Method and system for pronucleus detection in multi-focal embryo images based on optimal arc fusion

By adopting the optimal arc fusion technology in embryonic cell images, the problem of prokaryotic unclearity caused by multifocal length is solved, the detection accuracy and efficiency are improved, the clarity of prokaryotic in the fusion image is ensured, and the model training data is optimized.

CN118506115BActive Publication Date: 2024-11-19WUHAN MUTUAL UNITED TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410981378.8
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-07-22
Publication Date
2024-11-19
Estimated Expiration
2044-07-22

AI Technical Summary

Technical Problem

In embryonic cell images, multifocal length leads to unclear prokaryotic problems, affecting model detection and embryo quality assessment.

Method used

Using the multi-focal embryo image detection method based on the best arc fusion, the three images with the highest definition are selected through clarity evaluation to fuse, the images are evenly divided and elliptical arcs are recognized, and the rectangular block with the smallest error or the highest definition is selected for splicing. Form a high-quality fusion image for object detection.

Benefits of technology

It improves the accuracy and reliability of prokaryotic detection, reduces manual participation, reduces errors caused by human factors, and improves image processing speed and efficiency.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118506115B_ABST
    Figure CN118506115B_ABST
Patent Text Reader

Abstract

The present invention discloses a method and system for detecting pronuclei of multi-focal embryo images based on optimal arc fusion. The method evaluates the clarity of a group of multi-focal embryo images, selects three images with the highest clarity as target images, adjusts the three target images to the same size, divides each target image into rectangular blocks of equal size according to the same division method, performs edge detection and ellipse fitting on the rectangular blocks at the same position in the three target images to identify the optimal rectangular blocks and obtain a fused image, inputs the fused image into a target detection model, and detects the pronuclei in the multi-focal embryo images. The present invention solves the problem that the pronuclei of some images in multi-focal embryo cell images are not clear.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a multi-focal length embryo image pronucleus detection method and system based on optimal arc fusion. Background Art

[0002] In the medical field, especially in embryology research, manual identification of a large number of images is a time-consuming and professional task. Therefore, using computer technology to build visual models to assist doctors in quickly and accurately detecting pronuclear features has a profound impact on this research field. This technology can greatly improve work efficiency, reduce the workload of doctors, and improve the accuracy of diagnosis. However, the current embryonic pronuclear feature detection technology still faces challenges:

[0003] (1) In embryo cell imaging, determining the appropriate focal length is an important research topic. Different from blastomere detection, in the same device, the optimal focal length of blastomere is fixed, but the position of the pronucleus in the blastomere is uncertain, so its optimal focal length is uncertain;

[0004] (2) When multiple pronuclei exist at the same time and are located at different horizontal planes, different pronuclei require different optimal focal lengths to obtain a clear image. Therefore, when shooting with different focal lengths, one pronucleus may be clear while another is not.

[0005] These issues will affect the model's detection of prokaryotic cells and, in turn, the assessment of embryo quality. Summary of the invention

[0006] The present invention proposes a method and system for detecting pronuclei of multi-focal embryo images based on optimal arc fusion, which solves the problem that the pronuclei of some images in multi-focal embryo cell images are not clear.

[0007] In order to solve the above technical problems, the present invention provides a multi-focal length embryo image pronucleus detection method based on optimal arc fusion, comprising the following steps:

[0008] Step S1: Evaluate the clarity of a group of multi-focal length embryo images and select the three images with the highest clarity as target images;

[0009] Step S2: adjust the three target images to the same size, divide each target image into rectangular blocks of equal size according to the same division method, perform edge detection and ellipse fitting on the rectangular blocks at the same position in the three target images to identify the arcs in the rectangular blocks:

[0010] If only one rectangular block recognizes the arc, the rectangular block is taken as the best rectangular block; if multiple rectangular blocks recognize the arc, the rectangular block corresponding to the arc with the smallest error during ellipse fitting is taken as the best rectangular block; if no rectangular block recognizes the arc, the clarity of each rectangular block is evaluated, and the rectangular block with the highest clarity is selected as the best rectangular block; the best rectangular blocks at all positions are spliced ​​to obtain a fused image;

[0011] Step S3: inputting the fused image into the target detection model to detect the pronuclei in the multi-focal length embryo image.

