Method and apparatus for acquiring a cell classification model
By training a cell classification model and optimizing feature map weights, the problem of classifying objects with small inter-class differences was solved, achieving fast and accurate automated classification and improving the efficiency and accuracy of white blood cell classification.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- XIN HUA HOSPITAL AFFILIATED TO SHANGHAI JIAO TONG UNIV SCHOOL OF MEDICINE
- Filing Date
- 2020-10-23
- Publication Date
- 2026-06-16
AI Technical Summary
Existing classification methods are difficult to effectively distinguish objects with small differences between them, resulting in low efficiency and accuracy of manual classification. In particular, in the classification of white blood cells, relying on manual microscopic examination leads to a large workload and is prone to errors.
A pre-trained cell classification model is used, and feature maps are extracted and weighted through a convolutional neural network. An iterative algorithm is then used to optimize the weight distribution of the feature maps, thereby improving the feature discrimination ability and obtaining the optimal cell classification model.
It enables rapid and accurate classification of objects with small inter-class differences, improving classification efficiency and accuracy while reducing manual workload and error rate.
Smart Images
Figure CN114495096B_ABST
Abstract
Description
Technical Field
[0001] This article relates to classification techniques, particularly a method and apparatus for obtaining a cell classification model. Background Technology
[0002] Currently, there are many objects with small inter-class distances that require further classification. However, it is difficult to accurately classify objects with small inter-class distances using conventional classification methods or models. Most of these objects are still classified manually. For example, white blood cell classification still relies on manual microscopic examination, which is a large clinical workload, has low efficiency, and is prone to errors due to fatigue, resulting in low accuracy. Summary of the Invention
[0003] This application provides a method and apparatus for obtaining a cell classification model, which can classify targets with small inter-class differences, thereby improving classification efficiency and accuracy.
[0004] This application provides a method for obtaining a cell classification model, the method including:
[0005] Obtain a trained cell classification model;
[0006] Based on the test data and the cell classification model, obtain M feature maps of the target to be classified; M is a positive integer greater than or equal to 1; the test data includes one or more targets to be classified;
[0007] Obtain the weights of the M feature maps;
[0008] The cell classification model is optimized by inputting M weighted feature maps into it, thereby obtaining the optimal cell classification model.
[0009] In an exemplary embodiment of this application, obtaining M feature maps of the target to be classified based on test data and the cell classification model may include:
[0010] The test data is input into the cell classification model, the output of the cell classification model is obtained, and the correct classification result and the incorrect classification result are distinguished from the output result;
[0011] The correct classification result is input into the cell classification model, and the cell classification model extracts features from the correct classification result to obtain the M feature maps.
[0012] In an exemplary embodiment of this application, obtaining the weights of the M feature maps may include:
[0013] Perform feature classification on the M feature maps;
[0014] The weights of the M feature maps are calculated based on the feature classification.
[0015] In an exemplary embodiment of this application, classifying the M feature maps may include:
[0016] Multiple sets of feature maps are selected sequentially from the M feature maps in an iterative manner as cluster centers. Each set of feature maps includes a feature maps, and the a feature maps are used as cluster centers in the current iteration. The b feature maps that are not selected from the M feature maps are used as non-cluster centers; a + b = M, where a and b are both positive integers.
[0017] In each iteration, each feature map is treated as a one-dimensional vector, and the distance between the vector corresponding to each cluster center and the vector corresponding to each non-cluster center is calculated. All distances calculated in this iteration are divided into N distance intervals according to the distance size, and each distance interval is used as a feature classification. N is a positive integer greater than or equal to 2.
[0018] The iteration continues until the distance between any two feature maps is calculated and the distance intervals to which all M feature maps belong are obtained, at which point the iteration stops.
[0019] In an exemplary embodiment of this application, the step of iteratively selecting multiple sets of feature maps as cluster centers from the M feature maps may include:
[0020] In the initial iteration, a feature maps are randomly selected from the M feature maps as cluster centers; and in each iteration, the probability of each feature map in the b feature maps that are not cluster centers being selected as the cluster center in the next iteration is calculated according to a preset probability algorithm; in the next iteration, the a feature maps with the highest probability are selected as cluster centers.
[0021] In an exemplary embodiment of this application, the probability algorithm may include:
[0022] In each iteration, calculate the distance between the vector corresponding to each non-cluster center and the vector corresponding to any i-th non-cluster center b(i) among the b non-cluster centers. Select the minimum distance d from all the calculated distances for the i-th non-cluster center b(i). min (i), and calculate the minimum distance d. min (i) The ratio of the sum of all distances to the i-th non-cluster center b(i) is used as the probability that the i-th non-cluster center b(i) is selected as the cluster center in the next iteration; i is a positive integer, 1≤i≤b.
