Sperm key point data recognition model training method, data recognition method and device

Abstract

The application discloses a sperm key point data recognition model training method, a data recognition method and device. The model training method comprises the following steps: acquiring a plurality of sperm sample images and sample labels of the sperm sample images; iteratively performing the following training process until the sperm key point data recognition model is obtained: inputting the sperm sample images into the sperm key point data recognition model in training, obtaining key point data recognition results output by the sperm key point data recognition model, and training the sperm key point data recognition model in training based on the key point data recognition results and the sample labels, wherein the key point data recognition results comprise position region data of a region where a part key point of each part is located and region pixel data of the region where the part key point is located. The sperm recognition accuracy in the sperm sample without staining is improved.

Description

TECHNICAL FIELD

[0001] Embodiments of the present application relate to the technical field of computer technology, and in particular to a sperm key point data recognition model training method, data recognition method and device. BACKGROUND

[0002] With the influence of environmental pollution, life pressure and other factors, the incidence of male infertility is increasing year by year. Sperm quality analysis is an important examination to understand male fertility, and sperm motility is the main indicator of sperm motility.

[0003] In the late 1980s, computer-aided sperm quality analysis technology (CASA) has developed rapidly. It is found that using computer image analysis technology to automatically measure and evaluate various data of sperm has many advantages, which not only has simple operation, fast analysis speed, high calculation precision, good repeatability, provides accurate reference data for artificial insemination, improves the test level of the test doctor, reduces the workload of the test doctor, but also can overcome the shortcomings of traditional measurement methods, such as time-consuming, poor measurement accuracy, strong human subjectivity, etc.

[0004] In order to reduce the influence of staining on sperm activity in semen, the existing semen analysis process has omitted the staining step of the semen sample, and the semen sample is directly detected by a high-power microscope. In the detection process, the phase contrast microscope is used to obtain the cell image in the semen, and the sperm cells in the semen will be displayed as bright spots under the microscope. However, for high-concentration sperm specimen samples, the bright area is large, and general image processing is difficult to separate the adhered sperm, thereby affecting the accuracy of sperm recognition, and seriously affecting the subsequent sperm quality analysis. SUMMARY

[0005] The present application provides a sperm key point data recognition model training method, data recognition method and device to improve the accuracy of sperm recognition in sperm samples.

[0006] In a first aspect, the embodiments of the present application provide a sperm key point data recognition model training method, which comprises:

[0007] Obtaining multiple frames of multi-sperm sample images, and obtaining sample labels of the multi-sperm sample images; wherein the sample labels include position area labels of regions where part key points of at least one part of sperm in the multi-sperm sample images are located and sperm labels to which the part key points belong;

[0008] Iteratively performing the following training process until the training is completed to obtain a sperm key point data recognition model:

[0009] input the multi-sperm sample image into the sperm key point data recognition model in training, and obtain a key point data recognition result output by the sperm key point data recognition model in training; and train the sperm key point data recognition model in training based on the key point data recognition result and the sample label, wherein the key point data recognition result comprises position region data of a region where a part key point of each part is located and region pixel data of the region where the part key point is located.

[0010] Optionally, a sample label of the multi-sperm sample image is obtained, comprising:

[0011] For a part key point of any part, a key point coordinate of a current part key point in the multi-sperm image is obtained, and a preset position region containing the part key point in the multi-sperm image is determined based on the key point coordinate.

[0012] A position distance between each position coordinate in the preset position region and the key point coordinate is determined respectively.

[0013] A position region heat map of the preset position region where the current part key point is located is generated based on the each position coordinate and the corresponding position distance.

[0014] Optionally, the position region data where the part key point is located comprises position probability data of each position coordinate in the position region being the current part key point.

[0015] Correspondingly, after obtaining the key point data recognition result output by the sperm key point data recognition model in training, the method further comprises:

[0016] A preset key point position threshold is obtained, and a key point position of each part key point in the multi-sperm image is determined based on the position probability data in the position region data and the key point position threshold.

[0017] Optionally, the training of the sperm key point data recognition model in training based on the key point data recognition result and the sample label comprises:

[0018] A position region loss function generated based on the position region data where the part key point is located and a position region label of the region where the part key point is located, and a pixel loss function generated based on pixel data of the part key point position and a sperm label to which the part key point belongs are obtained, and the sperm key point data recognition model in training is trained based on the position region loss function and the pixel loss function.

[0019] Optionally, the pixel loss function is determined based on a pull-in loss function between position key points of the same sperm and a push-away loss function between position key points of different sperms.

[0020] Optionally, the pull-in loss function is determined based on pixel data of position key point positions of each position in the current sperm and average data of pixel data of the each position.

[0021] The push-away loss function is determined based on average data of each position in the current sperm and average data of each position in other sperms in the multi-sperm image.

[0022] Optionally, the sperm key point data recognition model comprises a pixel processing module, a feature extraction module, a feature splicing module and a recognition module.

[0023] The pixel processing module is configured to perform image scaling processing on the multi-sperm image.

[0024] The feature extraction module comprises at least one level of feature extraction sub-modules, and a feature extraction sub-module at a first level is connected with the pixel processing module.

[0025] The feature splicing module is connected with each level of feature extraction sub-modules, configured to perform feature splicing on output data of each level of feature extraction sub-modules after up-sampling of a preset step, to obtain spliced features output by the feature splicing module.

[0026] The recognition module is connected with the feature splicing module, configured to determine the key point data recognition result of the current multi-sperm image based on the spliced features obtained through feature splicing processing.

[0027] In a second aspect, an embodiment of the present application further provides a sperm key point data recognition method, which comprises:

[0028] obtaining an initial multi-sperm image, and performing image preprocessing on the initial multi-sperm image to obtain a multi-sperm image;

[0029] Based on a pre-trained sperm key point data recognition model, the key point data recognition result of the polysperm image is determined; wherein, the key point data recognition result includes the position area data of the area where the part key points of each part are located and the regional pixel data of the area where the part key points are located; the sperm key point data recognition model is trained based on the sperm key point data recognition model training method described in any of the above embodiments.

[0030] In a third aspect, an embodiment of the present invention further provides a sperm key point data recognition model training device, the device comprising:

[0031] A sample image and label acquisition module, configured to acquire multiple frames of polyspermia sample images and sample labels for the polyspermia sample images; wherein the sample labels include a location region label of a region where a key point of at least one part of the sperm in the polyspermia sample image is located and a sperm label to which the key point belongs;

[0032] The model training module is used to iteratively execute the following training process until the training is completed to obtain the sperm key point data recognition model:

[0033] The polysperm sample image is input into a sperm key point data recognition model under training, and a key point data recognition result output by the sperm key point data recognition model is obtained. The sperm key point data recognition model under training is trained based on the key point data recognition result and the sample label, wherein the key point data recognition result includes positional area data of the area where the key points of each part are located and regional pixel data of the area where the key points of each part are located.