[0012] Preferably, in step S1, the Laplacian operator is used to measure the clarity of the embryo image:

[0013] ;

[0014] In the formula, is the Laplace operator; f represents the gray value of the embryo image; represents the differential, represents the second derivative; , are the horizontal and vertical coordinates of the pixel points respectively.

[0015] Preferably, step S2 obtains an edge point set by performing edge detection on the rectangular block, including the following steps: smoothing the rectangular block; calculating the gradient of all pixel points in the rectangular block, setting a gradient threshold, and taking pixel points with gradients greater than the gradient threshold as edge points to obtain an edge point set.

[0016] Preferably, a Gaussian filter is used to smooth the rectangular block, and the expression of the Gaussian filter is:

[0017] ;

[0018] In the formula, is the pixel after smoothing; is the standard deviation of the Gaussian kernel; , are the horizontal and vertical coordinates of the pixel points respectively.

[0019] Preferably, the expression for calculating the gradient of the pixel points in the rectangular block is:

[0020] ;

[0021] ;

[0022] In the formula, , Pixels Gradient magnitude in horizontal and vertical directions; is the gray value of the rectangular block.

[0023] Preferably, step S2 performs ellipse fitting on the edge point set to detect whether there is a circular arc in the edge point set, including the following steps: using an ellipse fitting algorithm to fit the ellipse arc equation of the edge point set, calculating the fitting error between the fitted ellipse arc and the edge point, if the fitting error is less than a set error threshold, it is considered that a circular arc exists.

[0024] Preferably, the method for calculating the fitting error between the fitted elliptical arc and the edge point is: The horizontal coordinate of the edge point x Substitute the elliptical arc equation to obtain the fitting ordinate Y ,Will As the fitting error, the expression of the elliptical arc equation is:

[0025] ;

[0026] In the formula, A , B are the major and minor axes of the ellipse respectively; a , b are the horizontal and vertical coordinates of the center of the ellipse respectively.

[0027] The present invention also provides a multi-focal length embryo image pronucleus detection system based on optimal arc fusion, which is implemented based on the above-mentioned multi-focal length embryo image pronucleus detection method based on optimal arc fusion, and includes an image input module, a feature extraction module, a feature fusion module and a target detection module;

[0028] The image input module is used to preprocess the input embryo image and input the preprocessed image into the feature extraction module;

[0029] The feature extraction module is used to extract features from the preprocessed embryo image to obtain a corresponding feature map;

[0030] The feature fusion module is used to perform feature fusion on the feature map to obtain a fused feature map;

[0031] The target detection module is used to perform feature detection on the fused feature map.

[0032] Preferably, the image input module adopts adaptive image scaling technology to adjust all embryo images in the multi-focal embryo image to the same size; selects the three embryo images with the highest clarity in the multi-focal embryo image as the target image set; divides each target image in the target image set into the same number of rectangular blocks, performs arc segment detection on the rectangular blocks at the same positions of the three target images, and obtains the best rectangular block at each position according to the arc segment detection results; and splices the best rectangular blocks at all positions to obtain a fused image.

[0033] Preferably, the feature fusion module combines the feature pyramid network FPN and path aggregation networks PAN The combination enhances the feature fusion ability of the network.

