Chromosome identification method and system

Abstract

The present invention discloses a chromosome recognition method and system. The method comprises obtaining chromosome data; training a diffusion probability model according to the chromosome data; embedding a chromosome mask binary image and a chromosome grayscale normalized histogram of the chromosome image and a chromosome category into a feature vector, using a cross-attention mechanism to apply the feature vector to the diffusion probability model to generate chromosome expansion data; training a chromosome recognition network according to the chromosome expansion data; and identifying the chromosome data to be identified according to the chromosome recognition network; the data amplification method based on the diffusion probability model of the present invention not only retains the authenticity of the chromosome, but also successfully converts the image characteristics of the chromosome, and effectively expands the diversity of the data. Through this system, the recognition effect of chromosomes on different data sources has been significantly improved, providing a more accurate and reliable auxiliary tool for the field of medical diagnosis.

Description

Technical Field

[0001] The present invention relates to the field of image processing technology, and in particular to a chromosome recognition method and system. Background Art

[0002] In the field of medical diagnosis, especially in the detection of genetic diseases and blood diseases, chromosome recognition technology plays a vital role, and its effect is directly restricted by the quantity and quality of chromosome data. At present, chromosome data mainly comes from major hospitals, but the differences in the production technology level between hospitals are significant, resulting in the diversity of chromosome images in image characteristics such as color or brightness distribution. However, the training of chromosome recognition models is often limited to data sets with limited scale and relatively single image characteristics such as color or brightness distribution, which seriously restricts the generalization performance of the model when facing new data sources or chromosome images with different image characteristics. Traditional data augmentation methods, such as rotation and scaling, can increase the amount of data but fail to change the essential characteristics of the data; while methods based on generative models such as Generative Adversarial Networks (GAN), although they can generate data similar to the input, cannot effectively simulate the differences between different data sources. Summary of the invention

[0003] In the field of medical diagnosis, especially in the detection of genetic diseases and blood diseases, chromosome recognition technology plays a vital role, and its effect is directly restricted by the quantity and quality of chromosome data. At present, chromosome data mainly comes from major hospitals, but the differences in the production technology level between hospitals are significant, resulting in the diversity of chromosome images in image characteristics such as color or brightness distribution. However, the training of chromosome recognition models is often limited to data sets with limited scale and relatively single image characteristics such as color or brightness distribution, which seriously restricts the generalization performance of the model when facing new data sources or chromosome images with different image characteristics. Traditional data augmentation methods, such as rotation and scaling, can increase the amount of data but fail to change the essential characteristics of the data; while methods based on generative models such as Generative Adversarial Networks (GAN), although they can generate data similar to the input, cannot effectively simulate the differences between different data sources. ;

[0004] Where t represents the tth step of the diffusion process, α t , β t are artificially set hyperparameters, which respectively represent the proportion of original data information retained at time t and the intensity of noise added at time t. , ,∈ θ (X t, t) is the noise data predicted by the deep convolutional neural network, ∈ is a random sample in the standard Gaussian distribution.

[0005] Furthermore, the loss function of the diffusion probability model is:

[0006] ;

[0007] Where t represents the tth step of the diffusion process, L t represents the loss function value at the t-th step of diffusion, x0 is the target chromosome image during training, ∈ is a random sampling in the standard Gaussian distribution, is a manually set hyperparameter, is the mathematical expectation, ∈ θ It is the result predicted by the neural network.

[0008] Furthermore, in step 3, the cross attention mechanism splits the input tensor into two parts and , and then use one part as the query set and the other part as the key set and value set; the output is a tensor of size n×d2, and for each row vector, the attention weights for all row vectors are given;

[0009] Let Q = X1W Q and K=V=X2W K , then the cross attention is calculated as follows:

[0010] ;

[0011] in, , is the learned projection matrix, d K is the dimension of the key set, Q is the query feature vector for calculating the attention weight, K is the index feature vector for calculating the attention weight, V is the content feature vector for calculating the attention weight, and K T is the transpose of K, d1 and d2 are the dimensions of X1 and X2.

