained through the training process; confirming the accuracy of the detection result according to the lesion area to be verified, and adjusting the value of the parameter of the fundus macular lesion area detection model to be verified according to the accuracy of the detection result.

[0068] In some embodiments of the present application, multiple fundus image data may be divided according to a set ratio. For example, multiple fundus image data may be divided into a training data set, a verification data set, and a test data set according to a ratio of 7:2:1, and the three data sets may be adjusted to a format file suitable for training, verification, and testing of a fundus macular lesion area detection model. Afterwards, the verification data set is used to verify the fundus macular lesion area detection model to be verified. When the accuracy of the model's detection results remains unchanged, no longer improves, or reaches a preset number of training times, the model training and verification are completed.

[0069] Before actual application, in order to ensure the detection effect of the model. In some embodiments of the present application, after S240, the method further includes: S250 (not shown in the figure), repeating the following test process according to the test data set: inputting the original data of the fundus image in the test data set into the fundus macular lesion area detection model under test, and obtaining an image to be tested containing at least one detection area, wherein the fundus macular lesion area detection model under test corresponding to the first execution of the test process is the fundus macular lesion area detection model to be tested obtained through the verification process; confirming the image of the detection area to be tested according to at least one confidence level corresponding to the at least one detection area, wherein one confidence level corresponds to one detection area; confirming the accuracy of the detection result according to the image of the detection area to be tested, and adjusting the value of the parameter of the fundus macular lesion area detection model to be tested according to the accuracy of the detection result.

[0070] In some embodiments of the present application, when the fundus macular lesion area detection model under test is tested using the test data set in S250, after the original data of the fundus image is input, a test result to be tested containing at least one detection area and at least one confidence level are obtained. According to the analysis of the test results, it can be found that there is inclusion or partial overlap between lesions of the same category or between lesions of different categories. Based on this, the embodiment of the present application adopts post-processing technology (for example, non-maximum suppression algorithm) to remove redundant test results.

[0071] In some embodiments of the present application, in S250, the image of the detection area to be tested can be confirmed according to the non-maximum suppression algorithm and the confidence level, wherein the image of the detection area to be tested includes the fundus macular area and one or more lesion type areas.

[0072] For example, in some embodiments of the present application, in order to reasonably remove duplicate detection results, the confidences corresponding to the lesion type areas can be sorted from large to small or from small to large, and then the non-maximum suppression (NMS) algorithm is used to eliminate unreasonable results to obtain the image of the detection area to be tested. For example, this embodiment obtains 10 detection results, wherein the 10 detection results include lesions of the same category (for example, there are 6 detection results) and lesions of different categories (for example, there are 4 detection results). First, the NMS algorithm is used to process lesions of the same category. The 6 confidences corresponding to the 6 detection results of lesions of the same category are sorted in order from large to small, and then the NMS algorithm is used to calculate the overlap between the detection result with the first confidence in the sorting queue and the remaining 5 detection results to obtain 5 overlaps. The 5 overlaps are compared with the set thresholds respectively. If the overlap is greater than the set threshold, the detection result of the lesion type is eliminated, and if the overlap is less than or equal to the set threshold, the detection result of the lesion type is retained. Secondly, the NMS algorithm is used to process the detection results of different categories of lesions. The processing process of different categories is the same as that of the same category. To avoid repetition, it will not be described here.

[0073] In addition, in some embodiments of the present application, a target detection algorithm (eg, a soft-NMS algorithm) having the same function as the NMS algorithm may also be used to eliminate unreasonable detection results.

[0074] Combine the following Figure 4 Specific explanation Figure 2 The process of training the fundus macular lesion area detection model based on the artificial neural network model.

[0075] Please see attached Figure 4 , Figure 4A flow chart of a method for training a fundus macular lesion area detection model based on an artificial neural network model provided in an embodiment of the present application.

[0076] The above process is explained below as an example.

[0077] The first step is to obtain data and labels.

[0078] A plurality of fundus image data are acquired, wherein the fundus image data includes: original data and annotated area data of each fundus image.

[0079] For example, a large amount of fundus data is collected and cross-labeled by multiple experts to obtain more accurate labeling results. In order to achieve a balance between various types of data, the ratio of lesion data to non-lesion data in the data set is 1:1. The labeled data is Figure 5 and Figure 6 As shown, Figure 5 This is the fundus macular lesion data, which has MAC area and lesion area marked. Figure 6 This is the data of fundus macular lesion-free, in which only the MAC position is marked.