[0023] In an exemplary embodiment of this application, calculating the weights of the M feature maps based on the feature classification may include:
[0024] Each distance interval is assigned a different weight based on the distance it contains; and the feature maps contained within each distance interval are assigned corresponding weights accordingly.
[0025] The larger the distance value contained in each distance interval, the higher the weight assigned to that distance interval and the higher the weight assigned to the feature map contained in that distance interval; the smaller the distance value contained in each distance interval, the lower the weight assigned to that distance interval and the lower the weight assigned to the feature map contained in that distance interval.
[0026] In an exemplary embodiment of this application, the method may further include: normalizing all weights to a range of 0 to 1.
[0027] In an exemplary embodiment of this application, obtaining the trained cell classification model may include:
[0028] The cell classification model is obtained by training a pre-constructed convolutional neural network (CNN) for classification using preset training data.
[0029] The training data consists of the correct classification results obtained by classifying the target to be classified in advance.
[0030] This application also provides a cell classification model acquisition device, which may include a processor and a computer-readable storage medium. The computer-readable storage medium stores instructions that, when executed by the processor, implement the cell classification model acquisition method described in any one of the above embodiments.
[0031] This application embodiment includes: obtaining a trained cell classification model; obtaining M feature maps of the target to be classified based on test data and the cell classification model; M is a positive integer greater than or equal to 1; the test data includes one or more targets to be classified; obtaining the weights of the M feature maps; inputting the weighted M feature maps into the cell classification model to optimize the cell classification model, thereby obtaining the optimal cell classification model. This embodiment achieves the classification of targets with small inter-class differences, improving classification efficiency and accuracy.
[0032] Other features and advantages of this application will be set forth in the following description, and will be apparent in part from the description, or may be learned by practicing the application. Other advantages of this application can be realized and obtained by means of the solutions described in the description and the accompanying drawings. Attached Figure Description
[0033] The accompanying drawings are used to provide an understanding of the technical solutions of this application and constitute a part of the specification. They are used together with the embodiments of this application to explain the technical solutions of this application and do not constitute a limitation on the technical solutions of this application.
[0034] Figure 1 This is a flowchart illustrating the method for obtaining a cell classification model according to an embodiment of this application;
[0035] Figure 2 This is a schematic diagram of a method for obtaining a cell classification model according to an embodiment of this application;
[0036] Figure 3 This is a block diagram of the apparatus for acquiring a cell classification model according to an embodiment of this application. Detailed Implementation
[0037] This application describes several embodiments, but these descriptions are exemplary and not restrictive, and it will be apparent to those skilled in the art that many more embodiments and implementations are possible within the scope of the embodiments described herein. Although many possible combinations of features are shown in the drawings and discussed in the detailed description, many other combinations of the disclosed features are also possible. Unless specifically limited, any feature or element of any embodiment may be used in combination with, or may replace, any feature or element of any other embodiment.
[0038] This application includes and contemplates combinations of features and elements known to those skilled in the art. The embodiments, features, and elements disclosed in this application may also be combined with any conventional features or elements to form a unique inventive scheme as defined by the claims. Any feature or element of any embodiment may also be combined with features or elements from other inventive schemes to form another unique inventive scheme as defined by the claims. Therefore, it should be understood that any feature shown and / or discussed in this application may be implemented individually or in any suitable combination. Therefore, the embodiments are not limited except by the limitations imposed by the appended claims and their equivalents. Furthermore, various modifications and changes may be made within the scope of the appended claims.
[0039] Furthermore, in describing representative embodiments, the specification may have presented methods and / or processes as a specific sequence of steps. However, the method or process should not be limited to the specific order of steps described herein, to the extent that it does not depend on such a specific order. As will be understood by those skilled in the art, other sequences of steps are also possible. Therefore, the specific order of steps set forth in the specification should not be construed as a limitation of the claims. Moreover, the claims concerning the method and / or process should not be limited to the steps performed in the written order, and those skilled in the art will readily understand that these orders can be varied and still remain within the spirit and scope of the embodiments of this application.