[0012] Furthermore, generating chromosome expansion data in step 3 includes:

[0013] The first step is to extract the mask binary image C of chromosome C mask ; The mask binary image C mask The background is 0, the foreground is 1, and the size is the same as that of chromosome C, which is H×W. mask The category of chromosome C is used as two conditions and embedded into two feature vectors;

[0014] The second step is to extract the specified chromosome C styleThe chromosome grayscale normalized histogram in is obtained, and the chromosome grayscale normalized histogram is embedded into a feature vector; the calculation formula of the chromosome grayscale normalized histogram is as follows:

[0015] ;

[0016] in, p ( r k ) represents the relative frequency of each gray level in the image. r k Represents the gray level of a pixel, which is a discrete value between 0 and 255. n k Is grayscale r k The number of MN is the total number of pixels in the image, M , N Respectively represent the length and width of the image;

[0017] The feature vector of the chromosome grayscale normalized histogram has 256 dimensions. The calculated 256-dimensional feature vector is compressed to 16 dimensions. The calculation process is as follows:

[0018] Assume that the generated 256-dimensional vector is H c ,in H ci Represents the value of each gray level, the new 16-dimensional vector is H new ,in H newk Indicates k The sum of the pixel probabilities of the groups is calculated as follows:

[0019] ;

[0020] The newly generated 16-dimensional feature vector is re-normalized and calculated as follows:

[0021] ;

[0022] Among them, the length of the feature vector is 16, and the feature vector before normalization is recorded as H new , the normalized eigenvector is recorded as f , f(x) express f exist x The pixel level value is obtained by the vector H new Get the normalized feature vector f ;

[0023] The third step is to embed the feature vectors generated in the first and second steps into the diffusion probability model as control information;

[0024] The fourth step is to sample a noise image x of size H×W from the standard normal distribution. T , input into the diffusion probability model, and get x T-1 ;

[0025] Step 5: Repeat steps 3 and 4, and replace x in step 4 with T Replace with x T-1 , then the diffusion probability model output x is obtained T-2 , repeat the above process until the final output x0 is obtained.

[0026] Furthermore, in step 4, a resnet-34 deep convolutional neural network is constructed; chromosome expansion data is retrieved in a multi-threaded manner and distributed training of the resnet-34 deep convolutional neural network is performed on multiple GPUs to obtain a chromosome recognition network.

[0027] In a second aspect, the present invention provides a chromosome identification system, comprising:

[0028] A local chromosome image database, used to store chromosome data, wherein the chromosome data includes chromosome images and chromosome categories;

[0029] A diffusion probability model offline training module, used for training the diffusion probability model according to the chromosome data;

[0030] A chromosome data amplification module, which is used to embed the chromosome mask binary image and the chromosome grayscale normalized histogram of the chromosome image and the chromosome category into a feature vector, and use a cross-attention mechanism to apply the feature vector to the diffusion probability model to control the generation result of the diffusion probability model and generate chromosome expansion data;

[0031] A chromosome recognition module, used for training a chromosome recognition network according to the chromosome expansion data;

[0032] The user interaction module is used to identify the chromosome data to be identified according to the chromosome identification network.

[0033] The present invention has the following beneficial effects: a chromosome recognition method and system of the present invention, based on a diffusion probability model amplification data method, not only retains the authenticity of chromosomes, but also successfully converts the image characteristics of chromosomes, effectively expanding the diversity of data. Through this system, the recognition effect of chromosomes on different data sources has been significantly improved, providing a more accurate and reliable auxiliary tool for the field of medical diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] In order to more clearly illustrate the technical solution of the present invention, the drawings required for use in the embodiments are briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0035] Figure 1 A flow chart of the chromosome identification method provided by the present invention;

[0036] Figure 2 This is a schematic diagram of the chromosome data amplification module;

[0037] Figure 3 Schematic diagram for generating chromosome expansion data. DETAILED DESCRIPTION

[0038] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described in conjunction with the specific embodiments of the present invention and the corresponding drawings. 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 are within the scope of protection of the present invention. The technical solutions provided by the embodiments of the present invention are described in detail below in conjunction with the drawings.

[0039] See also Figure 1 The present invention provides a chromosome identification method, comprising:

[0040] Step 1: Acquire chromosome data, which includes chromosome images and chromosome categories.

[0041] Specifically, the data of the local chromosome image database of the present invention is derived from millions of chromosome images that have been analyzed by professional chromosome karyotype analysts, and are enhanced by a chromosome enhancement method to ensure the accuracy and clarity of the chromosome images.