[0080] The second step is to generate fundus data annotation files.

[0081] The multiple fundus image data in the first step are divided into a training data set, a validation data set and a test data set.

[0082] For example, all fundus images are divided into a training set: validation set: test set ratio of 7:2:1, and an annotation file format suitable for the model is generated based on the doctor's annotation results.

[0083] The third step is model training.

[0084] The following training process is repeated according to the training data set: the original data of the fundus image in the training data set is input into the fundus macular lesion area detection model being trained to obtain the predicted lesion area.

[0085] For example, the artificial neural network model is trained using the training set. The fundus data in the training set is used as the input of the artificial neural network model, and the detection result of the predicted lesion area is used as the output of the artificial neural network model to train the artificial neural network model. The center position and boundary position of the predicted lesion area are compared with the center position and boundary position of the true result (i.e., the marked lesion area). If the distance between the center position of the predicted lesion area and the true result is less than or equal to the set threshold and the boundary of the predicted lesion area is outside or overlaps the boundary of the true result, it is confirmed that the predicted lesion area includes the true result, the prediction result is accurate, and the value of IOU (as an example of a decision factor) is set to 1. If the distance between the center position of the predicted lesion area and the center position of the annotated lesion area is greater than a set threshold, and it is confirmed that the boundary of the predicted lesion area and the boundary of the annotated lesion area intersect or do not overlap, or the distance between the center position of the predicted lesion area and the center position of the annotated lesion area is greater than a set threshold, and it is confirmed that the boundary of the annotated lesion area is outside the boundary of the predicted lesion area or does not overlap, then it is confirmed that the predicted lesion area and the annotated lesion area for the fundus image partially overlap or do not overlap; or, if the distance between the center position of the predicted lesion area and the center position of the annotated lesion area is less than or equal to a set threshold, and it is confirmed that the boundary of the predicted lesion area and the boundary of the annotated lesion area intersect or do not overlap, then it is confirmed that the predicted lesion area and the annotated lesion area for the fundus image partially overlap or do not overlap, then the IOU value is calculated according to the conventional method, and the higher the value, the more accurate the result of predicting the lesion area. Figure 7 As shown in the figure, the true result (shown by the white line in the figure) falls within the predicted lesion area (shown by the black line in the figure), and the center positions of the true result and the predicted lesion area are basically consistent. At this time, the IOU value is set to 1, which represents the accuracy of the prediction result.

[0086] In addition, as another example of the present application, the positional relationship between the predicted lesion area and the real result is confirmed based on the fixed-point coordinate information of the circumscribed polygon of the predicted lesion area and the fixed-point coordinate information of the circumscribed polygon of the real result. Alternatively, the predicted lesion area can also be confirmed to include the real result by comparing the distance between the center position of the predicted lesion area output by the artificial neural network model in training and the center position of the real result with the set threshold and the boundary position. When the distance is greater than the set threshold and the boundary of the predicted lesion area and the boundary of the marked lesion area intersect or do not overlap, it is confirmed that the predicted lesion area partially overlaps or does not overlap with the real result, or the distance between the center position of the predicted lesion area and the center position of the marked lesion area is greater than the set threshold and it is confirmed that the boundary of the marked lesion area is outside the boundary of the predicted lesion area, or, if the distance between the center position of the predicted lesion area and the center position of the marked lesion area is less than or equal to the set threshold and it is confirmed that the boundary of the predicted lesion area and the boundary of the marked lesion area intersect or do not overlap, it is confirmed that the predicted lesion area and the marked lesion area for the fundus image partially overlap or do not overlap.

[0087] The fourth step is model verification.

[0088] After repeating the training process of the third step according to the training data set, the method further includes: repeating the following verification process according to the verification data set: using the verification data set to verify the fundus macular lesion area detection model to obtain the lesion area to be verified.

[0089] For example, the artificial neural network model to be verified is verified using the verification set, and the fundus data in the verification set is input into the artificial neural network model to be verified, and the lesion area to be verified output by the artificial neural network model to be verified is obtained. The accuracy of the lesion area to be verified is obtained by comparing the lesion area to be verified with the labeled lesion area data in the verification set. When the accuracy of the lesion area to be verified is low, the model parameters need to be adjusted. When the accuracy of the lesion area to be verified is no longer improved, the verification is completed, and the artificial neural network model to be tested is obtained and saved.

[0090] Step 5: Model testing.

[0091] Repeat the following test process for the artificial neural network model to be tested obtained in the third step using the test data set.