[0034] In a fourth aspect, an embodiment of the present invention further provides a sperm key point data identification device, the device comprising:

[0035] A polyspermia image acquisition module is used to acquire an initial polyspermia image and perform image preprocessing on the initial polyspermia image to obtain a polyspermia image;

[0036] A key point data recognition result determination module is used to determine the key point data recognition results of the polysperm image based on a pre-trained sperm key point data recognition model; wherein the key point data recognition results include position area data of the area where the key points of each part are located and pixel data of the area where the key points of the part are located; the sperm key point data recognition model is trained based on the sperm key point data recognition model training method provided in any embodiment of the present invention.

[0037] In a fifth aspect, an embodiment of the present invention further provides an electronic device, comprising:

[0038] one or more processors;

[0039] a storage device for storing one or more programs,

[0040] When the one or more programs are executed by the one or more processors, the one or more processors implement the sperm key point data recognition model training method or the sperm key point data recognition method provided in any embodiment of the present invention.

[0041] In a sixth aspect, an embodiment of the present invention further provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the sperm key point data recognition model training method or the sperm key point data recognition method provided in any embodiment of the present invention.

[0042] The model training method provided in this embodiment specifically includes obtaining multiple frames of polyspermia sample images and obtaining sample labels for the polyspermia sample images; wherein the sample labels include the location region labels of the region where the key points of at least one part of the sperm in the polyspermia sample images are located, and the sperm labels to which the key points belong; iteratively performing the following training process until the training is completed to obtain a sperm key point data recognition model: inputting the polyspermia sample images into the sperm key point data recognition model under training, and obtaining the key point data recognition results output by the sperm key point data recognition model; training the sperm key point data recognition model under training based on the key point data recognition results and the sample labels, wherein the key point data recognition results include the location region data of the region where the key points of each part are located, and the regional pixel data of the region where the key points are located. Through the above-mentioned model training method, a sperm recognition model with higher recognition accuracy is obtained, thereby improving the accuracy of sperm recognition. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] To more clearly illustrate the technical solutions of the exemplary embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings introduced here only illustrate some of the embodiments to be described by the present invention, and are not exhaustive. A person skilled in the art can derive other drawings based on these drawings without inventive effort.

[0044] Figure 1 1 is a flow chart of a sperm key point data recognition model training method provided in Example 1 of the present invention;

[0045] Figure 2 1 is a schematic structural diagram of a sperm key point data recognition model provided in Example 1 of the present invention;

[0046] Figure 3 1 is a flow chart of a sperm key point data identification method provided in Example 2 of the present invention;

[0047] Figure 4 is a structural schematic diagram of a sperm key point data recognition model training device provided by Embodiment Three of the present application;

[0048] Figure 5 is a structural schematic diagram of a sperm key point data recognition device provided by Embodiment Three of the present application;

[0049] Figure 6 is a structural schematic diagram of an electronic device provided by Embodiment Four of the present application. DETAILED DESCRIPTION

[0050] The present application will be further described below in conjunction with the accompanying drawings and embodiments. It can be understood that the specific embodiments described herein are only used to explain the present application, but not to limit the present application. In addition, it should be noted that, for the convenience of description, only the parts related to the present application are shown in the drawings, but not all the structures.

[0051] Embodiment One

[0052] Figure 1 is a flowchart of a sperm key point data recognition model training method provided by Embodiment One of the present application. The present embodiment can be applicable to the case of recognizing the key points of sperm in a sperm image. The method can be executed by a sperm key point data recognition model training device, which can be realized by software and / or hardware.

[0053] Before introducing the technical solutions of the embodiments of the present application, the application scenarios of the technical solutions of the embodiments are exemplarily introduced. Of course, the following application scenarios are only optional application scenarios, and the present embodiment can also be implemented in other application scenarios, and the present embodiment does not limit the application scenarios of the implemented technical methods. Specifically, the application scenarios include: with the influence of environmental pollution, life pressure and other factors, the incidence of male infertility increases year by year. Sperm quality analysis is an important examination to understand male fertility, and sperm motility is the main indicator of sperm movement ability. In order to reduce the influence of staining on sperm activity in semen, the existing semen analysis process has omitted the staining step of the semen sample, and the semen sample is directly detected by a high-power microscope. In the detection process, a phase contrast microscope is used to obtain the cell image in the semen. The sperm cells in the semen will be displayed as bright spots under the microscope. However, for high-concentration sperm specimen samples, the bright area is very large, and general image processing is difficult to separate the adhered sperm, thereby affecting the accuracy of sperm recognition, and seriously affecting the subsequent sperm quality analysis.

[0054] In view of the above technical problems, the technical scheme in the embodiment achieves the goal of identifying sperm by identifying key points in sperm through a sperm key point data identification model, thereby avoiding errors in sperm motility identification caused by sperm adhesion, and thus the problem of low sperm identification accuracy.

[0055] Before identifying sperm by using the sperm key point data identification model, the sperm key point data identification model is trained, specifically, the training method includes obtaining multiple frames of multi-sperm sample images and obtaining sample labels of the multi-sperm sample images; wherein the sample labels include position region labels of regions where key points of at least one part of sperm in the multi-sperm sample images are located and sperm labels to which the key points belong; the following training process is iteratively executed until the sperm key point data identification model is obtained through training: inputting the multi-sperm sample images into the sperm key point data identification model in training, and obtaining key point data identification results output by the sperm key point data identification model; training the sperm key point data identification model in training based on the key point data identification results and the sample labels, wherein the key point data identification results include position region data of regions where key points of each part are located and region pixel data of the regions where the key points are located. Through the above model training method, a sperm identification model with higher identification accuracy is obtained, thereby improving the accuracy of sperm identification.

[0056] As shown in the method, the method specifically includes the following steps: Figure 1

[0057] S110, obtaining multiple frames of multi-sperm sample images and obtaining sample labels of the multi-sperm sample images.

[0058] In the embodiment, the multi-sperm sample images can be sample images containing multiple sperm, and the sample images are used to train the sperm key point data identification model.

[0059] Specifically, the multi-sperm sample images can be images obtained by collecting fresh semen samples. The multiple frames of multi-sperm sample images can be obtained by continuously photographing a preset region of a semen smear under a preset magnification objective of a microscope equipped with a camera device, thereby obtaining multiple frames of multi-sperm sample images, or by recording a video using the above microscope, and then obtaining multiple frames of multi-sperm sample images from the sperm sample video.

[0060] Specifically, the multiple frames of multi-sperm sample images can be real-time collected images, and the method of obtaining the multiple frames of multi-sperm sample images can include: obtaining the collected images from the image collection device in real time. The multiple frames of multi-sperm sample images can also be pre-collected and stored in a local database or a server database, and the method of obtaining the multiple frames of multi-sperm sample images can further include: obtaining the images from the local database or the server database.​

[0061] In this embodiment, the sample image of the multiple sperm can be further labeled after being acquired, so as to acquire the sample label of the sample image of the multiple sperm. Alternatively, the sample image of the multiple sperm can be labeled before being acquired, and the sample image of the multiple sperm with the sample label is directly acquired in this embodiment. Similarly, the sample label can be a label based on manual labeling, or a label based on neural network or other computer labeling. The order of acquiring the sample image and the sample label, and the manner of acquiring the sample label are not limited in this embodiment.