[0042] Step 2: training a diffusion probability model based on the chromosome data.

[0043] Specifically, the core algorithm of the offline training of the diffusion probability model of the present invention is based on the diffusion probability model, and the algorithm gradually generates the outline and details of the target by simulating Gaussian distribution sampling in a random process. During forward diffusion, the model continuously adds noise to the input data and learns the characteristics of the noise until the data completely becomes noise; during reverse diffusion, the noise data learned through forward diffusion is gradually removed to eventually generate an ordered data sample. However, the generation process of the model is random, so it is necessary to adjust the parameters of the model through certain control conditions to control the content generated by the model. Therefore, the present invention proposes an improved diffusion probability model, such as Figure 2 As shown in Figure 2, the diffusion probability model is essentially a random Gaussian noise X T After T times of diffusion, the process to the target image X0, where T is a manually set hyperparameter. Figure 2 The dotted box describes a process of taking two adjacent intermediate images X t To X t-1 In step 2, the diffusion probability model has a diffusion process in which the input image is X t , the output image is X t-1 , processed by a deep convolutional neural network, from X t To X t-1 The recursive formula is as follows:

[0044] ;

[0045] Where t represents the tth step of the diffusion process, α t , β t are artificially set hyperparameters, which respectively represent the proportion of original data information retained at time t and the intensity of noise added at time t. , ,∈ θ (X t , t) is the noise data predicted by the deep convolutional neural network, ∈ is a random sample in the standard Gaussian distribution.

[0046] The loss function of the diffusion probability model is:

[0047] ;

[0048] Where t represents the tth step of the diffusion process, L t represents the loss function value at the t-th step of diffusion, x0 is the target chromosome image during training, ∈ is a random sampling in the standard Gaussian distribution, is a manually set hyperparameter, is the mathematical expectation, ∈ θ It is the result predicted by the neural network.

[0049] Step three, embed the chromosome mask binary image, chromosome grayscale normalized histogram and chromosome category of the chromosome image into the feature vector, and use the cross-attention mechanism to apply the feature vector to the diffusion probability model to control the generation result of the diffusion probability model and generate chromosome expansion data.

[0050] Specifically, in order to limit the generated chromosome categories and contours while changing image characteristics such as image color and clarity, the present invention draws on the feature embedding technology in natural language processing to embed chromosome categories, chromosome mask binary images, and chromosome grayscale normalized histograms into feature vectors, and then uses the cross-attention mechanism to apply the feature vectors to the diffusion model to control the generation results of the model. Among them, the chromosome category is the chromosome number, such as chromosome 1 and chromosome 2; the chromosome contour is the chromosome mask binary image, 0 represents the background, and 1 represents the foreground; the image characteristics of the chromosome are generated by extracting the chromosome grayscale normalized histogram of the chromosome image to generate a feature vector. Finally, the feature vectors generated by the three control conditions are embedded in the diffusion probability model to control the generation effect of the model.

[0051] In natural language processing, cross attention is to calculate the attention on different sequences to process the semantic relationship between sequences. The cross attention mechanism splits the input tensor into two parts and , and then use one part as the query set and the other part as the key set and value set; the output is a tensor of size n×d2, and for each row vector, the attention weights for all row vectors are given;

[0052] Let Q = X1W Q and K=V=X2W K , then the cross attention is calculated as follows:

[0053] ;

[0054] in, , is the learned projection matrix, d K is the dimension of the key set, Q is the query feature vector for calculating the attention weight, K is the index feature vector for calculating the attention weight, V is the content feature vector for calculating the attention weight, and K T is the transpose of K, d1 and d2 are the dimensions of X1 and X2.