[0034] The benefits of the present invention include at least:

[0035] 1. Evaluate the clarity of multi-focal embryo images and select the three images with the highest clarity for subsequent image fusion, ensuring the quality of the selected images and providing high-quality input for subsequent pronucleus detection;

[0036] 2. By evenly dividing the image and identifying arcs, the pronucleus detection algorithm can focus on the most important part of the image, that is, the area where the pronucleus is located, thereby optimizing the data used for model training;

[0037] 3. The automated image processing and fusion process reduces manual involvement, reduces errors caused by human factors, and improves image processing speed and efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 Schematic diagram of multi-nuclear targets in embryonic cells at different focal lengths;

[0039] Figure 2 It is a flowchart of the prokaryotic detection algorithm;

[0040] Figure 3 Schematic diagram of the process of image preprocessing method in an embodiment of the present invention;

[0041] Figure 4 Schematic diagram of the clarity of different pronuclei at different focal lengths;

[0042] Figure 5 It is a flowchart of the optimal arc area fusion algorithm;

[0043] Figure 6 This is a comparison chart of the effects of prokaryotic detection. DETAILED DESCRIPTION

[0044] The following is a clear and complete description of the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work belong to the protection scope of the present invention.

[0045] The present invention successfully collected 2000 groups of embryo images taken at different focal lengths from multiple reproductive centers, 11 in each group, and these images were taken in a time-difference incubator, covering the embryonic development process of the D1-D3 stage. This large amount of data provides a solid foundation for the research of the present invention, and under the guidance of professional knowledge and experienced embryologists, the pronucleus is marked. When multiple pronuclei coexist, because different pronuclei are located on different focal planes, when observed at different focal lengths, some pronuclei will be clearly visible, while others are blurred. This has brought great confusion to conventional pronucleus detection. To solve this problem, the present invention adopts a simple image fusion technology, by fusing the pronucleus images taken at different focal lengths, the advantages of clear pronuclei under each focal length can be comprehensively utilized, thereby obtaining a pronucleus image with full focus and clarity. This helps to improve the accuracy and reliability of pronucleus detection.

[0046] Figure 1 A set of images of multinucleated embryonic cells at different focal lengths are shown. Figure 1 It can be seen that embryonic cells have transparent or translucent optical properties under the irradiation of an electron microscope, so that the pronuclei in the cells can be observed. However, the clarity of the image is different at different focal lengths. Due to the different relative horizontal positions of the pronuclei, the best focal lengths for different embryonic cells and different pronuclei in the same embryonic cell are different. The best focal length of one embryonic cell may not be the best focal length of another embryonic cell. Figure 1 In the figure, we can see that in some images taken at different focal lengths, one pronucleus is clear while the other is not, as shown in (4) and (6). The pronucleus with the highest clarity is marked in the figure. These problems pose great challenges to the subsequent detection of the pronucleus.

[0047] To address these challenges, the present invention first selects the three clearest images at multiple focal lengths using a clarity evaluation operator, then uses an image fusion method to comprehensively utilize the clear areas in the three images, and finally uses the fused images to join the training set to obtain a high-quality model. The algorithm flow is as follows: Figure 2 As shown, the following is the implementation process of this example:

[0048] 1. Data Collection Phase

[0049] The present invention collected tens of thousands of images from different reproductive centers. These images were taken in a time-difference incubator, covering embryo images from D1 to D3, which contain multinucleated embryonic cells. Considering that different pronuclei require different optimal focal lengths to obtain clear images, 11 images were taken at different focal lengths for embryonic cells in each state as a group, and 2,000 groups of embryonic images containing pronuclei were obtained.

[0050] 2. Data preprocessing stage

[0051] like Figure 3 As shown, the present invention further improves the recognizability of the pronucleus in the image by selecting the rectangular block with the smallest ellipse fitting error or the highest clarity from multiple rectangular blocks in three pictures as the best rectangular block, and splicing these best rectangular blocks to obtain a fused image.