[0062] In this embodiment, the sample label includes a position region label of a region where a part key point of at least one part of a sperm in the sample image of the multiple sperm is located, and a sperm label to which the part key point belongs. Since each frame of the sample image of the multiple sperm in this embodiment contains multiple sperms, each sperm is labeled with multiple parts, for example, a sperm head and a sperm tail, and for example, the sperm head further includes a sperm head vertex and a sperm head tail point. The above-mentioned parts are only exemplary optional parts in this embodiment, and other parts of the sperm can also be included, which are not enumerated one by one in this embodiment. In order to more clearly display the position region label of the region where the part key point of at least one part in the sample image of the multiple sperm is located in this embodiment, a position region heat map of each part key point is generated based on the position region of the region where each part key point is located in the multiple sperm image.

[0063] Optionally, for any part key point, the key point coordinate of the current part key point in the multiple sperm image is acquired, and a preset position region in the multiple sperm image containing the part key point is determined based on the key point coordinate; the position distance between each position coordinate in the preset position region and the key point coordinate is determined respectively; and a position region heat map of the preset position region where the current part key point is located is generated based on each position coordinate and the corresponding position distance.

[0064] Specifically, one sperm part in any frame of the sample image of the multiple sperm is taken as an example to introduce the generation of the position region heat map of the part key point of the part. For example, in the current frame of the sample image of the multiple sperm, the head vertex of the sperm head is acquired as the key point labeling result of the part key point of the first part, the key point coordinate of the part key point of the first part in the multiple sperm image is determined based on the key point labeling result of the part key point, and the preset position region in the multiple sperm image containing the part key point is determined based on the key point coordinate. Further, the position distance between each position coordinate in the preset position region and the key point coordinate of the first part key point is determined respectively; and the position region heat map of the preset position region where the current part key point is located is generated based on each position coordinate, the corresponding position distance, and the preset heat map generation expression.

[0065] Specifically, a location area heat map of a preset location area where the key point of the current part is located can be generated according to the following expression; illustratively, the expression includes:

[0066]

[0067] in, The key point heat map representing the preset location area of ​​the generated part key points, Represents the position data of the key point in the sample image, Indicates The position data of any point in the sample area with a size of N×N in the center; Indicates the bias parameter. Optional, N and The value of can be set according to actual needs, and this embodiment does not limit the numerical setting; for example, in this embodiment, based on the previous experimental data results, N is set to 15, and the parameter Set to 4.

[0068] It should be noted that the aforementioned key points for the first part are merely illustrative examples in this embodiment. Key points corresponding to other parts of the sperm image may also be determined as key points based on actual circumstances, such as using the head and tail points as key points for the second part. This embodiment does not impose any restrictions on the location, position, or number of key points.

[0069] Furthermore, based on the above method, the regional position heat map of the area where the key points of each part in each frame of the polysperm image are located is determined, and the regional position heat map is further used as a sample label in the training process of the sperm key point data recognition model to train the model to obtain a more accurate sperm key point data recognition model.

[0070] S120, iteratively executing the model training process until the training is completed to obtain a sperm key point data recognition model.

[0071] In an embodiment of the present invention, the training process of the model specifically includes: inputting the polysperm sample image into the sperm key point data recognition model under training, and obtaining the key point data recognition results output by the sperm key point data recognition model, and training the sperm key point data recognition model under training based on the key point data recognition results and sample labels, wherein the key point data recognition results include the position area data of the area where the part key points of each part are located and the area pixel data of the area where the part key points are located.

[0072] Specifically, based on the sample label of the multi-sperm sample image and the key point data recognition result of each iteration of the model, a model loss function in each iteration process is generated, and the parameters in the sperm key point data recognition model in training are adjusted multiple times based on the model loss function until the model training of the sperm key point data recognition model is completed.

[0073] The training method of the sperm key point data recognition model provided in this embodiment is introduced below by taking the model training of any round as an example.

[0074] Specifically, after obtaining the multi-frame multi-sperm sample image and the sample label of the multi-sperm sample image, the pixel value of each pixel point in the multi-sperm sample image is divided by 255 and normalized to obtain the preprocessed multi-sperm sample image, so that the recognition result of the sperm key point data recognition model trained based on the sample image is more accurate.

[0075] Further, the preprocessed multi-sperm sample image is input into the sperm key point data recognition model in training, and the sample key point data recognition result output by the sperm key point data recognition model is obtained. The key point data recognition result includes position region data of the region where the part key point is located and region pixel data of the region where the part key point is located.

[0076] In this embodiment, the sperm key point data recognition model includes a pixel processing module, a feature extraction module, a feature splicing module, and a recognition module. The pixel processing module is used for image scaling processing of the multi-sperm image. The feature extraction module includes at least one level of feature extraction submodule, and the feature extraction submodule of the first level is connected with the pixel processing module. Any level feature extraction submodule is used to take the output data of the pixel processing module or at least one feature extraction submodule of the previous level as the input data of the current level feature extraction submodule after down-sampling processing with a preset stride, and to perform feature extraction on the input data and output. The output data of the current level feature extraction submodule is used as the input data of the next level feature extraction submodule and / or the output data of the current level feature extraction submodule after down-sampling with a preset stride. The feature splicing module is connected with each level feature extraction submodule, and is used to perform feature splicing after up-sampling of the output data of each level feature extraction submodule with a preset stride, to obtain spliced features output by the feature splicing module. The recognition module is connected with the feature splicing module, and is used to determine the key point data recognition result of the current frame multi-sperm image based on the spliced features obtained by the feature splicing processing.

[0077] Specifically as Figure 2As shown, the polyspermic image is input into the model, and first passes through a basic module, which is a pixel processing module. Specifically, the pixel processing module can be composed of two convolution layers with a stride of 2 and a kernel of 3*3. The convolution layer is defined as the combination of convolution operation, BN operation and ReLU activation operation. Specifically, after the polyspermic sample image passes through the pixel processing module, a pixel processing sample image with a resolution of 1 / 4 of the original polyspermic sample image is obtained.

[0078] Specifically, the pixel processing sample image is input into the feature extraction module. The feature extraction module in this embodiment is composed of four levels of feature extraction sub-modules, and the corresponding resolutions of each feature extraction sub-module are 1 / 4, 1 / 8, 1 / 16 and 1 / 32 of the resolution of the original polyspermic image.