[0055] The following is a detailed explanation of the process of generating a chromosome image with specified image characteristics such as color or brightness distribution. Assume that there is a real chromosome 1 C, and the goal is to generate a chromosome with the outline of the real chromosome 1, but the image characteristics such as color or brightness distribution are different from the input C.style The chromosomes have the same color or brightness distribution as chromosome 1. The chromosome expansion data generated in step 3 includes:

[0056] The first step is to extract the mask binary image C of chromosome C mask ; The mask binary image C mask The background is 0, the foreground is 1, and the size is the same as that of chromosome C, which is H×W. mask The category of chromosome C is used as two conditions and embedded into two feature vectors;

[0057] The second step is to extract the specified chromosome C style The chromosome grayscale normalized histogram in is obtained, and the chromosome grayscale normalized histogram is embedded into a feature vector; the calculation formula of the chromosome grayscale normalized histogram is as follows:

[0058] ;

[0059] in, p ( r k ) represents the relative frequency of each gray level in the image. r k Represents the gray level of a pixel, which is a discrete value between 0 and 255. n k Is grayscale r k The number of MN is the total number of pixels in the image, M , N Respectively represent the length and width of the image;

[0060] The feature vector of the chromosome grayscale normalized histogram has 256 dimensions, and each dimension corresponds to a pixel level. Such pixel granularity is too high, which makes it difficult for the model to learn. Therefore, the calculated 256-dimensional feature vector is compressed to 16 dimensions. The calculation process is as follows:

[0061] Assume that the generated 256-dimensional vector is H c ,in H ci Represents the value of each gray level, the new 16-dimensional vector is H new ,in H newk Indicates k The sum of the pixel probabilities of the groups is calculated as follows:

[0062] ;

[0063] The newly generated 16-dimensional feature vector is re-normalized and calculated as follows:

[0064]

[0065] Among them, the length of the feature vector is 16, and the feature vector before normalization is recorded as H new , the normalized eigenvector is recorded as f , f(x) express f exist x The pixel level value is obtained by the vector H new Get the normalized feature vector f ;

[0066] The third step is to embed the feature vectors generated in the first and second steps into the diffusion probability model as control information;

[0067] The fourth step is to sample a noise image x of size H×W from the standard normal distribution. T , input into the diffusion probability model, and get x T-1 ;

[0068] Step 5: Repeat steps 3 and 4, and replace x in step 4 with T Replace with x T-1 , then the diffusion probability model output x is obtained T-2 , repeat the above process until the final output x0 is obtained.

[0069] from Figure 2 As can be seen from the figure, the generated x0 is consistent with the real C in category and shape, but different from C in color and brightness. style The chromosomes are similar. This shows that when generating chromosomes, the present invention not only effectively retains the authenticity of the chromosomes, but also successfully achieves flexible conversion of image characteristics such as chromosome color and brightness. More specifically, Figure 3 In the example diagram of chromosome generation with different image characteristics shown, three different chromosome image characteristics are selected as input. After model reasoning, chromosome images corresponding to these input image characteristics are generated. Although the image characteristics such as color and brightness are different, all generated chromosomes maintain the same classification attributes and contour structure, which strongly proves that the present invention can stably maintain the basic characteristics and authenticity of chromosomes when converting image characteristics such as chromosome color and brightness, and will not affect the core information of chromosomes due to changes in image characteristics such as color and brightness. This generation capability helps to provide rich and diverse real data for chromosome recognition models, improve the generalization ability of the model, and enhance the practical application value of the system.

[0070] Step 4: training a chromosome recognition network based on the chromosome expansion data.

[0071] Specifically, in step 4, a resnet-34 deep convolutional neural network is constructed; chromosome expansion data is retrieved in a multi-threaded manner and distributed training of the resnet-34 deep convolutional neural network is completed on multiple GPUs to obtain a chromosome recognition network.

[0072] Step five: identifying the chromosome data to be identified based on the chromosome identification network.

[0073] Specifically, the user interaction interface of the present invention is used to assist doctors in identifying a model of chromosome categories. For input data, the chromosome recognition network will output chromosome category information to assist doctors in diagnosing a disease.

[0074] The present invention can provide more abundant and diverse training data for chromosome recognition network, which is helpful to improve the generalization of chromosome recognition network. Through the introduction of diffusion probability model, the image characteristics of chromosome can be converted, and chromosome data with different image characteristics such as image color and brightness can be generated, which is helpful to enhance the adaptability and robustness of chromosome recognition network for different data sources. Through the above-mentioned chromosome expansion data, a chromosome recognition network with high generalization and robustness can be trained to accurately identify chromosome data with different sources or different colors, brightness and other image characteristics.

