A sample detection device and its particle classification method

By combining a sample detection device with a target detection model and a classification model, white blood cells are automatically identified and classified, solving the problems of time-consuming, labor-intensive, and experience-dependent traditional methods, and achieving efficient and accurate white blood cell detection.

CN123065918APending Publication Date: 2026-11-10SHENZHEN DYMIND BIOTECH
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202510592325.1
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2025-05-08
Publication Date
2026-11-10

AI Technical Summary

Technical Problem

Traditional white blood cell testing methods are time-consuming and labor-intensive, rely on doctors' experience, are easily affected by subjective factors, and are inefficient.

Method used

A sample detection device is used to identify particle images by combining a target detection model and a classification model. The preliminary classification results and location information are obtained through the preset target detection model, and the classification model is used to further identify the region image, so as to obtain accurate particle classification results.

Benefits of technology

It improves the accuracy and efficiency of white blood cell classification, reduces the need for manual intervention, and saves labor costs.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123065918A_ABST
    Figure CN123065918A_ABST
Patent Text Reader

Abstract

This application discloses a sample detection device. The device includes a detection component, an image acquisition component, and a processor. The image acquisition component is disposed on one side of the detection component and is used to acquire images of the sample to be tested, obtaining particle images, which include multiple types of particles. The processor is connected to the image acquisition component and is used to: identify the multiple particles in the particle image using a preset target detection model to obtain preliminary classification results and position information of the multiple particles; obtain region images of at least some particles based on the position information of the multiple particles; identify the region images of at least some particles using a preset classification model to obtain a second classification result of at least some particles; and obtain a final classification result of the multiple particles based on the preliminary classification result and / or the second classification result. The sample detection device of this application can reliably classify particles, improving the accuracy of particle classification.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This application relates to the field of in vitro diagnostic technology, and in particular to a sample detection device and a particle classification method thereof. Background Technology

[0002] White blood cells in the blood are the most important immune cells in the human body, playing a vital role in maintaining the body's immune function. There are five types of white blood cells in the blood: neutrophils, eosinophils, basophils, lymphocytes, and monocytes. The relative changes in the number of each type of white blood cell have become an important basis for doctors to diagnose diseases.

[0003] Traditional white blood cell testing requires doctors to observe the specimen slide under a microscope. This method is not only time-consuming and labor-intensive, but also inefficient. In addition, it relies to some extent on the doctor's experience and is easily affected by subjective experience and human factors. Summary of the Invention

[0004] This application provides a sample detection device and a particle classification method thereof to solve the technical problem that in the prior art, the detection of white blood cells requires doctors to observe the specimen slide under a microscope. This method is not only time-consuming and labor-intensive, but also inefficient. In addition, it depends to a certain extent on the doctor's experience and is easily affected by subjective experience and human factors.

[0005] To address the aforementioned technical problems, this application provides a sample detection device comprising: a detection component, an image acquisition component, and a processor. The detection component receives a sample to be tested; the image acquisition component is disposed on one side of the detection component and is used to acquire an image of the sample to be tested, obtaining a particle image containing multiple particles; the processor is connected to the image acquisition component and is used to: identify the multiple particles in the particle image using a preset target detection model to obtain preliminary classification results and position information of the multiple particles; obtain region images of at least some particles based on the position information of the multiple particles; identify the region images of at least some particles using a preset classification model to obtain a second classification result of at least some particles; and obtain a final classification result of the multiple particles based on the preliminary classification result and / or the second classification result.

[0006] In some specific embodiments, after obtaining the preliminary classification results of multiple particles, the processor is further configured to: obtain the confidence level of the preliminary classification result of each particle in the multiple particles; and obtain the particles in the multiple particles whose confidence level of the preliminary classification result is less than a preset value, so as to obtain at least some particles.

[0007] In some specific embodiments, the processor is further configured to: when it is confirmed that the confidence level of the preliminary classification result of at least one particle among multiple particles is greater than or equal to a preset value, then take the preliminary classification result of at least one particle as the final classification result of at least one particle.

[0008] In some specific embodiments, the processor is further configured to: obtain the confidence levels of the preliminary classification results of multiple particles and the confidence levels of the second classification results of at least some particles; and obtain the final classification results of multiple particles based on the confidence levels of the preliminary classification results of multiple particles and the confidence levels of the second classification results of at least some particles.