[0092] For example, a fundus data in the test set is input into the artificial neural network model to be tested, and the output result of the artificial neural network model to be tested is obtained, which includes an image to be tested of at least one detection area and at least one confidence level, wherein one detection area corresponds to one confidence level. Figure 8 and Fig. 9 As shown, Figure 8 is the annotation result of the test image in the test dataset, Fig. 9 This is a schematic diagram of the results of the image of the tested area, where the results include the MAC area with a confidence level of 0.38 and the lesion area HBBBB with a confidence level of 0.34.

[0093] It is understandable that since the lesion-free data in the test set marks the location of the MAC, the test results will detect the location of the MAC while detecting the lesion area. If the MAC location is detected with a high confidence level, it will be displayed in the test results.

[0094] Step 6: Analyze and post-process the test results.

[0095] The image to be tested including at least one detection area and at least one confidence level in the fifth step is analyzed and redundancy is removed using post-processing technology.

[0096] For example, in the test images obtained for various lesion types, the study found that there were inclusions or intersections between the same lesions or between different types of lesions. Fig.10 As shown, Fig.10 This is a display diagram of the detection results containing multiple lesion type areas. It can be seen that the figure includes multiple lesion type marked areas.

[0097] In order to reasonably remove duplicate test results. As an example of this application, the NMS algorithm is used to remove Fig.10 For lesions of the same category, the NMS algorithm is used to Fig.10 The test results with basically the same position but lower confidence than the set confidence threshold are eliminated. The results after elimination are as follows Fig.11 As shown in , the detection area with higher confidence is retained. Fig.11 For different types of lesions, the NMS algorithm is used to eliminate the results whose position overlap is greater than the set threshold and whose confidence is lower than the set confidence threshold. The results after elimination are as follows: Fig.12 shown.

[0098] In addition, for the detection results of only the macular area, in addition to applying the NMS algorithm, the one with the largest confidence is selected as the final detection result of the MAC area, so that one fundus image contains only one MAC area.

[0099] Step 7: Analyze the processing results and adjust the model parameters.

[0100] The accuracy of the above-mentioned image of the detection area to be tested is analyzed to adjust the value of the parameters of the fundus macular lesion area detection model to be tested. For example, the associated parameters (such as the learning rate) in the model can be fine-tuned according to the accuracy to improve the application effect of the model.

[0101] Step 8: Obtain the target artificial neural network model (as a specific example of a fundus macular lesion area detection model).

[0102] In addition, some embodiments of the present application also provide a method for detecting a macular lesion area on the fundus, which can detect the macular lesion area based on the fundus image to be detected and the macular lesion area to be detected. Figure 2 The fundus macular lesion target detection model obtained by the corresponding method of any embodiment in the invention is used to obtain a fundus macular lesion detection result including at least one detection area and at least one confidence level corresponding to the at least one detection area; the fundus macular lesion detection result is confirmed according to the non-maximum suppression algorithm and the confidence level, wherein the fundus macular lesion detection result includes a fundus macular lesion area and one or more lesion type areas. The above method can be used to quickly and accurately obtain the fundus macular lesion detection result, which is convenient for doctors to view or refer to. Moreover, the above method can be used to efficiently and accurately detect batches of fundus data.

[0103] Please refer to Fig.13 , Fig.13 FIG. 1 is a block diagram showing a device for training a model for detecting a fundus macular lesion region provided in an embodiment of the present application. It should be understood that the device for training a model for detecting a fundus macular lesion region is similar to the above-mentioned Figure 2 Corresponding to the method embodiment, each step involved in the above method embodiment can be executed. The specific functions of the device for training the fundus macular lesion area detection model can be found in the description above. To avoid repetition, the detailed description is appropriately omitted here.

[0104] Fig.13The training fundus macular lesion area detection model includes at least one software function module that can be stored in a memory in the form of software or firmware or solidified in the training fundus macular lesion area detection model device, and the training fundus macular lesion area detection model includes: S1310 data acquisition module, configured to acquire multiple fundus image data, wherein the fundus image data includes: original data and annotated area data of each fundus image; S1320 training data set module, configured to obtain a training data set from the multiple fundus image data; S1330 model training module, configured to repeat the following training process according to the training data set: input the original data of the fundus image in the training data set into the training data set; The method comprises the following steps: inputting a fundus macular lesion area detection model in training to obtain a predicted lesion area; confirming the accuracy of the detection result according to the predicted lesion area, and adjusting the value of the trained parameter according to the accuracy of the detection result, wherein the accuracy is characterized by a decision factor, and if it is confirmed that the predicted lesion area includes the annotated lesion area for the fundus image or it is confirmed that the annotated lesion area for the fundus image includes the predicted lesion area, the value of the decision factor is set to a target value, and the target value is used to characterize that the detection result is accurate; or if it is confirmed that the predicted lesion area and the annotated lesion area for the fundus image partially overlap or have no overlap, the accuracy of the detection result is determined according to the calculated value of the decision factor.