[0079] In this embodiment, the pixel processing sample image is input into the first level of feature extraction sub-module in the feature extraction module. The output result of the first convolution layer in the first level is used as the input image of the second convolution layer, and after down-sampling with a stride of 2, it is used as the input image of the first convolution layer in the second level of feature extraction sub-module. Further, the feature image output by the last convolution layer in the first level is used as the output result of the first level of feature extraction sub-module.

[0080] Specifically, the output result of the second convolution layer in the second level of feature extraction sub-module is used as the input image of the third convolution layer, and after down-sampling with a stride of 2, it is down-sampled with a stride of 4 together with the feature image output by the second convolution layer in the first level, and is used as the input image of the first convolution layer in the third level of feature extraction sub-module. Further, the feature image output by the last convolution layer in the second level is used as the output result of the second level of feature extraction sub-module.

[0081] Specifically, the output result of the second convolution layer in the third level of feature extraction sub-module is used as the input image of the third convolution layer, and after down-sampling with a stride of 2, it is down-sampled with a stride of 4 together with the feature image output by the third convolution layer in the second level and the feature sample image output by the third convolution layer in the first level, and is used as the input image of the first convolution layer in the fourth level. Further, the feature image output by the last convolution layer in the third level is used as the output result of the third level of feature extraction sub-module.

[0082] Further, the feature image output by the last convolution layer in the fourth level is used as the output result of the fourth level of feature extraction sub-module.

[0083] The feature images of the output of the last convolution layer in the four-level feature extraction sub-module are up-sampled to the corresponding resolution and input to the stitching model, so that the up-sampled feature images have the same resolution as the pixel processing image input to the feature extraction module. Specifically, the output of the second level is up-sampled by a stride of 2, the output of the third level is up-sampled by a stride of 4, and the output of the fourth level is up-sampled by a stride of 8. The feature images after up-sampling are input to the stitching model for image stitching; further, each feature sample image is subjected to a convolution layer operation to obtain a stitching image output by the stitching model.

[0084] Specifically, the stitching image is input to the recognition model and subjected to a convolution operation with a channel number of 4 to obtain a key point data recognition result output by the recognition model.

[0085] It should be noted that the number of parts corresponding to the part key points to be recognized in the present embodiment is 2, so a convolution operation with a channel number of 4 is required in the recognition model. If the number of parts corresponding to the part key points to be recognized is other values, a convolution operation with a channel number twice the number of parts is required. The channel number is not limited in the present embodiment.

[0086] The resolution of the sample result image in the sample key point data recognition result in the present embodiment is (H / 4, W / 4, 4). The first two channels are used to obtain the region position data of the region where each part key point is located. This position information is used to determine the position information of each part key point. The last two channels are used to obtain the pixel data of the part key point, which is used to calculate the correlation information between each part key point.

[0087] It should be noted that the technical solution of the present embodiment reduces the resolution of the sample result image obtained based on the sperm key point data recognition model by performing different degrees of down-sampling on the multi-sperm sample image. In order to obtain an accurate loss function based on the sample result image, it is necessary to ensure the consistency of the resolution of the sample label and the sample result image. After obtaining the position region data of each part key point in the model output, the technical solution of the present embodiment also generates a position region image corresponding to the position region data in the key point recognition result, and performs image scaling on the position region image to make the scaled position region image have the same resolution as the input image of the sperm key point data recognition model.

[0088] In the embodiment of the present application, the position region data of the part key point includes position probability data of each position coordinate in the position region that is the position of the part key point; accordingly, after obtaining the key point data recognition result output by the sperm key point data recognition model, the method further includes: obtaining a preset key point position threshold, and determining the key point position of each part key point in the multi-sperm image based on the position probability data in the position region data and the key point position threshold.

[0089] Specifically, taking the determination of the key point position of the part key point of the first part in the multi-sperm image as an example, the position region data of the region where the part key point of the first part is located is obtained, that is, the position region data of the part key point of the first channel in the key point data recognition result is obtained. Specifically, the position region data of the part key point of the first part includes: the position probability data of each position coordinate in the position region that is the position of the part key point of the first part. The position coordinate corresponding to the position probability data greater than the threshold t in the position probability data of the part key point of the first part in the position region is determined as the part key point coordinate of the first part. The position coordinate corresponding to the position probability data greater than the threshold t in the position probability data of the part key point of the second part in the position region where the part key point of the second part is located can also be determined as the part key point coordinate of the second part. It should be noted that, since the number of sperms in a frame of multi-sperm image is multiple, the number of key point coordinates of the same part in the multi-sperm image is also multiple. In this embodiment, the value of the threshold t is not limited, and can be set according to the size of the position probability data of the part key point; for example, according to the historical experimental data result, t is set to 0.01 in this embodiment.

[0090] Further, after determining the key point position of the part key point, the pixel data of the part key point position is determined based on the region position data of the region where the part key point is located and the pixel data of the part key point. Further, the model is trained based on the key point data recognition result, the pixel data of the part key point position in the recognition result, and the sample label generation model loss function.

[0091] Specifically, the position region loss function generated based on the position region data of the region where the part key point is located and the position region label of the region where the part key point is located, and the pixel loss function generated based on the pixel data of the part key point position and the label of the sperm to which the part key point belongs are obtained, and the sperm key point data recognition model in training is trained based on the position region loss function and the pixel loss function.

[0092] In the embodiment, the pixel loss function is determined based on a pull-in loss function between the position key points of the same sperm and a push-away loss function between the position key points of different sperms. Specifically, the pull-in loss function is determined based on the key point pixel data of the position key point position of each part in the current sperm and the average data of the pixel data of each part in the current sperm; and the push-away loss function is determined based on the average data of each part in the current sperm and the average data of each part in other sperms in the multi-sperm image.

[0093] Specifically, the position region data of the region where the position key point of each first part in the first channel in the key point data recognition result is located and the position region data of the region where the position key point of each second part in the second channel is located are obtained, as well as the position region heat map of the region where the position key point of the first part is located and the position region heat map of the region where the position key point of the second part is located, and the position region loss function is determined based on the following expression; for example, the position region loss function is as follows:

[0094]

[0095] wherein, represents the position region data of the position key point in the i-th channel in the key point data recognition result, represents the position region heat map of the position key point in the i-th part.

[0096] Specifically, the pixel data of the position key point position of each first part in the third channel in the key point data recognition result and the pixel data of the position key point position of each second part in the fourth channel are obtained, and the pull-in loss function is determined based on the following expression; for example, the pull-in loss function is as follows:

[0097]

[0098] wherein, the definition is and are the results of the third channel and the fourth channel in the key point data recognition result, represents the position key point position of the k-th part of the n-th sperm, then represents the pixel data of the position key point position of the first part and the pixel data of the position key point position of the second part in the third channel and the fourth channel in the key point data recognition result, respectively, an average of pixel data representing the position of the part key point of the first part and the position of the part key point of the second part of the nth sperm, and the data is taken as the reference prediction of the sperm. Further, based on the pixel data of the position of each part key point of the nth sperm and the reference prediction of the sperm, a square loss is determined as the zoom-in loss function of the sperm, and the effect of determining the zoom-in loss function based on the above method is that the pixel data of the part key points of all parts of the sperm is pulled to the average, so that the pixel data of the position of the part key point of all parts of the sperm is similar.