[0009] In some specific embodiments, the processor is further configured to: acquire a training dataset of a preset classification model; augment the training data of some / all types of particles in the training dataset using image processing methods and / or generative artificial intelligence models to obtain an augmented training dataset; and train the preset classification model using the augmented training dataset.

[0010] In some specific embodiments, the processor is also configured to: augment the training data of one or more particles with the smallest number in the training dataset using image processing methods and / or generative artificial intelligence models to obtain an augmented training dataset.

[0011] In some specific embodiments, the particle image is a white blood cell image, and various particles include: neutrophils, eosinophils, basophils, monocytes and lymphocytes. The processor is also used to: amplify the training data of basophils in the training dataset by 8-12 times, the training data of eosinophils by 4-8 times, and the training data of monocytes by 2-5 times.

[0012] In some specific embodiments, the processor is also used to: adjust the size of images in the training dataset and the augmented training dataset by padding the images with zeros around them or cropping them when augmenting the training dataset and when training a preset classification model.

[0013] In some specific embodiments, the preset classification model includes a residual neural network model; the preset target detection model includes: YOLO series models, fast region convolutional neural network models, or single-shot multi-box detector models.

[0014] To address the aforementioned technical problems, this application provides a particle classification method for a sample detection device. Based on any of the above embodiments, the particle classification method includes: acquiring a particle image, which includes multiple particles; identifying the multiple particles in the particle image using a preset target detection model to obtain preliminary classification results and position information of the multiple particles; obtaining region images of at least some particles based on the position information of the multiple particles; identifying the region images of the particles using a preset classification model to obtain a second classification result of at least some particles; and obtaining a final classification result of the multiple particles based on the preliminary classification result and the second classification result.

[0015] The beneficial effects of this application are as follows: Unlike existing technologies, the sample detection device provided in this application includes a detection component, an image acquisition component, and a processor. The detection component receives the sample to be tested; the image acquisition component is disposed on one side of the detection component and is used to acquire images of the sample to obtain particle images, which include multiple particles; the processor is connected to the image acquisition component and is used to: identify multiple particles in the particle image using a preset target detection model to obtain preliminary classification results and position information of multiple particles; obtain region images of at least some particles based on the position information of multiple particles; identify the region images of at least some particles using a preset classification model to obtain a second classification result of at least some particles; and obtain the final classification result of multiple particles based on the preliminary classification result and / or the second classification result. The sample detection device of this application can identify and classify particles in the sample to be tested, with high detection efficiency. Moreover, the detection process combines a preset target detection model and a preset classification model, does not rely on manual labor, and can reliably classify particles, improving the accuracy of particle classification. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a schematic diagram of the framework of an embodiment of a sample detection device provided in this application;

[0018] Figure 2 This is a schematic flowchart of an embodiment of a particle classification method for a sample detection device provided in this application;

[0019] Figure 3This is a schematic flowchart of another embodiment of a particle classification method for a sample detection device provided in this application;

[0020] Figure 4 This is a schematic flowchart of another embodiment of a particle classification method for a sample detection device provided in this application;

[0021] Figure 5 yes Figure 2 A flowchart illustrating an embodiment of the training method for the preset classification model in step S23. Detailed Implementation

[0022] To make the above-mentioned objectives, features, and advantages of this application more apparent and understandable, the specific embodiments of this application will be described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the scope of this application. Furthermore, it should be noted that, for ease of description, only the parts relevant to this application are shown in the accompanying drawings, not the entire structure. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without inventive effort are within the scope of protection of this application.

[0023] The terms "first," "second," etc., used in this application are used to distinguish different objects, not to describe a specific order. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or apparatus that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products, or apparatuses.

[0024] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0025] This application provides a sample detection device capable of classifying and detecting particles in a sample with high accuracy. The sample can be a blood sample, urine sample, or sweat sample, etc. For example, the device can identify and classify white blood cells in a blood sample. The following will use white blood cell classification as an example to provide a detailed description of the sample detection device and its particle classification method provided in this application.

[0026] Please see Figure 1 As shown, Figure 1 This is a schematic diagram of an embodiment of a sample detection device provided in this application. Specifically, the sample detection device 10 includes a detection component 11, an image acquisition component 12, and a processor 13.

[0027] The detection component 11 is used to receive the sample to be tested. When the sample detection device 10 is used to classify white blood cells in blood, the detection component 11 is used to receive the blood sample. To more accurately distinguish different types of white blood cells, the sample to be tested can be a stained blood sample. The detection component 11 may include a sample receiving position, on which the sample to be tested is placed.