[0040] This application provides a method for obtaining a cell classification model, such as... Figure 1 , Figure 2 As shown, the method may include steps S101-S104:
[0041] S101. Obtain the trained cell classification model;
[0042] S102. Obtain M feature maps of the target to be classified based on the test data and the cell classification model; M is a positive integer greater than or equal to 1; the test data includes one or more targets to be classified;
[0043] S103. Obtain the weights of the M feature maps;
[0044] S104. Input the M weighted feature maps into the cell classification model to optimize the cell classification model and obtain the optimal cell classification model.
[0045] In an exemplary embodiment of this application, white blood cell classification is a medical detection method that involves classifying and counting white blood cells during blood microscopy. There are five types of white blood cells in blood: neutrophils, eosinophils, basophils, lymphocytes, and monocytes. In bone marrow, there are more than 40 different types, each with its own physiological function. White blood cell classification is an important indicator in environmental medicine research. However, the differences between various white blood cell types are very small and difficult to distinguish. Currently, white blood cell classification relies on manual microscopic examination, which is labor-intensive, inefficient, and prone to errors due to fatigue. The optimal model of this application embodiment can quickly and accurately extract various cell types, effectively improving work efficiency and accuracy.
[0046] In an exemplary embodiment of this application, when the optimal cell classification model of this application embodiment is used for cell classification, the test data can be one or more cell images to be classified, and the feature map can be one or more feature images of various cell types.
[0047] In an exemplary embodiment of this application, the optimal cell classification model of this application embodiment can also be used for classifying car brands, classifying different categories of the same animal or plant species (such as sheep) (such as goats, sheep, antelopes, etc.), and classifying any objects with small differences between categories.
[0048] In an exemplary embodiment of this application, obtaining the trained cell classification model may include:
[0049] The cell classification model is obtained by training a pre-constructed convolutional neural network (CNN) for classification using preset training data.
[0050] The training data consists of the correct classification results obtained by classifying the target to be classified in advance.
[0051] In the exemplary embodiments of this application, Convolutional Neural Networks (CNNs) are a type of feedforward neural network that includes convolutional computation and has a deep structure, and are one of the representative algorithms of deep learning. A CNN can be composed of structures such as convolutional layers and pooling layers.
[0052] In an exemplary embodiment of this application, for cell classification, the training data can be pre-classified images of various cells.
[0053] In the exemplary embodiments of this application, the cell classification model described above can accurately classify targets with large inter-class differences (such as bicycles and cars). However, it cannot meet the requirements for classifying objects with small inter-class differences, such as various cells, cars of various brands, and sheep of different breeds. Therefore, the cell classification model can be further optimized to obtain the optimal cell classification model.
[0054] In an exemplary embodiment of this application, in order to obtain the optimal cell classification model, features of each type of object in the target body to be classified can be extracted by a convolutional neural network (CNN), and feature weights are set for the features extracted by the CNN. The role of these feature weights is to improve the speed and accuracy of feature selection. The features that can significantly distinguish the different types of the target body to be classified have larger weights, while the features that are more common and cannot significantly distinguish the different types of the target body to be classified have smaller weights.
[0055] In the exemplary embodiments of this application, the above steps S102-S104 are cell classification model optimization steps, and each optimization step will be described in detail below.
[0056] In an exemplary embodiment of this application, obtaining M feature maps of the target to be classified based on test data and the cell classification model may include:
[0057] The test data is input into the cell classification model, the output of the cell classification model is obtained, and the correct classification result and the incorrect classification result are distinguished from the output result;
[0058] The correct classification result is input into the cell classification model, and the cell classification model extracts features from the correct classification result to obtain the M feature maps.
[0059] In an exemplary embodiment of this application, the number of feature maps M obtained can be determined based on the different types of target objects to be classified and the number of distinguishing features of different objects. The specific value of M is not limited here. For example, for white blood cell classification, M can include several thousand types.
[0060] In an exemplary embodiment of this application, obtaining the weights of the M feature maps may include:
[0061] Perform feature classification on the M feature maps;
[0062] The weights of the M feature maps are calculated based on the N1+N2 class feature maps divided by the feature classification.
[0063] In an exemplary embodiment of this application, feature classification of the M feature maps can divide them into N1+N2 categories of feature maps; wherein, N1 category refers to N1 target feature categories that can distinguish different types of the target to be classified, and N2 category refers to N2 non-target feature categories that are commonly contained in different types of the target to be classified, and that cannot or are difficult to distinguish between different types of the target to be classified; N1 and N2 can both be positive integers greater than or equal to 1.