[0099] Specifically, the zoom-out loss function is determined based on the following expression; and exemplarily, the zoom-out loss function is as follows:

[0100]

[0101] wherein, is a transpose matrix of , so that the interval between the reference predictions of different sperms is increased as much as possible, so that different sperms are pushed away from each other.

[0102] In the embodiment, and act simultaneously, so that the pixel data of the position of the part key point of different parts in the same sperm is similar, and the difference between the pixel data of the position of the part key point in different sperms is large, so that the association between the part key points of each part of the same sperm is more accurate.

[0103] Further, the loss function of the model can be expressed as: The value of can be set in advance, and the embodiment does not limit the value of ; and exemplarily, according to historical experimental results, the values of are respectively set to 1, 0.01 and 0.01.

[0104] Further, based on the above loss function, the Adam optimizer is used to train the sperm key point data recognition model in training. The process is repeated until the loss function reaches a preset threshold or the number of iterations reaches a preset threshold, and the training of the sperm key point data recognition model is stopped, and the sperm key point data recognition model is obtained.

[0105] The model training method provided in the embodiment specifically comprises: acquiring a plurality of sperm sample images, and acquiring sample labels of the plurality of sperm sample images; wherein the sample labels comprise position region labels of regions of at least one part of sperm in the plurality of sperm sample images and sperm labels to which the part key points belong; and the following training process is iteratively executed until a sperm key point data recognition model is obtained: inputting the plurality of sperm sample images into the sperm key point data recognition model in training, and obtaining key point data recognition results output by the sperm key point data recognition model; and training the sperm key point data recognition model in training based on the key point data recognition results and the sample labels, wherein the key point data recognition results comprise position region data of the regions of the part key points and region pixel data of the regions of the part key points. Through the above model training method, a sperm recognition model with higher recognition accuracy is obtained, thereby improving the accuracy of sperm recognition.

[0106] Embodiment Two

[0107] Figure 3 A flowchart of a sperm key point data recognition method provided for Embodiment Two of the present application, the embodiment can be applicable to the case of recognizing key points of sperm in a sperm image. The method can be executed by a sperm key point data recognition device, which can be realized by software and / or hardware. As shown in the figure, the method specifically comprises the following steps: Figure 3

[0108] S210, acquiring an initial multi-sperm image, and performing image preprocessing on the initial multi-sperm image to obtain a multi-sperm image.

[0109] In the embodiment, the multi-sperm image can be an image containing a plurality of sperm. The initial multi-sperm image can be an image obtained by collecting a fresh semen sample, and the fresh semen sample can be an undyed sample.

[0110] Optionally, the initial multi-sperm image can be a real-time collected image, and the method of acquiring the initial multi-sperm image can comprise: acquiring the collected image from the image collection device in real time. The multi-sperm image can also be an image pre-collected and stored in a local database or a server database, and the method of acquiring the initial multi-sperm image can further comprise: acquiring the image from the local database or the server database.

[0111] Specifically, the method of performing image preprocessing on the initial multi-sperm image can comprise: dividing the pixel value of each pixel point in the multi-sperm image by 255 and performing normalization to obtain the preprocessed multi-sperm image, so that the sperm key point data recognition result obtained based on the multi-sperm image for image recognition is more accurate.

[0112] ​S220, determining the key point data recognition result of the multi-sperm image based on the pre-trained sperm key point data recognition model.

[0113] In the embodiment, the method for determining the key point data recognition result of the multi-sperm image can include: for each frame of the multi-sperm image in the continuous frames of multi-sperm images, performing image preprocessing on the current frame of multi-sperm image, and inputting the preprocessed multi-sperm image into the pre-trained sperm key point data recognition model to obtain the key point data recognition result of the current frame of multi-sperm image output by the sperm key point data recognition model.

[0114] In the embodiment, the key point data recognition result includes position area data of a region where a key point of at least one part is located and region pixel data of the region where the key point of the part is located; and the sperm key point data recognition model is trained based on the sperm key point data recognition model training method in any of the above embodiments.

[0115] Based on the trained sperm key point data recognition model provided in the above embodiments, further, in the embodiment, the sperm key point data recognition result is obtained based on the sperm key point data recognition model, and the position area data of at least one part key point in the sperm image and the pixel data of each part key point are determined based on the key point data recognition result, which improves the accuracy and reliability of sperm recognition.

[0116] The following are embodiments of the sperm key point data recognition model training device and sperm key point data recognition device provided by the embodiments of the present application. The device belongs to the same inventive concept as the sperm key point data recognition model training method and sperm key point data recognition method of the above embodiments. Details not described in the embodiments of the sperm key point data recognition model training device and sperm key point data recognition device can be referred to the embodiments of the sperm key point data recognition model training method and sperm key point data recognition method.

[0117] Embodiment Three

[0118] Figure 4 The structure diagram of the sperm key point data recognition model training device provided for the third embodiment of the present application. The present embodiment can be applied to the case of recognizing the key points of sperm in sperm images. Referring to Figure 4 The specific structure of the sperm key point data recognition model training device includes: a sample image and label acquisition module 310 and a model training module 320; wherein,

[0119] The sample image and label obtaining module 310 is configured to obtain a plurality of multi-sperm sample images and sample labels of the multi-sperm sample images, wherein the sample labels comprise position region labels of regions of position key points of at least one part of sperm in the multi-sperm sample images and sperm labels to which the position key points belong.

[0120] The model training module 320 is configured to iteratively perform the following training process until a sperm key point data recognition model is obtained through training:

[0121] The multi-sperm sample images are input into the sperm key point data recognition model under training, and key point data recognition results output by the sperm key point data recognition model are obtained, and the sperm key point data recognition model under training is trained based on the key point data recognition results and the sample labels, wherein the key point data recognition results comprise position region data of the regions of the position key points of the parts and region pixel data of the regions of the position key points.

[0122] The model training method provided in the embodiment specifically comprises obtaining a plurality of multi-sperm sample images and sample labels of the multi-sperm sample images, wherein the sample labels comprise position region labels of regions of position key points of at least one part of sperm in the multi-sperm sample images and sperm labels to which the position key points belong, and iteratively performing the following training process until a sperm key point data recognition model is obtained through training: the multi-sperm sample images are input into the sperm key point data recognition model under training, and sample key point data recognition results output by the sperm key point data recognition model are obtained, and the sperm key point data recognition model under training is trained based on the sample key point data recognition results and the sample labels, wherein the sample key point data recognition results comprise position region data of the regions of the position key points of the parts and region pixel data of the regions of the position key points. Through the above model training method, a sperm recognition model with higher recognition accuracy is obtained, thereby improving the accuracy of sperm recognition.