[0075] Based on the above chromosome identification method, the present invention provides a chromosome identification system, comprising:

[0076] A local chromosome image database, used to store chromosome data, wherein the chromosome data includes chromosome images and chromosome categories;

[0077] A diffusion probability model offline training module, used for training the diffusion probability model according to the chromosome data;

[0078] A chromosome data amplification module, which is used to embed the chromosome mask binary image and the chromosome grayscale normalized histogram of the chromosome image and the chromosome category into a feature vector, and use a cross-attention mechanism to apply the feature vector to the diffusion probability model to control the generation result of the diffusion probability model and generate chromosome expansion data;

[0079] A chromosome recognition module, used for training a chromosome recognition network according to the chromosome expansion data;

[0080] The user interaction module is used to identify the chromosome data to be identified according to the chromosome identification network.

[0081] The embodiment of the present invention further provides a storage medium, wherein a computer program is stored in the storage medium, and when the computer program is executed by a processor, some or all of the steps in each embodiment of the chromosome identification method provided by the present invention are implemented. The storage medium may be a magnetic disk, an optical disk, a read-only memory (ROM) or a random access memory (RAM).

[0082] Those skilled in the art can clearly understand that the technology in the embodiments of the present invention can be implemented by means of software plus a necessary general hardware platform. Based on this understanding, the technical solution in the embodiments of the present invention can be essentially or partly contributed to the prior art in the form of a software product, which can be stored in a storage medium such as ROM / RAM, a magnetic disk, an optical disk, etc., and includes several instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in the various embodiments of the present invention or certain parts of the embodiments.

[0083] The above-described embodiments of the present invention do not limit the protection scope of the present invention.

Claims

1. A chromosome identification method, characterized in that: include: Step 1, obtaining chromosome data, wherein the chromosome data includes chromosome images and chromosome categories; Step 2, training a diffusion probability model according to the chromosome data; Step 3: embed the chromosome mask binary image, chromosome grayscale normalized histogram and chromosome category of the chromosome image into a feature vector, and use the cross attention mechanism to apply the feature vector to the diffusion probability model to control the generation result of the diffusion probability model and generate chromosome expansion data; The chromosome expansion data generated in step 3 includes: The first step is to extract the mask binary image C of chromosome C mask ; The mask binary image C mask The background is 0, the foreground is 1, and the size is the same as that of chromosome C, which is H×W. mask The category of chromosome C is used as two conditions and embedded into two feature vectors; The second step is to extract the specified chromosome C style The chromosome grayscale normalized histogram in is obtained, and the chromosome grayscale normalized histogram is embedded into a feature vector; the calculation formula of the chromosome grayscale normalized histogram is as follows: ; in, p ( r k ) represents the relative frequency of each gray level in the image. r k Represents the gray level of a pixel, which is a discrete value between 0 and 255. n k Is grayscale r k The number of MN is the total number of pixels in the image, M , N Respectively represent the length and width of the image; The feature vector of the chromosome grayscale normalized histogram has 256 dimensions. The calculated 256-dimensional feature vector is compressed to 16 dimensions. The calculation process is as follows: Assume that the generated 256-dimensional vector is H c ,in H ci Represents the value of each gray level, the new 16-dimensional vector is H new ,in H newk Indicates k The sum of the pixel probabilities of the groups is calculated as follows: ; The newly generated 16-dimensional feature vector is re-normalized and calculated as follows: ; Among them, the length of the feature vector is 16, and the feature vector before normalization is recorded as H new , the normalized eigenvector is recorded as f , f(x) express f exist x The pixel level value is obtained by the above formula from the vector H new Get the normalized feature vector f ; The third step is to embed the feature vectors generated in the first and second steps into the diffusion probability model as control information; The fourth step is to sample a noise image x of size H×W from the standard normal distribution. T , input into the diffusion probability model, and get x T-1 ; Step 5: Repeat steps 3 and 4, and replace x in step 4 with T Replace with x T-1 , then the diffusion probability model output x is obtained T-2 , repeat the above process until the final output x0 is obtained; Step 4, training a chromosome recognition network according to the chromosome expansion data; Step five: identifying the chromosome data to be identified based on the chromosome identification network.