[0064] In an exemplary embodiment of this application, a higher weight can be assigned to the N1 class of target features, and a lower weight can be assigned to the N2 class of non-target features. Furthermore, different weights can be assigned based on the salience (the characteristic used to distinguish different classes; higher salience corresponds to higher weight, and lower salience corresponds to lower weight) of the different target and non-target feature classes, where higher salience corresponds to higher weight, and lower salience corresponds to lower weight.
[0065] In an exemplary embodiment of this application, feature classification of the M feature maps can be performed using either an unsupervised classification method or a supervised classification method, and no specific implementation method is limited.
[0066] In the exemplary embodiments of this application, the embodiments of the unsupervised classification method are described in detail below.
[0067] In an exemplary embodiment of this application, classifying the M feature maps may include:
[0068] Multiple sets of feature maps are selected sequentially from the M feature maps in an iterative manner as cluster centers. Each set of feature maps includes a feature maps, and the a feature maps are used as cluster centers in the current iteration. The b feature maps that are not selected from the M feature maps are used as non-cluster centers; a + b = M, where a and b are both positive integers.
[0069] In each iteration, each feature map is treated as a one-dimensional vector, and the distance between the vector corresponding to each cluster center and the vector corresponding to each non-cluster center is calculated. All distances calculated in this iteration are divided into N distance intervals according to the distance size, and each distance interval is used as a feature classification. N is a positive integer greater than or equal to 2.
[0070] The iteration continues until the distance between any two feature maps is calculated and the distance intervals to which all M feature maps belong are obtained, at which point the iteration stops.
[0071] In an exemplary embodiment of this application, the value of 'a' can be defined according to different classification accuracies and actual needs.
[0072] In an exemplary embodiment of this application, the step of iteratively selecting multiple sets of feature maps as cluster centers from the M feature maps may include:
[0073] In the initial iteration, a feature maps are randomly selected from the M feature maps as cluster centers; and in each iteration, the probability of each feature map in the b feature maps that are not cluster centers being selected as the cluster center in the next iteration is calculated according to a preset probability algorithm; in the next iteration, the a feature maps with the highest probability are selected as cluster centers.
[0074] In an exemplary embodiment of this application, the probability algorithm may include:
[0075] In each iteration, calculate the distance between the vector corresponding to each non-cluster center and the vector corresponding to any i-th non-cluster center b(i) among the b non-cluster centers. Select the minimum distance d from all the calculated distances for the i-th non-cluster center b(i). min (i), and calculate the minimum distance d. min (i) The ratio of the sum of all distances to the i-th non-cluster center b(i) is used as the probability that the i-th non-cluster center b(i) is selected as the cluster center in the next iteration; i is a positive integer, 1≤i≤b.
[0076] In an exemplary embodiment of this application, a cluster centers C1, C2, ..., Ca can be selected from M feature maps, and the distance between each cluster center C1, C2, ..., Ca and each of the b non-cluster centers can be calculated. For example, the distances between the cluster centers C1, C2, ..., Ca and the i-th non-cluster center b(i) can be denoted as d[b(i), C1], d[b(i), C2], ..., d[b(i), Ca], respectively. The minimum value among d[b(i), C1], d[b(i), C2], ..., d[b(i), Ca] can be denoted as d min (i), that is: dmin(i)=min{d[b(i), C1], d[b(i), C2], ..., d[b(i), Ca]}.
[0077] In an exemplary embodiment of this application, in this iteration, the minimum distance d between the i-th non-cluster center b(i) and the cluster centers C1, C2, ..., Ca is obtained. min After (i), the minimum distance d can be used as a basis. min (i) Calculate the probability P that the i-th non-cluster center b(i) is selected as the cluster center in the next iteration:
[0078]
[0079] In an exemplary embodiment of this application, a higher P value indicates a higher probability that the i-th non-cluster center b(i) will be selected as the cluster center in the next iteration; conversely, a lower P value indicates a lower probability that the i-th non-cluster center b(i) will be selected as the cluster center in the next iteration. After calculating the probability P of each non-cluster center, a non-cluster centers with the highest probability P can be selected from the calculated probabilities P as the cluster centers for the next iteration.
[0080] In an exemplary embodiment of this application, after obtaining a cluster centers in each iteration, the distance between the a cluster centers and each non-cluster center can be calculated in each iteration, thereby allowing the distance between any two feature maps to be calculated.