[0123] On the basis of the above embodiments, the sample image and label obtaining module 310 comprises:

[0124] The preset position region obtaining unit is configured to, for the position key point of any part, obtain a key point coordinate of the current position key point in the multi-sperm image, and determine a preset position region in the multi-sperm image containing the position key point based on the key point coordinate;

[0125] The position distance determining unit is configured to respectively determine position distances between each position coordinate in the preset position region and the key point coordinate;

[0126] Generate a position area heat map of a preset position area where the current part key point is located based on the position coordinates and the corresponding position distance.

[0127] On the basis of the above embodiments, the position area data of the part key point includes the position probability data of each position coordinate in the position area.

[0128] Correspondingly, the device also includes:

[0129] The key point position determination unit is configured to obtain a preset key point position threshold, and determine the key point position of each part key point in the multi-sperm image based on the position probability data in the position area data and the key point position threshold.

[0130] On the basis of the above embodiments, the model training module 320 includes:

[0131] The model training submodule is configured to obtain a position area loss function generated based on the position area data of the region where the part key point is located and the position area label of the region where the part key point is located, and a pixel loss function generated based on the region pixel data of the region where the part key point is located and the sperm label to which the part key point belongs, and train the sperm key point data recognition model in the training based on the position area loss function and the pixel loss function.

[0132] On the basis of the above embodiments, the pixel loss function is determined based on a pull-in loss function between part key points of the same sperm and a push-away loss function between part key points of different sperms.

[0133] On the basis of the above embodiments,

[0134] The pull-in loss function is determined based on the pixel data of the position of each part key point in the current sperm and the average data of the pixel data of each part.

[0135] The push-away loss function is determined based on the average data of each part in the current sperm and the average data of each part in other sperms in the multi-sperm image.

[0136] On the basis of the above embodiments, the sperm key point data recognition model includes a pixel processing module, a feature extraction module, a feature splicing module, and a recognition module.

[0137] The pixel processing module is configured to perform image scaling processing on the multi-sperm image.

[0138] The feature extraction module comprises at least one level of feature extraction sub-module, and the feature extraction sub-module of the first level is connected with the pixel processing module; any level of feature extraction sub-module is used for taking the output data of the pixel processing module or at least one feature extraction sub-module of the previous level as the input data of the current level feature extraction sub-module after performing down-sampling processing with a preset step, performing feature extraction on the input data and outputting; wherein the output data of the feature extraction sub-module of the current level is used as the input data of the feature extraction sub-module of the next level after performing down-sampling with a preset step and / or the output data of the feature extraction sub-module of the current level;

[0139] The feature splicing module is connected with each level of feature extraction sub-module, and is used for performing feature splicing after performing up-sampling with a preset step on the output data of each level of feature extraction sub-module, to obtain spliced features output by the feature splicing module;

[0140] The recognition module is connected with the feature splicing module, and is used for determining the key point data recognition result of the current frame of multi-sperm image based on the spliced features obtained through feature splicing processing.

[0141] The sperm key point data recognition model training device provided in the embodiments of the present application can execute the sperm key point data recognition model training method provided in any embodiment of the present application, and has the corresponding function modules and beneficial effects of the execution method.

[0142] It is worth noting that in the embodiments of the sperm key point data recognition model training device described above, each unit and module included is only divided according to the function logic, but is not limited to the above division, as long as the corresponding functions can be realized; in addition, the specific names of each functional unit are only for the convenience of mutual differentiation, and do not limit the protection scope of the present application.

[0143] Embodiment four

[0144] Figure 5 The structure diagram of the sperm key point data recognition device provided in the fourth embodiment of the present application can be applied to the case of identifying the key points of sperm in the sperm image. Referring to Figure 5 The specific structure of the sperm key point data recognition device comprises a multi-sperm image acquisition module 410 and a key point data recognition result determination module 420; wherein,

[0145] The multi-sperm image acquisition module 410 is used for acquiring an initial multi-sperm image, and obtaining a multi-sperm image by performing image preprocessing on the initial multi-sperm image;

[0146] The key point data recognition result determination module 420 is configured to determine the key point data recognition result of the multi-sperm image based on the pre-trained sperm key point data recognition model, wherein the key point data recognition result comprises position region data of a region where each part key point is located and region pixel data of the region where the part key point is located; and the sperm key point data recognition model is trained based on the sperm key point data recognition model training method in any of the above embodiments.

[0147] Based on the trained sperm key point data recognition model provided in the above embodiments, further, in the embodiment, the key point data recognition result of the sperm image is obtained based on the sperm key point data recognition model, and the position region data of at least one part key point in the sperm image and the pixel data of each part key point are determined based on the key point data recognition result, thereby improving the accuracy and reliability of sperm recognition.

[0148] The sperm key point data recognition apparatus provided in the embodiment of the present application can execute the sperm key point data recognition method provided in any of the embodiments of the present application, and has the corresponding function modules and beneficial effects of the execution method.

[0149] It should be noted that, in the embodiments of the sperm key point data recognition apparatus, each unit and module included is only divided according to the function logic, but is not limited to the above division, as long as the corresponding function can be realized; in addition, the specific names of each functional unit are only for the convenience of mutual differentiation, and do not limit the protection scope of the present application.

[0150] Embodiment five

[0151] Figure 6 A structural schematic diagram of an electronic device provided for the embodiment five of the present application. Figure 6 A block diagram of an exemplary electronic device 12 suitable for use in implementing embodiments of the present application is shown. Figure 6 The electronic device 12 shown is merely one example and should not be construed as limiting the scope of the embodiments of the present application.

[0152] As shown in Figure 6 The electronic device 12 is shown in the form of a general computing electronic device. Components of the electronic device 12 can include, but are not limited to, one or more processors or processing units 16, system memory 28, and a bus 18 that connects different system components, including the system memory 28 and the processing unit 16.

[0153] Bus 18 represents one or more of several types of bus structures, including a memory bus or memory controller, a peripheral bus, a graphics acceleration bus, a processor or local bus using any of a variety of bus architectures. By way of example, these architectures include Industry Standard Architecture (ISA) bus, Micro Channel Architecture (MCA) bus, Enhanced ISA (EISA) bus, Video Electronics Standards Association (VESA) local bus, and Peripheral Component Interconnect (PCI) bus.

[0154] Electronic device 12 typically includes a variety of computer system readable media. These media can be any available media that is accessible by electronic device 12 and includes both volatile and non-volatile media, removable and non-removable media.