[0028] The image acquisition component 12 is disposed on one side of the detection component 11 and is used to acquire images of the sample to be tested to obtain a particle image, which includes various particles. When the particle image is a white blood cell image, the various particles include: neutrophils, eosinophils, basophils, monocytes, and lymphocytes.

[0029] Processor 13 is connected to image acquisition component 12. Processor 13 executes a computer program to implement the particle classification method of the sample detection apparatus according to any of the following embodiments. Processor 13 can be an integrated circuit chip with signal processing capabilities. Processor 13 can also be a general-purpose processor, digital signal processor, application-specific integrated circuit, off-the-shelf programmable gate array or other programmable logic device, discrete gate or transistor logic device, or discrete hardware component. The general-purpose processor can be a microprocessor or any conventional processor, etc. The particle classification method of the sample detection apparatus provided in this application will be described in detail below.

[0030] Please see Figure 2 As shown, Figure 2 This is a flowchart illustrating an embodiment of a particle classification method for a sample detection device provided in this application. Specifically, the processor 13 can be used to:

[0031] S21: The pre-set target detection model is used to identify multiple particles in the particle image to obtain preliminary classification results and position information of multiple particles.

[0032] The acquired particle image is input into a preset target detection model to perform preliminary identification of various particles in the image. This yields preliminary classification results and positional information for the particles. The positional information refers to the predicted location of each particle.

[0033] The preset object detection model may include: YOLO (You Only Look Once) series models, fast region convolutional neural network models, or single-shot multi-box detector models. For example, the YOLOv5 object detection model in AI (Artificial Intelligence) computer vision can be used. In other embodiments, semantic segmentation / instance segmentation models, etc., can also be used.

[0034] In one specific embodiment, when the particle image is a white blood cell image, various particles include: neutrophils, eosinophils, basophils, monocytes, and lymphocytes. The trained AI recognition model YOLO is used to make a preliminary prediction of the particle classification in the particle image, and the resulting confusion matrix is ​​shown in Table 1.

[0035] Table 1. Confusion matrix obtained based on the target detection model.

[0036]

[0037] The data in Table 1 represent recall rates; a high recall rate indicates a low false negative rate. The recall rates for basophils and monocytes in the table are poor. Recall rate indicates the proportion of samples that were actually positive but were correctly predicted.

[0038] As can be seen from Table 1, the YOLO model is not effective in classifying basophils and monocytes, especially for basophils, where the recall rate is 0.

[0039] Therefore, the preliminary classification results of some particles using the preset target detection model may be inaccurate.

[0040] S22: Based on the position information of multiple particles, obtain a region image of at least some particles.

[0041] The processor can obtain a region image of a particle based on its position information. When the particle image is a white blood cell image, the region image refers to an image of a single cell region. In other words, the processor extracts an image of a single cell region based on the particle's position information.

[0042] The processor can obtain region images for each type of particle, or it can obtain region images for only a subset of particles. For example, in the above embodiment, the accuracy of the preliminary prediction results for monocytes and basophils is low, so at least some particles here include monocytes and basophils.

[0043] S23: Identify the region images of at least some particles using a preset classification model to obtain a second classification result for at least some particles.

[0044] The region images of at least some particles obtained in step S22 are input into a preset classification model for recognition to obtain a second classification result for at least some particles.

[0045] For example, in the above embodiments, although the preset target detection model is not very accurate in classifying basophils and monocytes in white blood cells, it has a strong cell detection capability. This application adds a classification model after the target detection model. The region images of particles obtained based on the preset target detection model are input into the preset classification model to continue recognizing the region images of particles, thereby obtaining a second classification result for at least some particles.

[0046] Preferably, the preset classification model may include a ResNet (Residual Neural Network) model. In other embodiments, other neural network models may also be selected as the preset classification model.

[0047] S24: Based on the preliminary classification results and / or the second classification results, the final classification results of various particles are obtained.

[0048] After obtaining the preliminary classification results and the second classification results, the final classification results of various particles can be obtained based on the preliminary classification results and / or the second classification results.

[0049] In the above embodiments, particles are initially classified using a preset target detection model, and then the region image of a single particle is identified using a preset classification model to classify the particles again. Based on the above two classification results, the final classification result of the particles is obtained. This method can significantly improve the accuracy of particle classification.