[0081] In an exemplary embodiment of this application, the similarity between two feature maps (each feature map can represent a distinguishing feature) can be determined by calculating the distance between any two feature maps, that is, the similarity between two distinguishing features can be obtained; wherein, the smaller the similarity, the better the two feature maps can distinguish the different types of the target to be classified, and the larger the similarity, the lower the ability of the two feature maps to distinguish the different types of the target to be classified.
[0082] In an exemplary embodiment of this application, the calculated distances can be divided into different distance intervals according to their magnitude. Each distance interval contains one or more distances of a certain magnitude, and the magnitudes of the distances contained in different distance intervals are different.
[0083] In an exemplary embodiment of this application, different distance intervals represent sets of feature maps with different saliency; the distance intervals containing larger distance values are sets of feature maps with higher saliency, and the distance intervals containing smaller distance values are sets of feature maps with lower saliency.
[0084] In an exemplary embodiment of this application, calculating the weights of the M feature maps based on the feature classification may include:
[0085] Each distance interval is assigned a different weight based on the distance it contains; and the feature maps contained within each distance interval are assigned corresponding weights accordingly.
[0086] The larger the distance value contained in each distance interval, the higher the weight assigned to that distance interval and the higher the weight assigned to the feature map contained in that distance interval; the smaller the distance value contained in each distance interval, the lower the weight assigned to that distance interval and the lower the weight assigned to the feature map contained in that distance interval.
[0087] In an exemplary embodiment of this application, since the distance intervals containing large distance values are a set of feature maps with high saliency, these distance intervals can be assigned higher weights to increase the importance of feature maps within these distance intervals in feature differentiation, thereby improving classification speed and accuracy. Conversely, since the distance intervals containing small distance values are a set of feature maps with low saliency, these distance intervals can be assigned lower weights to reduce the importance of feature maps within these distance intervals in feature differentiation, thereby reducing the influence of feature maps that are not conducive to class differentiation in the classification process.
[0088] In an exemplary embodiment of this application, different weights are set for different distance intervals, and corresponding weights are assigned to the feature maps contained within the distance intervals. For multiple feature maps contained in the same distance interval, they can all be assigned the same weight, or different weights can be assigned according to the distance between different feature maps.
[0089] In the exemplary embodiments of this application, the specific weight values can be defined according to different needs, and no specific restrictions are imposed here.
[0090] In an exemplary embodiment of this application, the method may further include: normalizing all weights to a range of 0 to 1.
[0091] In an exemplary embodiment of this application, after obtaining the weights of all M feature maps, a weight vector with M dimensions can be obtained. The weights of each dimension in the obtained weight vector can be summed and normalized to between 0 and 1.
[0092] In an exemplary embodiment of this application, the normalization method may include: performing the following calculations for each dimension weight: Where Xj refers to the weight of the j-th dimension, The sum of all weights is j, where j is a positive integer between 1 and M.
[0093] In an exemplary embodiment of this application, after obtaining the weight of each feature map, the M feature maps with weights can be input into the cell classification model as the basis for the cell classification model to perform classification, thereby obtaining the optimal cell classification model.
[0094] In an exemplary embodiment of this application, data (such as images) of the target body to be classified (such as cells, cars, etc.) are input into the optimal cell classification model, thereby enabling the target body to be classified to be classified quickly and accurately.
[0095] This application embodiment also provides a cell classification model acquisition device 1, such as... Figure 3 As shown, it may include a processor 11 and a computer-readable storage medium 12, wherein the computer-readable storage medium 12 stores instructions that, when executed by the processor 11, implement the method for obtaining the cell classification model as described above.
[0096] In the exemplary embodiments of this application, any of the embodiments in the method embodiments described above are applicable to the device embodiments, and will not be described in detail here.
[0097] It will be understood by those skilled in the art that all or some of the steps, systems, or apparatuses disclosed above, and their functional modules / units, can be implemented as software, firmware, hardware, or suitable combinations thereof. In hardware implementations, the division between functional modules / units mentioned above does not necessarily correspond to the division of physical components; for example, a physical component may have multiple functions, or a function or step may be performed collaboratively by several physical components. Some or all components may be implemented as software executed by a processor, such as a digital signal processor or microprocessor, or as hardware, or as an integrated circuit, such as an application-specific integrated circuit (ASIC). Such software may be distributed on a computer-readable medium, which may include computer storage media (or non-transitory media) and communication media (or transient media). As is known to those skilled in the art, the term computer storage media includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information (such as computer-readable instructions, data structures, program modules, or other data). Computer storage media include, but are not limited to, RAM, ROM, EEPROM, flash memory or other memory technologies, CD-ROM, digital versatile disc (DVD) or other optical disc storage, magnetic cartridges, magnetic tape, disk storage or other magnetic storage devices, or any other medium that can be used to store desired information and can be accessed by a computer. Furthermore, it is well known to those skilled in the art that communication media typically contain computer-readable instructions, data structures, program modules, or other data in modulated data signals such as carrier waves or other transmission mechanisms, and may include any information delivery medium.