[0155] System memory 28 can include computer system readable media in the form of volatile memory, such as random access memory (RAM) 30 and / or cache memory 32. Electronic device 12 can further include other removable / non-removable, volatile / non-volatile computer system storage media. By way of example only, storage system 34 can be provided for reading from and writing to a non-removable, non-volatile magnetic media (e.g., a "hard drive"). Figure 6 not shown in FIG. 1, can provide volatile memory for storage of information and / or instructions to be executed by processing unit 20. Although exemplary embodiments of the present application are described in the foregoing context of a fully functioning computer system, those of ordinary skill in the art will appreciate that the acts and parameter values described herein can be carried out in a computer system configured to function as a specific device, such as a virtual machine, a device driver, or a particular storage device. Figure 6 not shown in FIG. 1, can provide volatile memory for storage of information and / or instructions to be executed by processing unit 20. Although exemplary embodiments of the present application are described in the foregoing context of a fully functioning computer system, those of ordinary skill in the art will appreciate that the acts and parameter values described herein can be carried out in a computer system configured to function as a specific device, such as a virtual machine, a device driver, or a particular storage device.

[0156] Program / utility 40 having a set (at least one) of program modules 42 can be stored in, for example, system memory 28 by way of example, such program modules 42 include, but are not limited to, an operating system, one or more application programs, other program modules, and program data, each of which or a combination can include implementation of a network environment. Program modules 42 generally carry out the functions and / or methodologies of embodiments of the present application as described herein.

[0157] The electronic device 12 can also be in communication with one or more external devices 14 such as a keyboard, a pointing device, a display 24, etc.; can also be in communication with one or more devices that enable a user to interact with the electronic device 12; and / or can be in communication with any devices (such as a network card, a modem or the like) that enable the electronic device 12 to communicate with one or more other computing devices. Such communication can be via Input / Output (I / O) interface 22. Still yet, the electronic device 12 can be in communication with one or more networks (such as a local area network (LAN), a wide area network (WAN), and / or the Internet) through network adapter 20. As Figure 6 illustrated, network adapter 20 communicates to the other components of the electronic device 12 via bus 18. It should be appreciated that although not shown, other hardware and / or software modules could be used in conjunction with the electronic device 12. For example, microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data archival storage systems, etc., can be used in conjunction with the electronic device 12. Figure 6

[0158] Processing unit 16 executes instructions, which can be stored in system memory 28, as well as one or more sample data acquisition and various functional applications. For example, the processing unit 16 can execute instructions to implement a method for training a sperm key point data recognition model and a method for sperm key point data recognition. The method for training a sperm key point data recognition model can include:

[0159] obtaining a plurality of sperm sample images, and obtaining sample labels of the plurality of sperm sample images; wherein the sample labels include position region labels of regions where at least one part key point of a part of a sperm in the plurality of sperm sample images is located and sperm labels to which the part key point belongs.

[0160] iteratively performing the following training process until a sperm key point data recognition model is obtained:

[0161] inputting the plurality of sperm sample images into the sperm key point data recognition model being trained, and obtaining key point data recognition results output by the sperm key point data recognition model; and training the sperm key point data recognition model being trained based on the key point data recognition results and the sample labels, wherein the key point data recognition results include position region data of the regions where the part key points of the parts are located and region pixel data of the regions where the part key points are located.

[0162] The method for sperm key point data recognition can include:

[0163] obtaining an initial plurality of sperm images, and performing image preprocessing on the initial plurality of sperm images to obtain a plurality of sperm images.

[0164] ​obtain the key point data recognition result of the multi-sperm image based on the pre-trained sperm key point data recognition model; wherein the key point data recognition result comprises position region data of a region where a part key point of each part is located and region pixel data of the region where the part key point is located; and the sperm key point data recognition model is trained based on the sperm key point data recognition model training method in any of the above embodiments.

[0165] Of course, those skilled in the art can understand that the processor can also implement the technical solutions of the sample data acquisition method provided in any of the embodiments of the application.

[0166] Embodiment six

[0167] The embodiment six provides a computer readable storage medium, and a computer program is stored on the computer readable storage medium. The program is executed by a processor to implement steps of, for example, a sperm key point data recognition model training method and a sperm key point data recognition method provided in the embodiment. Optionally, the sperm key point data recognition model training method comprises:

[0168] a plurality of multi-sperm sample images are obtained, and sample labels of the multi-sperm sample images are obtained; wherein the sample labels comprise position region labels of a region where a part key point of at least one part of a sperm in a multi-sperm sample image is located and sperm labels to which the part key point belongs.

[0169] iteratively performing the following training process until the sperm key point data recognition model is obtained:

[0170] the multi-sperm sample images are input into the sperm key point data recognition model in training, and a key point data recognition result output by the sperm key point data recognition model is obtained, and the sperm key point data recognition model in training is trained based on the key point data recognition result and the sample labels, wherein the key point data recognition result comprises position region data of a region where a part key point of each part is located and region pixel data of the region where the part key point is located.

[0171] Optionally, the sperm key point data recognition method comprises:

[0172] an initial multi-sperm image is obtained, and the initial multi-sperm image is image preprocessed to obtain a multi-sperm image.

[0173] The key point data recognition model of the sperm is trained based on the sperm key point data recognition model training method in any of the preceding embodiments.

[0174] The computer storage medium of the embodiments of the present application can adopt any combination of one or more computer readable media. The computer readable medium can be a computer readable signal medium or a computer readable storage medium. The computer readable storage medium may, for example, but is not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or apparatus, or any combination thereof. More specific examples (non-exhaustive list) of the computer readable storage medium include an electrical connection having one or more wires, a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In this document, the computer readable storage medium can be any tangible medium containing or storing a program that can be used by or in connection with an instruction execution system, device or apparatus.

[0175] The computer readable signal medium can include a data signal propagating in a baseband or as part of a carrier wave propagating through transmission media, in which the computer readable program code is embodied. Such a propagating data signal can take many forms, including but not limited to electro-magnetic, optical, or any suitable combination thereof. The computer readable signal medium can also be any computer readable medium that is not a computer readable storage medium and that can communicate, propagate or transport program for use by or in connection with an instruction execution system, apparatus or device.

[0176] The program code contained on the computer readable medium can be transmitted by any suitable medium, including but not limited to wireless, wire, cable, RF, etc., or any suitable combination thereof.

[0177] Computer program code for carrying out operations of the present application can be written in any combination of one or more programming languages, including an object oriented programming language such as Java, Smalltalk, C++ or the like, and conventional procedural programming languages, such as the "C" programming language or similar programming languages. The program code can execute entirely on the user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer or entirely on the remote computer or server. In the latter scenario, the remote computer can be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or the connection can be made to an external computer (for example, through the Internet using an Internet Service Provider).

[0178] Those skilled in the art will appreciate that the modules or steps of the application described above can be implemented in a general purpose computer, and can be centralized in a single computer or distributed over multiple computers in a network, and optionally, they can be implemented in program code executable by a computer, and thus can be stored in a storage device and executed by a computer, or they can be made into individual integrated circuit modules, or a plurality of modules or steps can be made into a single integrated circuit module. Thus, the present application is not limited to any particular combination of hardware and software.