[0105] Some embodiments of the present application also provide a system, the system comprising one or more computers and one or more storage devices storing instructions, when the instructions are executed by the one or more computers, the one or more computers execute Figure 2 The operation of the corresponding method of any embodiment.

[0106] Some embodiments of the present application also provide one or more computer storage media storing instructions, which, when executed by one or more computers, enable the one or more computers to execute Figure 2 The operation of the corresponding method in any one of the embodiments.

[0107] The above description is only an embodiment of the present application and is not intended to limit the scope of protection of the present application. For those skilled in the art, the present application may have various changes and variations. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application should be included in the scope of protection of the present application. It should be noted that similar reference numerals and letters represent similar items in the following drawings, so once an item is defined in one drawing, it does not need to be further defined and explained in the subsequent drawings.

[0108] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

[0109] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

Claims

1. A method for training a fundus macular lesion area detection model, characterized in that: The method comprises: Acquire multiple fundus image data, wherein the fundus image data includes: original data and annotated area data of each fundus image; Obtaining a training data set from the plurality of fundus image data; Repeat the following training process based on the training data set: Inputting the original data of the fundus image in the training data set into the fundus macular lesion area detection model being trained to obtain a predicted lesion area; Confirming the accuracy of the detection result according to the predicted lesion area, and adjusting the value of the trained parameter according to the accuracy of the detection result, wherein the accuracy is characterized by a decision factor, and if it is confirmed that the predicted lesion area includes the annotated lesion area for the fundus image or it is confirmed that the annotated lesion area for the fundus image includes the predicted lesion area, then the value of the decision factor is set as a target value, and the target value is used to characterize that the detection result is accurate; or if it is confirmed that the predicted lesion area and the annotated lesion area for the fundus image partially overlap or have no overlap, then determining the accuracy of the detection result according to the calculated value of the decision factor; The confirming that the predicted lesion area includes the annotated lesion area of ​​the fundus image or confirming that the annotated lesion area of ​​the fundus image includes the predicted lesion area includes: The boundary of the circumscribed polygon corresponding to the predicted lesion area includes the boundary of the circumscribed polygon corresponding to the marked lesion area, or the boundary of the circumscribed polygon corresponding to the marked lesion area includes the boundary of the circumscribed polygon corresponding to the predicted lesion area.

2. The method according to claim 1, characterized in that The predicted lesion region is confirmed to include the annotated lesion region or the annotated lesion region is confirmed to include the predicted lesion region based on the relationship between the position information of the predicted lesion region and the position information of the marked lesion region.

3. The method according to any one of claims 1 to 2, characterized in that: The confirming that the predicted lesion area includes the annotated lesion area of ​​the fundus image or confirming that the annotated lesion area of ​​the fundus image includes the predicted lesion area includes: If the distance between the center position of the predicted lesion area and the center position of the marked lesion area is less than or equal to a set threshold, and it is confirmed that the boundary of the predicted lesion area is outside or overlaps with the boundary of the marked lesion area, then it is confirmed that the predicted lesion area includes the marked lesion area, or if the distance between the center position of the predicted lesion area and the center position of the marked lesion area is less than or equal to a set threshold, and it is confirmed that the boundary of the marked lesion area is outside or overlaps with the boundary of the predicted lesion area, then it is confirmed that the marked lesion area includes the predicted lesion area.

4. The method according to claim 1, characterized in that The boundary of the circumscribed polygon is characterized by the coordinates of the fixed points included in the boundary.