[0052] (1) Clarity screening

[0053] Since each set of images contains 11 different focal lengths of shooting information, there are some poor quality pictures, which are not needed in the subsequent image fusion and model training stage. Conventional manual selection of high-definition pictures requires a lot of time and energy, and the speed is far from meeting the requirements of real-time detection. The present invention measures the clarity of embryo images by the Laplace operator:

[0054] ;

[0055] In the formula, is the Laplace operator; f is the grayscale value or brightness value function of the embryo image; represents the differential, Represents the second derivative, which describes the image at the pixel point The second differential characteristic at , that is, the degree of change of local grayscale and brightness; , are the horizontal and vertical coordinates of the pixel points respectively.

[0056] Taking into account the influence of background and algorithm accuracy, the three images with the highest clarity are selected as the sample set used for fusion. Through the above screening, 2000 groups of sample sets can be obtained, each of which includes three embryo images containing pronuclei.

[0057] (2) Image fusion

[0058] Figure 4A set of sample sets is shown. It can be clearly seen that the pronucleus outline on the left side of the first picture is clearly visible, the pronucleus edge on the right is relatively blurred, and the second picture is the opposite. It can be seen that the optimal focal length required for shooting different pronuclei is different. How to comprehensively utilize the advantages shown by two pictures with different focal lengths and learn from each other's strengths is an important topic of research. The current mainstream image fusion methods, such as feature-level fusion, multi-scale fusion and other algorithms, have achieved good results in many scenes that need to be fused. However, unlike the images in daily scenes, the pronucleus in the embryonic image is a small target, and the proportion of the whole picture is very small. Secondly, the texture of the pronucleus and other features are very similar to the background, so the performance of these fusion schemes in embryonic cells is often unsatisfactory. In this regard, the present invention designs a simple optimal arc segment area fusion for the morphological characteristics of the pronucleus, which effectively avoids the challenges of the pronucleus target being too small, the morphological difference being small, etc.

[0059] like Figure 5 As shown in the figure, firstly, the embryo images in a set of samples are adjusted to the same size, and then the three embryo images are cut to make each The image is cut into Then, arc segment detection is performed on the small rectangles cut out from the same positions of the three pictures.

[0060] For each rectangular block, the Canny operator is used for edge detection. First, the rectangular block is smoothed by Gaussian filtering:

[0061] ;

[0062] In the formula, is the pixel after smoothing; is the standard deviation of the Gaussian kernel, which can be set to 1-5 and is an adjustable variable; , are the horizontal and vertical coordinates of the pixel points respectively.

[0063] Then the Sobel operator is used to calculate the horizontal and vertical gradient amplitudes of all pixels in the rectangular block to obtain the gradient of the pixel:

[0064] ;

[0065] ;

[0066] In the formula, , Pixels Gradient magnitude in horizontal and vertical directions; is the gray value of the rectangular block.

[0067] By setting a gradient threshold , taking the pixel points whose gradient is greater than the gradient threshold as edge points, a series of edge point sets are obtained. For these edge point sets, an ellipse arc equation is fitted using the ellipse fitting algorithm in MATLAB:

[0068] ;

[0069] In the formula, A , B are the major and minor axes of the ellipse respectively; a , b are the horizontal and vertical coordinates of the center of the ellipse respectively.

[0070] For the fitted elliptical arc, the horizontal coordinates of the edge points are x Substitute into the elliptical arc equation to get the fitting ordinate Y ,Will As the fitting error, if the fitting error is less than the set error threshold, it is considered that the arc exists.

[0071] For all the rectangular blocks in the above three pictures, the rectangular blocks at the corresponding positions are compared. If only one rectangular block recognizes the arc and the other two do not, the quality of the rectangular block that can detect the arc is considered to be the best, and this rectangular block is taken as the best rectangular block; if multiple rectangular blocks recognize the arc, by comparing the average fitting error between the fitted arc and all pixel points, the rectangular block with the smaller average fitting error is taken as the best rectangular block; if no rectangular block recognizes the arc, the Laplace operator is used to calculate the clarity of each rectangular block, and the rectangular block with the highest clarity is selected as the best rectangular block. Finally, the best rectangular blocks at all positions are spliced ​​and fused to obtain the desired image.