[0179] Note that the above are only the preferred embodiments of the present application and the principles of the applied technology. Those skilled in the art will understand that the present application is not limited to the specific embodiments described herein, and that various obvious changes, re-adjustments and substitutions can be made by those skilled in the art without departing from the scope of the present application. Therefore, although the present application has been described in detail through the above embodiments, the present application is not limited to the above embodiments, and can include more other equivalent embodiments without departing from the concept of the present application, and the scope of the present application is determined by the scope of the appended claims.

Claims

1. A method for training a sperm key point data recognition model, characterized in that, The method comprises: obtaining a plurality of images of a plurality of sperm samples, and obtaining sample labels of the plurality of images of the plurality of sperm samples; wherein the sample labels comprise position region labels of regions in which at least one part of a sperm in the plurality of images of the plurality of sperm samples is located and sperm labels to which the part of the sperm belongs; obtaining the sample labels of the plurality of images of the plurality of sperm samples comprises: for any part of a part of a key point, obtaining a key point coordinate of a current part of a key point in the plurality of images of the plurality of sperm samples, and determining a preset position region in the plurality of images of the plurality of sperm samples containing the part of the key point based on the key point coordinate; determining a position distance between each position coordinate in the preset position region and the key point coordinate; generating a position region heat map of the preset position region in which the current part of the key point is located based on the each position coordinate and the corresponding position distance; iteratively performing the following training process until the training is completed to obtain a sperm key point data recognition model: inputting the plurality of images of the plurality of sperm samples into the sperm key point data recognition model in training, and obtaining a key point data recognition result output by the sperm key point data recognition model in training, and training the sperm key point data recognition model in training based on the key point data recognition result and the sample labels, wherein the key point data recognition result comprises position region data of regions in which the parts of the key points are located and region pixel data of the regions in which the parts of the key points are located; the position region data in which the part of the key point is located comprises position probability data of each position coordinate in the position region being the current part of the key point; correspondingly, after obtaining the key point data recognition result output by the sperm key point data recognition model, the method further comprises: obtaining a preset key point position threshold, and determining a key point position of each part of the key point in the plurality of images of the plurality of sperm samples based on the position probability data in the position region data and the key point position threshold.

2. The method of claim 1, wherein, the training of the sperm key point data recognition model in training based on the key point data recognition result and the sample labels comprises: obtaining a position region loss function generated based on the position region data of the regions in which the parts of the key points are located and the position region labels of the regions in which the parts of the key points are located, and a pixel loss function generated based on the pixel data of the positions of the parts of the key points and the sperm labels to which the parts of the key points belong, and training the sperm key point data recognition model in training based on the position region loss function and the pixel loss function.

3. The method of claim 2, wherein, the pixel loss function is determined based on a pull-in loss function between parts of key points of the same sperm and a push-away loss function between parts of key points of different sperm.

4. The method of claim 3, wherein: the pull-in loss function is determined based on pixel data of positions of parts of key points in a current sperm and average data of the pixel data of the parts; the push-away loss function is determined based on average data of the parts in the current sperm and average data of the parts in other sperm in the plurality of images of the plurality of sperm.

5. The method of claim 1, wherein, The sperm key point data recognition model comprises a pixel processing module, a feature extraction module, a feature splicing module and a recognition module. The pixel processing module is configured to perform image scaling processing on the multi-sperm image. The feature extraction module comprises at least one level of feature extraction sub-modules, and the feature extraction sub-modules of the first level are connected with the pixel processing module. Any level of feature extraction sub-module is configured to take the output data of the pixel processing module or at least one feature extraction sub-module of the previous level as input data after performing down-sampling processing with a preset stride, perform feature extraction on the input data and output the feature extraction result. The feature splicing module is connected with each level of feature extraction sub-module, and is configured to perform feature splicing on the output data of each level of feature extraction sub-module after performing up-sampling processing with a preset stride, and obtain spliced features output by the feature splicing module. 6.A method for identifying sperm key point data, characterized in that, The recognition module is connected with the feature splicing module, and is configured to determine the key point data recognition result of the current frame of multi-sperm image based on the spliced features obtained through feature splicing processing. The method comprises the following steps: An initial multi-sperm image is obtained, and image preprocessing is performed on the initial multi-sperm image to obtain a multi-sperm image. 7.A device for training a sperm key point data recognition model, characterized in that, A sperm key point data recognition model is determined based on a pre-trained sperm key point data recognition model, wherein the key point data recognition result comprises position region data of a region where a part key point of each part is located and region pixel data of the region where the part key point is located. The sperm key point data recognition model is trained based on the sperm key point data recognition model training method of any one of claims 1-5. The method comprises the following steps: A sample image and label acquisition module is configured to acquire a plurality of multi-sperm sample images and sample labels of the multi-sperm sample images, wherein the sample labels comprise position region labels of a region where a part key point of at least one part of a sperm in a multi-sperm sample image is located and sperm labels to which the part key point belongs. The sample image and label acquisition module comprises: A preset position region acquisition unit is configured to acquire key point coordinates of a current part key point in a multi-sperm sample image for any part key point, and determine a preset position region in the multi-sperm sample image containing the part key point based on the key point coordinates. A position distance determination unit is configured to determine position distances between each position coordinate in the preset position region and the key point coordinates. A position region heat map of the preset position region where the current part key point is located is generated based on the position region and the corresponding position distances. A model training module is configured to iteratively perform the following training process until a sperm key point data recognition model is obtained: The multi-sperm sample image is input into the sperm key point data recognition model in training, and key point data recognition results output by the sperm key point data recognition model are obtained, and the sperm key point data recognition model in training is trained based on the key point data recognition results and the sample label, wherein the key point data recognition results include position region data of a region where each part key point is located and region pixel data of the region where each part key point is located; The position region data where the part key point is located includes position probability data of each position coordinate in the position region being a current part key point; Correspondingly, the apparatus further includes: A key point position determination unit configured to obtain a preset key point position threshold, and determine key point positions of each part key point in the multi-sperm image based on position probability data in the position region data and the key point position threshold.

8. A sperm key point data recognition apparatus, characterized by, The apparatus includes: A multi-sperm image acquisition module configured to acquire an initial multi-sperm image, and perform image preprocessing on the initial multi-sperm image to obtain a multi-sperm image; A key point data recognition result determination module configured to determine key point data recognition results of the multi-sperm image based on a sperm key point data recognition model pre-trained, wherein the key point data recognition results include position region data of a region where each part key point is located and pixel data of the region where each part key point is located, and the sperm key point data recognition model is obtained by training based on the sperm key point data recognition model training method in any one of claims 1-5.

9. An electronic device, comprising: The apparatus includes: One or more processors; A storage device configured to store one or more programs, When the one or more programs are executed by the one or more processors, the one or more processors implement the sperm key point data recognition model training method in any one of claims 1-5 and / or the sperm key point data recognition method in claim 6.

10. A computer-readable storage medium having stored thereon a computer program, characterized in that, The program is executed by the processor to implement the sperm key point data recognition model training method in any one of claims 1-5 and / or the sperm key point recognition method in claim 6.

Images