5. The method according to claim 1, characterized in that The confirming that the predicted lesion area and the annotated lesion area of ​​the fundus image partially overlap or do not overlap includes: If the distance between the center position of the predicted lesion area and the center position of the annotated lesion area is greater than a set threshold and it is confirmed that the boundary of the predicted lesion area intersects or does not overlap with the boundary of the annotated lesion area, or the distance between the center position of the predicted lesion area and the center position of the annotated lesion area is greater than a set threshold and it is confirmed that the boundary of the annotated lesion area is outside the boundary of the predicted lesion area or does not overlap, then it is confirmed that the predicted lesion area and the annotated lesion area for the fundus image partially overlap or do not overlap, or, if the distance between the center position of the predicted lesion area and the center position of the annotated lesion area is less than or equal to the set threshold and it is confirmed that the boundary of the predicted lesion area intersects or does not overlap with the boundary of the annotated lesion area, then it is confirmed that the predicted lesion area and the annotated lesion area for the fundus image partially overlap or do not overlap.

6. The method according to claim 1, characterized in that The setting range of the target value is 0.9 to 1.

7. The method according to claim 1, characterized in that Dividing the plurality of fundus image data into the training data set, the verification data set and the test data set; in, After repeating the following training process according to the training data set, the method further includes: Repeat the following validation process based on the validation dataset: Inputting the original data of the fundus image included in the verification data set into the fundus macular lesion area detection model in verification to obtain the lesion area to be verified, wherein the fundus macular lesion area detection model in verification corresponding to the first execution of the verification process is the fundus macular lesion area detection model to be verified obtained through the training process; The accuracy of the detection result is confirmed according to the lesion area to be verified, and the value of the detection model parameter of the fundus macular lesion area to be verified is adjusted according to the accuracy of the detection result.

8. The method according to claim 7, characterized in that After repeating the following verification process according to the verification data set, the method further includes: Repeat the following test process based on the test data set: Inputting the original data of the fundus image in the test data set into the fundus macular lesion area detection model under test to obtain an image to be tested containing at least one detection area, wherein the fundus macular lesion area detection model under test corresponding to the first execution of the test process is the fundus macular lesion area detection model to be tested obtained through the verification process; Confirming the image of the detection area to be tested according to at least one confidence level corresponding to the at least one detection area, wherein one confidence level corresponds to one detection area; The accuracy of the detection result is confirmed according to the image of the detection area to be tested, and the value of the detection model parameter of the fundus macular lesion area to be tested is adjusted according to the accuracy of the detection result.

9. The method according to claim 8, characterized in that Confirming the image of the detection area to be tested according to at least one confidence level corresponding to the at least one detection area includes: The image of the detection area to be tested is confirmed according to the non-maximum suppression algorithm and the confidence level, wherein the image of the detection area to be tested includes the fundus macular area and one or more lesion type areas.

10. A method for detecting a macular lesion area of ​​a fundus, characterized in that: include: According to the fundus image to be detected and the fundus macular lesion target detection model obtained by the method according to any one of claims 1 to 9, obtaining a fundus macular detection result including at least one detection area and at least one confidence level corresponding to the at least one detection area; The fundus macula detection result is confirmed according to the non-maximum suppression algorithm and the confidence level, wherein the fundus macula detection result includes the fundus macula area and one or more lesion type areas.

11. A device for training a fundus macular lesion area detection model, characterized in that: The device is used to perform the method according to claim 1, comprising: A data acquisition module is configured to acquire a plurality of fundus image data, wherein the fundus image data includes: original data and annotated area data of each fundus image; A training data set module is configured to obtain a training data set from the plurality of fundus image data; The model training module is configured to repeat the following training process according to the training data set: input the original data of the fundus image in the training data set into the fundus macular lesion area detection model under training to obtain the predicted lesion area; confirm the accuracy of the detection result according to the predicted lesion area, and adjust the value of the trained parameter according to the accuracy of the detection result, wherein the accuracy is characterized by a decision factor, and if it is confirmed that the predicted lesion area includes the annotated lesion area for the fundus image or it is confirmed that the annotated lesion area of ​​the fundus image includes the predicted lesion area, the value of the decision factor is set to a target value, and the target value is used to characterize that the detection result is accurate; or if it is confirmed that the predicted lesion area and the annotated lesion area for the fundus image partially overlap or have no overlap, the accuracy of the detection result is determined according to the calculated value of the decision factor.

12. A system comprising one or more computers and one or more storage devices storing instructions, wherein when the instructions are executed by the one or more computers, the one or more computers are caused to perform the operations of the corresponding method according to any one of claims 1-10.

13. One or more computer storage media storing instructions, which, when executed by one or more computers, cause the one or more computers to perform the operations of the corresponding method according to any one of claims 1-10.

Citation Information

Patent Citations

  • Three-dimensional target detection method, training method of model of three-dimensional target detection method, related device and equipment

    CN111179247A