[0050] In one specific embodiment, the object detection model and the classification model were trained separately and validated on the dataset. The resulting confusion matrix is ​​shown in Table 2. It can be seen that classifying particles using the classification model solves the problem that the general method almost cannot correctly classify basophils, increasing the recall rate from 0 to 42.4%. At the same time, the recall rate for the other four cell types is also significantly improved, especially for monocytes and lymphocytes.

[0051] Table 2. Confusion matrix obtained based on classification model.

[0052]

[0053] Please see Figure 3 As shown, Figure 3 This is a flowchart illustrating another embodiment of a particle classification method for a sample detection device provided in this application. Specifically, the particle classification method may include:

[0054] S31: The pre-set target detection model is used to identify multiple particles in the particle image to obtain preliminary classification results and position information of multiple particles.

[0055] Step S31 is the same as step S21, and will not be repeated here.

[0056] S32: Obtain the confidence level of the preliminary classification results for each particle in a variety of particles.

[0057] After obtaining preliminary classification results for multiple particles, the confidence level of the preliminary classification result for each particle is calculated.

[0058] Confidence score is a quantitative metric that measures how certain a model is about a given prediction. Optionally, the confidence score of the initial classification result for each particle can be obtained based on a deep learning model. In other embodiments, uncertainty can also be estimated using ensemble or Bayesian methods and then used as the confidence score.

[0059] S33: When it is confirmed that the confidence level of the preliminary classification result of at least one particle among multiple particles is greater than or equal to a preset value, the preliminary classification result of at least one particle shall be taken as the final classification result of at least one particle.

[0060] When the confidence level of the preliminary classification result of a particle is greater than or equal to a preset value, it can be considered as a high confidence level. In this case, the preliminary classification result of the particle can be used as the final classification result of the particle. For example, if the confidence level of the preliminary classification result of some particles is greater than or equal to the preset value, then the preliminary classification result of those particles can be used as the final classification result of those particles.

[0061] The preset value can be greater than 0.85, for example, it can be 0.9. When the confidence level of the preliminary classification result of a particle is greater than or equal to 0.9, the preliminary classification result of the particle is taken as the final classification result of the particle.

[0062] S34: Obtain particles from a variety of particles whose confidence level in the preliminary classification result is less than a preset value, so as to obtain at least a subset of particles.

[0063] If, among a variety of particles, the confidence level of a particle's preliminary classification result is lower than a preset value, then the preliminary classification result for that particle is considered to have low confidence. Particles with preliminary classification results having a confidence level lower than the preset value are collected to obtain at least a subset of particles.

[0064] S35: Based on the position information of multiple particles, obtain a region image of at least some particles.

[0065] Step S35 is the same as step S22, and will not be repeated here.

[0066] S36: Recognize the region image of at least some particles using a preset classification model to obtain a second classification result for at least some particles, and use the second classification result of at least some particles as the final classification result for at least some particles.

[0067] The step of "identifying the region images of at least some particles using a preset classification model to obtain the second classification result of at least some particles" in this step is the same as step S23, and will not be repeated here.

[0068] After obtaining the second classification result for at least some particles, this second classification result is directly used as the final classification result for at least some particles. Here, "at least some particles" refers to particles whose confidence level in the preliminary classification result is less than a preset value.

[0069] In the above embodiments, a preset target detection model and a preset classification model are combined to classify particles. Specifically, the particles are first classified using the preset target detection model to obtain a preliminary classification result. Then, the confidence level of the preliminary classification result is calculated. If the confidence level of the preliminary classification result is high, the preliminary classification result is directly output as the final classification result for the particle. If the confidence level of the preliminary classification result is low, a single region image of the particle is input into the preset classification model for classification to obtain a second classification result. This second classification result is then used as the final classification result for the particle. For example, in one embodiment, if the preliminary classification result of the particle predicts the category as eosinophil, its confidence level is relatively high, and the result can be directly output. If the preliminary classification result of the particle predicts the category as monocyte, its confidence level is relatively low, and the particle can be further classified using the classification model before the result is output. Through this method, the accuracy of particle classification can be significantly improved.

[0070] In another embodiment, such as Figure 4 As shown, Figure 4 This is a schematic flowchart of another embodiment of a particle detection method for a sample detection device provided in this application. In this embodiment, the particle detection method includes:

[0071] S41: The pre-set target detection model is used to identify multiple particles in the particle image to obtain preliminary classification results and position information of multiple particles.