[0072] (3) Dataset division and production stage

[0073] The fused images in step (2) were divided into two data sets at a ratio of 9:1. The former and the images screened by clarity in step (1) were added to data set 1 and annotated under the guidance of an embryologist. The most appropriate labels were selected by voting as the data set, which was then divided into a training set and a validation set at a ratio of 9:1. The latter was used as a test set for model detection and evaluation.

[0074] 3. Model training phase

[0075] (1) Algorithm model construction

[0076] In the field of embryo detection, the practical application of the invention needs to ensure the operability of the detection. The present invention uses Yolov5 as the backbone network. The framework of Yolov5 includes four different components: input, backbone, neck and prediction. It is particularly noteworthy that the input component adopts adaptive image scaling technology, which can seamlessly adapt to embryo images of different sizes; in YOLOv5, this adaptive anchor frame is embedded in the entire training code, so before each training starts, it will adaptively calculate the optimal anchor frame value required by different training sets according to different data sets. The backbone network is usually a feature extraction network, which bears the important responsibility of extracting relevant information from the input image for subsequent network use. By extracting features from the input image containing the target, the corresponding feature map is obtained. In target detection, the backbone network is the basic part of the deep learning model, which is responsible for extracting features from the input image for subsequent tasks. In addition, the neck component adopts a structure that combines the feature pyramid network and the path aggregation network, thereby realizing a powerful fusion of the feature pyramid network (FPN) and the path aggregation network (PAN) method. In addition, by adopting the CSP2 structure designed by CSPNet, the network feature fusion capability is enhanced.

[0077] (2) Data augmentation

[0078] In the experimental part of this embodiment, an exploratory study was conducted on the data enhancement technology for images. Specifically, two common data expansion methods were focused on: image cropping and image flipping. First, the original image data set was randomly cropped, and different cropping ratios were set during the random cropping process to generate cropped images of different sizes. The cropping ratios were set to 0.8, 0.6, and 0.4, respectively, and the sizes of the generated cropped images were 80%, 60%, and 40% of the original images, respectively. Through this random cropping, the position change of the target object in the image can be simulated, and the robustness of the model to the change of the target position is enhanced. Secondly, the original image data set was horizontally flipped and vertically flipped. Horizontal flipping can simulate the left-right direction change of the target object, while vertical flipping can simulate the change of the up-down direction. Through this flipping operation, the number of training samples can be greatly increased, thereby improving the generalization ability of the model.

[0079] (3) Feature extraction

[0080] The training set The image is taken as input, and the feature information is extracted through the backbone network in a forward propagation manner to obtain the corresponding feature map; the feature maps obtained in certain layers of the backbone network are then sent to the neck area for feature fusion, generally some feature pyramids are used for feature fusion; finally, the fused feature maps are sent to the detection head for target classification, identification and positioning.

[0081] (4) Model training

[0082] For the existing data set, the embodiment of the present invention performs training for 100 time steps epochs, and the initial learning rate is 0.01. The software environment used in the experiment includes Ubuntu 18.04, CUDA 10.2, Python 3.7 and Torch 1.8.

[0083] 3. Image detection stage

[0084] The input image is divided into fixed-size grid cells, each of which contains multiple predicted bounding boxes. The algorithm determines the target category prediction for each cell by selecting the bounding box with the highest confidence as the primary prediction box, and uses a softmax classifier for multi-category classification. The non-maximum suppression method NMS is applied to remove overlapping detection boxes and only retain the most confident bounding boxes. Finally, the bounding boxes with lower confidence are filtered out according to the set confidence threshold, and the detected target category and location information are output. This method combines the confidence and category prediction of the bounding box to achieve the target detection task.