2. The chromosome identification method according to claim 1, characterized in that: In step 2, the diffusion probability model has an input image X in a diffusion process. t , the output image is X t-1 , processed by a deep convolutional neural network, from X t To X t-1 The recursive formula is as follows: ; Where t represents the tth step of the diffusion process, α t , β t are artificially set hyperparameters, which respectively represent the proportion of original data information retained at time t and the intensity of noise added at time t. , ∈ θ (X t , t) is the noise data predicted by the deep convolutional neural network, ∈ is a random sample in the standard Gaussian distribution.

3. The chromosome identification method according to claim 2, characterized in that: The loss function of the diffusion probability model is: ; Where t represents the tth step of the diffusion process, L t represents the loss function value at the t-th step of diffusion, x0 is the target chromosome image during training, ∈ is a random sampling in the standard Gaussian distribution, is a manually set hyperparameter, is the mathematical expectation, ∈ θ It is the result predicted by the neural network.

4. The chromosome identification method according to claim 3, characterized in that: In step 3, the cross-attention mechanism splits the input tensor into two parts and , and then use one part as the query set and the other part as the key set and value set; the output is a tensor of size n×d2, and for each row vector, the attention weights for all row vectors are given; Let Q = X1W Q and K=V= W K , then the cross attention is calculated as follows: ; in, , is the learned projection matrix, d K is the dimension of the key set, Q is the query feature vector for calculating the attention weight, K is the index feature vector for calculating the attention weight, V is the content feature vector for calculating the attention weight, and K T is the transpose of K, d1 and d2 are the dimensions of X1 and X2.

5. The chromosome identification method according to claim 1, characterized in that: In step 4, a resnet-34 deep convolutional neural network is constructed; chromosome expansion data is retrieved in a multi-threaded manner and distributed training of the resnet-34 deep convolutional neural network is performed on multiple GPUs to obtain a chromosome recognition network.

6. A chromosome identification system, characterized in that: include: A local chromosome image database, used to store chromosome data, wherein the chromosome data includes chromosome images and chromosome categories; A diffusion probability model offline training module, used for training the diffusion probability model according to the chromosome data; A chromosome data amplification module, which is used to embed the chromosome mask binary image and the chromosome grayscale normalized histogram of the chromosome image and the chromosome category into a feature vector, and use a cross-attention mechanism to apply the feature vector to the diffusion probability model to control the generation result of the diffusion probability model and generate chromosome expansion data; Generating chromosome expansion data includes: The first step is to extract the mask binary image C of chromosome C mask ; The mask binary image C mask The background is 0, the foreground is 1, and the size is the same as that of chromosome C, which is H×W. mask The category of chromosome C is used as two conditions and embedded into two feature vectors; The second step is to extract the specified chromosome C style The chromosome grayscale normalized histogram in is obtained, and the chromosome grayscale normalized histogram is embedded into a feature vector; the calculation formula of the chromosome grayscale normalized histogram is as follows: ; in, p ( r k ) represents the relative frequency of each gray level in the image. r k Represents the gray level of a pixel, which is a discrete value between 0 and 255. n k Is grayscale r k The number of MN is the total number of pixels in the image, M , N Respectively represent the length and width of the image; The feature vector of the chromosome grayscale normalized histogram has 256 dimensions. The calculated 256-dimensional feature vector is compressed to 16 dimensions. The calculation process is as follows: Assume that the generated 256-dimensional vector is H c ,in H ci Represents the value of each gray level, the new 16-dimensional vector is H new ,in H newk Indicates k The sum of the pixel probabilities of the groups is calculated as follows: ; The newly generated 16-dimensional feature vector is re-normalized and calculated as follows: ; Among them, the length of the feature vector is 16, and the feature vector before normalization is recorded as H new , the normalized eigenvector is recorded as f , f(x) express f exist x The pixel level value is obtained by the above formula from the vector H new Get the normalized feature vector f ; The third step is to embed the feature vectors generated in the first and second steps into the diffusion probability model as control information; The fourth step is to sample a noise image x of size H×W from the standard normal distribution. T , input into the diffusion probability model, and get x T-1 ; Step 5: Repeat steps 3 and 4, and replace x in step 4 with T Replace with x T-1 , then the diffusion probability model output x is obtained T-2 , repeat the above process until the final output x0 is obtained; A chromosome recognition module, used for training a chromosome recognition network according to the chromosome expansion data; The user interaction module is used to identify the chromosome data to be identified according to the chromosome identification network.

Images