[0072] Step S41 is the same as step S21, and will not be repeated here.

[0073] S42: Based on the position information of multiple particles, obtain a region image of at least some particles.

[0074] Step S42 is the same as step S22, and will not be described again here.

[0075] S43: Identify the region images of at least some particles using a preset classification model to obtain a second classification result for at least some particles.

[0076] Step S43 is the same as step S23, and will not be repeated here.

[0077] S44: Obtain the confidence level of the preliminary classification results for multiple particles and the confidence level of the secondary classification results for at least some particles.

[0078] After obtaining the preliminary classification results of multiple particles and the secondary classification results of at least some particles, the confidence levels of the preliminary classification results and the secondary classification results of the particles can be calculated.

[0079] In some embodiments, a second classification result for each particle can be obtained through a preset classification model, and then the confidence level of the preliminary classification result and the confidence level of the second classification result for each particle can be calculated.

[0080] In other embodiments, a second classification result for some particles can be obtained through a preset classification model, and the confidence level of the second classification result for these particles and the confidence level of the initial classification result for these particles can be calculated.

[0081] S45: Based on the confidence levels of the preliminary classification results of multiple particles and the confidence levels of the secondary classification results of at least some particles, the final classification results of multiple particles are obtained.

[0082] After obtaining the confidence levels of the preliminary classification results for multiple particles and the confidence levels of the secondary classification results for at least some particles, the classification result with the higher confidence level can be selected as the final classification result for the particle by comparing the confidence levels of the preliminary classification results and the secondary classification results.

[0083] For example, in the above embodiment, if the preliminary classification result of a particle is predicted to be a lymphocyte, and the classification model outputs a result of basophil, and the confidence level of the classification model output is high, then the final classification result of the particle is considered to be a basophil.

[0084] In the above embodiments, by comparing the confidence levels of the preliminary classification results and the second classification results of the particles, the classification result with the higher confidence level is selected as the final classification result of the particles. In this way, by combining the two classification methods, a more accurate particle classification result is obtained.

[0085] In other embodiments, such as Figure 5 As shown, Figure 5 yes Figure 2A flowchart illustrating an embodiment of the training method for the preset classification model in step S23, the training method including:

[0086] S51: Obtain the training dataset for the preset classification model.

[0087] Collect the training dataset of the pre-defined classification model. The training data of various particles in the training dataset may be imbalanced.

[0088] For example, under normal circumstances, when the training data consists of white blood cell data, the proportions of the five cell types in blood cells vary considerably. In a specific embodiment, the proportions of each particle in the training dataset are: basophils 0.9%, eosinophils 1.8%, lymphocytes 29.1%, monocytes 6.8%, and neutrophils 61.4%. Since samples of some cells, such as eosinophils and basophils, are relatively scarce, it is impractical to expand the dataset through clinical trials in a short period. To improve this situation, data augmentation can be used.

[0089] S52: Augment the training data of some / all types of particles in the training dataset using image processing methods and / or generative artificial intelligence models to obtain an augmented training dataset.

[0090] In some embodiments, data augmentation can be performed on certain types of particles in the training dataset. For example, the training data of one or more particles with the smallest proportion can be augmented. Specifically, data augmentation can be performed on samples of eosinophils and basophils, where data is relatively scarce.

[0091] During the experiment, it was also found that although the data augmentation method improved the results compared to not using data augmentation, the proportion of data augmentation also affected the results. In a specific embodiment, the training data for basophils was augmented 8-12 times, the training data for eosinophils was augmented 4-8 times, and the training data for monocytes was augmented 2-5 times. In this way, the white blood cell classification samples were more balanced, which could significantly improve the accuracy of the model's particle classification results.

[0092] In other embodiments, data augmentation can be performed on each particle to obtain an augmented training dataset, ensuring that each particle has sufficient training data.

[0093] In one specific embodiment, training data can be augmented using conventional image processing methods, including flipping / rotating, contrast adjustment, noise reduction, random cropping, and other affine transformations or combinations thereof.

[0094] In other embodiments, the training data for one or more particles can be augmented using a generative artificial intelligence model to obtain an augmented training dataset. The generative artificial intelligence model can use a generative adversarial network model, a diffusion model, or a variational autoencoder model.

[0095] S53: Train the pre-defined classification model using the expanded training dataset.

[0096] After obtaining the augmented training dataset, the pre-defined classification model is trained using the augmented training dataset.