[0085] Based on the above research, the present invention proposes a multi-focal length embryo image pronucleus detection method based on optimal arc fusion, comprising the following steps:

[0086] S1 collects embryo images containing multi-nucleus phenomena, photographs embryo cells at different focal lengths, and constructs an embryo image dataset;

[0087] S2 uses the Laplace operator to measure the clarity of each group of multi-focal length images, selects the three clearest images as the target images of each group, and constructs data set 1;

[0088] S3 fuses the three images into one image through the optimal arc area, constructs dataset 2, and divides it into training set and test set according to 9:1;

[0089] S4 adds the dataset 1 obtained by S2 to the training set obtained by S3, annotates it under the guidance of an embryologist, and then divides it into a training set and a validation set in a ratio of 9:1;

[0090] S5 uses Yolov5 to train the prokaryotic detection model;

[0091] S6 uses the model obtained in S5 to perform prokaryotic detection on the test set, and the statistical results of prokaryotic detection are as follows: Figure 6 .

[0092] pass Figure 6 It can be seen that the method of the present invention overcomes the problem that the model is difficult to identify the prokaryotic target detection of embryonic cells from two aspects:

[0093] 1. When multiple pronuclei exist at the same time and are located on different horizontal planes, the horizontal position of the phase will cause different pronuclei to require different optimal focal lengths to obtain a clear image; therefore, when shooting with different focal lengths, one pronucleus may be clear while the other is not. To address this problem, the Laplace operator is used to calculate the clarity of images with different focal lengths, and the three images with the highest clarity are selected for subsequent image fusion, thus solving the focusing problem of shooting pronuclei.

[0094] 2. Different from blastomere detection, in the same device, the optimal shooting focal length of blastomere is fixed, but the position of pronucleus in blastomere is uncertain, so its optimal focal length is uncertain. To solve this problem, the regional block fusion method based on optimal arc is adopted to comprehensively utilize the clarity advantage areas under different focal lengths, which greatly improves the detection accuracy.

[0095] The present invention provides a multi-focal length embryo image pronucleus detection method based on optimal arc fusion, which obtains a plurality of rectangular blocks by evenly dividing the image, and selects the rectangular block with the smallest ellipse fitting error or the highest clarity as the optimal rectangular block among the rectangular blocks, so that the target detection algorithm can focus on the most important part of the image, that is, the area where the pronucleus is located, thereby optimizing the data used for model training, further improving the recognizability of the pronucleus in the image, effectively solving the problem of inconsistent pronucleus clarity caused by different focal lengths, and ensuring the clarity of the pronucleus in the final fused image.

[0096] The technical features of the above embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. Only the preferred embodiments of the present invention are expressed. The description is more specific and detailed, but it cannot be understood as limiting the scope of the patent of the present invention. As long as there is no contradiction in the combination of these technical features, they should be considered as the scope recorded in this specification.

[0097] It should be noted that, for those skilled in the art, several modifications and improvements can be made without departing from the concept of the present invention, and these modifications and improvements all belong to the protection scope of the present invention. Therefore, the protection scope of the patent of the present invention shall be subject to the attached claims.

Claims

1. A method for pronucleus detection of multi-focal embryo images based on optimal arc fusion, characterized in that: The following steps are involved: Step S1: Evaluate the clarity of a group of multi-focal length embryo images and select the three images with the highest clarity as target images; Step S2: adjust the three target images to the same size, divide each target image into rectangular blocks of equal size according to the same division method, perform edge detection and ellipse fitting on the rectangular blocks at the same position in the three target images to identify the arcs in the rectangular blocks: The method for identifying the circular arc in the rectangular block is as follows: performing edge detection on the rectangular block to obtain an edge point set, using an ellipse fitting algorithm to fit an ellipse arc equation of the edge point set, and calculating the fitting error between the fitted ellipse arc and the edge point. If the fitting error is less than a set error threshold, it is considered that the circular arc exists; For all the rectangular blocks of the three target images, the rectangular blocks at the corresponding positions are compared. If only one rectangular block recognizes the arc and the other two do not, the quality of the rectangular block that can detect the arc is considered to be the best, and this rectangular block is taken as the best rectangular block; if multiple rectangular blocks recognize the arc, by comparing the average fitting error between the fitted arc and all pixel points, the rectangular block with the smaller average fitting error is taken as the best rectangular block; if no rectangular block recognizes the arc, the clarity of each rectangular block is evaluated, and the rectangular block with the highest clarity is selected as the best rectangular block; the best rectangular blocks at all positions are spliced ​​to obtain a fused image; Step S3: inputting the fused image into the target detection model to detect the pronuclei in the multi-focal length embryo image.