Claims
1. A method for obtaining a cell classification model, characterized in that, The method includes: Obtain a trained cell classification model; Based on the test data and the cell classification model, obtain M feature maps of the target to be classified; M is a positive integer greater than or equal to 1; the test data includes one or more targets to be classified; Obtain the weights of the M feature maps; M weighted feature maps are input into the cell classification model to optimize the cell classification model and obtain the optimal cell classification model. The weights for obtaining the M feature maps include: Perform feature classification on the M feature maps; The weights of the M feature maps are calculated based on the feature classification. The classification of the M feature maps includes: Multiple sets of feature maps are selected sequentially from the M feature maps in an iterative manner as cluster centers. Each set of feature maps includes a feature maps, and the a feature maps are respectively used as cluster centers in the current iteration. The b feature maps that are not selected from the M feature maps are used as non-cluster centers; a + b = M, where a and b are both positive integers. In each iteration, each feature map is treated as a one-dimensional vector, and the distance between the vector corresponding to each cluster center and the vector corresponding to each non-cluster center is calculated. All distances calculated in this iteration are divided into N distance intervals according to the distance size, and each distance interval is used as a feature classification. N is a positive integer greater than or equal to 2. The iteration continues until the distance between any two feature maps is calculated and the distance intervals to which all M feature maps belong are obtained, at which point the iteration stops.
2. The method for obtaining a cell classification model according to claim 1, characterized in that, The process of obtaining M feature maps of the target to be classified based on the test data and the cell classification model includes: The test data is input into the cell classification model, the output of the cell classification model is obtained, and the correct classification result and the incorrect classification result are distinguished from the output result; The correct classification result is input into the cell classification model, and the cell classification model extracts features from the correct classification result to obtain the M feature maps.
3. The method for obtaining a cell classification model according to claim 1, characterized in that, The step of iteratively selecting multiple sets of feature maps as cluster centers from the M feature maps includes: In the initial iteration, a feature maps are randomly selected from the M feature maps as cluster centers; and in each iteration, the probability of each feature map in the b feature maps that are not cluster centers being selected as the cluster center in the next iteration is calculated according to a preset probability algorithm; in the next iteration, the a feature maps with the highest probability are selected as cluster centers.
4. The method for obtaining a cell classification model according to claim 3, characterized in that, The probability algorithm includes: In each iteration, calculate the distance between the vector corresponding to each non-cluster center and the vector corresponding to any i-th non-cluster center b(i) among the b non-cluster centers. Select the minimum distance d from all the calculated distances for the i-th non-cluster center b(i). min (i), and calculate the minimum distance d. min (i) The ratio of the sum of all distances to the i-th non-cluster center b(i) is used as the probability that the i-th non-cluster center b(i) is selected as the cluster center in the next iteration; i is a positive integer, 1≤i≤b.
5. The method for obtaining a cell classification model according to claim 1, characterized in that, The step of calculating the weights of the M feature maps based on the feature classification includes: Each distance interval is assigned a different weight based on the distance it contains; and the feature maps contained within each distance interval are assigned corresponding weights accordingly. The larger the distance value contained in each distance interval, the higher the weight assigned to that distance interval and the higher the weight assigned to the feature map contained in that distance interval; the smaller the distance value contained in each distance interval, the lower the weight assigned to that distance interval and the lower the weight assigned to the feature map contained in that distance interval.
6. The method for obtaining a cell classification model according to any one of claims 1-5, characterized in that, The method also includes: normalizing all weights to between 0 and 1.
7. The method for obtaining a cell classification model according to any one of claims 1-5, characterized in that, The process of obtaining the trained cell classification model includes: The cell classification model is obtained by training a pre-constructed convolutional neural network (CNN) for classification using preset training data. The training data consists of the correct classification results obtained by classifying the target to be classified in advance.
8. A device for acquiring a cell classification model, comprising a processor and a computer-readable storage medium, wherein the computer-readable storage medium stores instructions, characterized in that, When the instruction is executed by the processor, the method for obtaining the cell classification model as described in any one of claims 1-7 is implemented.