[0097] In the white blood cell classification process, considering the different sizes of different cells, when augmenting the training dataset and training the pre-defined classification model, image resizing (changing cell size) cannot be used. Instead, the images in both the training and augmented datasets are resized by padding with zeros around the edges or cropping to ensure consistent image size. Verification has shown that this method yields significantly better particle classification results than preprocessing methods that resize cells, especially for identifying monocytes (which are larger than the other four cell types).

[0098] In the above embodiments, data augmentation is used to address the problem of imbalanced training samples, reduce data collection costs, and train the model using the augmented training dataset, which can significantly improve the model's performance and robustness.

[0099] In summary, this application, by combining a pre-set target detection model and a classification model, can reliably classify particle types based on effective particle identification, thereby improving the accuracy of particle classification and the efficiency of detection. Moreover, the entire process requires no human intervention, greatly saving labor costs.

[0100] The above description is merely an embodiment of this application and does not limit the patent scope of this application. Any equivalent structural or procedural transformations made using the content of this application's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of this application.

Claims

1. A sample detection device, characterized in that, The sample detection device includes: The detection component is used to receive the sample to be tested; An image acquisition component is disposed on one side of the detection component and is used to acquire images of the sample to be tested to obtain a particle image, wherein the particle image includes a variety of particles; A processor, connected to the image acquisition component, is used for: The particle image is identified by a preset target detection model to obtain preliminary classification results and position information of the particles. Based on the position information of the various particles, a region image of at least some of the particles is obtained; The region images of at least some of the particles are identified by a preset classification model to obtain a second classification result for the at least some of the particles; Based on the preliminary classification results and / or the second classification results, the final classification results of the various particles are obtained.

2. The sample detection device according to claim 1, characterized in that, After obtaining the preliminary classification results of the various particles, the processor is further configured to: Obtain the confidence level of the preliminary classification result for each of the various particles; Among the various particles, the particles whose confidence level of the preliminary classification result is less than a preset value are obtained to obtain the at least part of the particles.

3. The sample detection device according to claim 2, characterized in that, The processor is also used for: When it is confirmed that the confidence level of the preliminary classification result of at least one particle among the multiple particles is greater than or equal to the preset value, the preliminary classification result of the at least one particle is taken as the final classification result of the at least one particle.

4. The sample detection device according to claim 1, characterized in that, The processor is also used for: The confidence levels of the preliminary classification results of the various particles and the confidence levels of the second classification results of at least some of the particles are obtained. Based on the confidence levels of the preliminary classification results of the various particles and the confidence levels of the second classification results of at least some of the particles, the final classification results of the various particles are obtained.

5. The sample detection device according to claim 1, characterized in that, The processor is also used for: Obtain the training dataset of the preset classification model; The training data of some / all types of particles in the training dataset are augmented using image processing methods and / or generative artificial intelligence models to obtain an augmented training dataset. The pre-defined classification model is trained using the augmented training dataset.

6. The sample detection device according to claim 5, characterized in that, The processor is also used for: The training data of one or more particles with the fewest number in the training dataset is augmented using image processing methods and / or generative artificial intelligence models to obtain the augmented training dataset.

7. The sample detection device according to claim 6, characterized in that, The particle image is a white blood cell image, and the various particles include: neutrophils, eosinophils, basophils, monocytes, and lymphocytes. The processor is also used for: The training data of basophils in the training dataset are amplified 8-12 times, the training data of eosinophils are amplified 4-8 times, and the training data of monocytes are amplified 2-5 times.

8. The sample detection device according to claim 5, characterized in that, The processor is also used for: When augmenting the training dataset and training the preset classification model, the images in the training dataset and the augmented training dataset are resized by padding with zeros around the images or by cropping.

9. The sample detection device according to claim 1, characterized in that, The preset classification model includes a residual neural network model; the preset target detection model includes: YOLO series models, fast region convolutional neural network models, or single-shot multi-frame detector models.

10. A particle classification method for a sample detection device, characterized in that, Based on the sample detection device according to any one of claims 1-9, the particle classification method includes: Acquire particle images, which include a variety of particles; The particle image is identified by a preset target detection model to obtain preliminary classification results and position information of the particles. Based on the position information of the various particles, a region image of at least some of the particles is obtained; The region image of the particles is identified by a preset classification model to obtain a second classification result for at least some of the particles; Based on the preliminary classification results and the second classification results, the final classification results of the various particles are obtained.