2. A multi-focal length embryo image prokaryotic detection method based on optimal arc fusion according to claim 1, characterized in that: In step S1, the Laplacian operator is used to measure the clarity of the embryo image: Where, Laplace is the Laplace operator; f represents the gray value of the embryo image; represents the differential, Represents the second derivative; x and y are the horizontal and vertical coordinates of the pixel point respectively.

3. A multi-focal length embryo image prokaryotic detection method based on optimal arc fusion according to claim 1, characterized in that: Step S2 obtains an edge point set by performing edge detection on the rectangular block, including the following steps: smoothing the rectangular block; calculating the gradient of all pixel points in the rectangular block, setting a gradient threshold, and taking pixel points with gradients greater than the gradient threshold as edge points to obtain an edge point set.

4. A multi-focal length embryo image prokaryotic detection method based on optimal arc fusion according to claim 3, characterized in that: Gaussian filtering is used to smooth the rectangular block. The expression of the Gaussian filtering is: Where G(x,y) is the pixel after smoothing; σ is the standard deviation of the Gaussian kernel; x and y are the horizontal and vertical coordinates of the pixel, respectively.

5. A multi-focal length embryo image prokaryotic detection method based on optimal arc fusion according to claim 3, characterized in that: The expression for calculating the gradient of a pixel in a rectangular block is: In the formula, G x (x,y),G y (x, y) are the gradient amplitudes in the horizontal and vertical directions of the pixel point (x, y); I gray is the gray value of the rectangular block.

6. A multi-focal length embryo image prokaryotic detection method based on optimal arc fusion according to claim 3, characterized in that: The method for calculating the fitting error between the fitted elliptical arc and the edge point is: substitute the abscissa x of the edge point with coordinates (x, y) into the elliptical arc equation to obtain the fitting ordinate Y, and substitute (yY) 2 As the fitting error, the expression of the elliptical arc equation is: Where A and B are the major and minor axes of the ellipse respectively; a and b are the horizontal and vertical coordinates of the center of the ellipse respectively.

7. A multi-focal length embryo image prokaryotic detection system based on optimal arc fusion, which is implemented based on a multi-focal length embryo image prokaryotic detection method based on optimal arc fusion as claimed in any one of claims 1 to 6, characterized in that: It includes image input module, feature extraction module, feature fusion module and target detection module; The image input module is used to preprocess the input embryo image and input the preprocessed image into the feature extraction module; The feature extraction module is used to extract features from the preprocessed embryo image to obtain a corresponding feature map; The feature fusion module is used to perform feature fusion on the feature map to obtain a fused feature map; The target detection module is used to perform feature detection on the fused feature map.

8. A multi-focal length embryo image prokaryotic detection system based on optimal arc fusion according to claim 7, characterized in that:The image input module adopts adaptive image scaling technology to adjust all embryo images in the multi-focal embryo image to the same size; selects three embryo images with the highest clarity in the multi-focal embryo image as a target image set; divides each target image in the target image set into an equal number of rectangular blocks, performs arc segment detection on the rectangular blocks at the same position of the three target images, and obtains the best rectangular block at each position according to the arc segment detection result; and splices the best rectangular blocks at all positions to obtain a fused image.

9. The multi-focal length embryo image pronucleus detection system based on optimal arc fusion according to claim 7, characterized in that: The feature fusion module combines the feature pyramid network FPN and the path aggregation network PAN to enhance the feature fusion capability of the network.