Multi-level hybrid model construction method and blood image stratification method

Abstract

This application relates to the field of blood analysis technology, specifically providing a method for constructing a multi-level hybrid model and a blood image layering method, aiming to solve the problem of existing blood image layering models lacking active learning capabilities. The method includes: continuously acquiring blood images; processing the blood images using a first model to obtain a first layering result, wherein the first model is an unsupervised clustering model; acquiring a first dataset generated after user correction of the first layering result; training a second model using the first dataset, wherein the second model is a supervised classification model based on machine learning; when the amount of data in the generated first dataset and the performance of the trained second model meet a first preset standard, using the trained second model to replace the first model as the layering engine of the multi-level hybrid model. This application constructs a feedback loop that continuously improves model performance using user-corrected data, effectively solving the limitation of existing static models whose performance becomes fixed after deployment.

Description

Technical Field

[0001] This application relates to the field of blood analysis technology, specifically providing a method for constructing a multi-level hybrid model and a method for layering blood images. Background Technology

[0002] Currently, methods for stratifying blood images typically employ static models, whose performance cannot be improved once deployed. This approach cannot actively learn from and utilize the valuable experience gained through manual corrections by doctors or technicians, and is prone to repeating the same errors, resulting in low stratification accuracy.

[0003] Accordingly, the field needs a new method for constructing multi-level hybrid models to address the aforementioned issues. Summary of the Invention

[0004] This application aims to solve the aforementioned technical problem, namely, to address the lack of active learning capabilities in existing blood image layering models.

[0005] In a first aspect, this application provides a method for constructing a multi-level hybrid model, the multi-level hybrid model including a first model and a second model, wherein the first model is deployed as the initial layering engine of the multi-level hybrid model, the method comprising: continuously acquiring blood images; processing the blood images using the first model to obtain a first layering result, wherein the first model is an unsupervised clustering model; acquiring a first dataset generated by a user after correcting the first layering result; training the second model using the first dataset, wherein the second model is a supervised classification model based on machine learning; and when the amount of data in the generated first dataset and the performance of the trained second model meet a first preset standard, using the trained second model to replace the first model as the layering engine of the multi-level hybrid model.

[0006] In one technical solution of the above-mentioned multi-level hybrid model construction method, the multi-level hybrid model further includes a third model, and the method further includes: processing blood images using the second model to obtain a second layering result; obtaining a second dataset generated after the user corrects the second layering result; training the third model using the second dataset, wherein the third model is a deep neural network model; when the amount of data in the generated second dataset and the performance of the trained third model meet a second preset standard, using the trained third model to replace the second model as the layering engine of the multi-level hybrid model.

[0007] In one technical solution of the above-mentioned multi-level hybrid model construction method, the method further includes: processing the blood image using the third model to obtain a third layer result; obtaining a third dataset generated after the user corrects the third layer result; and continuously training the third model using the third dataset.

[0008] In one technical solution of the above-mentioned multi-level hybrid model construction method, the step of processing the blood image using the first model to obtain a first layered result includes: performing layered sampling on the acquired blood image based on pixel gradients to obtain a set of pixel samples covering multiple regions in the vertical direction of the blood image; and performing unsupervised clustering on the set of pixel samples using the first model to obtain the first layered result.

[0009] In one technical solution of the above-mentioned multi-level hybrid model construction method, the first model includes a batch clustering algorithm. The step of using the first model to perform unsupervised clustering on the pixel sample set to obtain a first hierarchical result includes: extracting a multi-dimensional feature vector corresponding to each pixel sample in the pixel sample set, wherein the multi-dimensional feature vector includes at least color features and normalized vertical position features; obtaining a clustering result based on the multi-dimensional feature vector using the batch clustering algorithm; and obtaining a first hierarchical result based on the clustering result using a preset cluster correction rule and statistical methods.

[0010] In one technical solution of the above-mentioned multi-level hybrid model construction method, the statistical method includes quantile statistics. The step of obtaining the first stratification result based on the clustering result using a preset cluster correction rule and statistical methods includes: determining the actual number of layers of the blood image based on the clustering result using the preset cluster correction rule; determining the boundary of each layer in the actual number of layers based on the actual number of layers and the clustering result using the quantile statistics method; and obtaining the first stratification result based on the actual number of layers and the boundary of each layer in the actual number of layers.

[0011] In one technical solution of the above-mentioned multi-level hybrid model construction method, the first dataset includes at least: multiple feature vectors generated by the user after correcting the first layering result, wherein any feature vector includes color features, normalized vertical position features and neighborhood statistical features; the second dataset includes at least: a layered mask map generated by the user after correcting the second layering result.

[0012] In a second aspect, a blood image layering method is provided, the method comprising: acquiring a blood image to be analyzed and a multi-level hybrid model constructed according to the method in the first aspect or any corresponding technical solution; and calling the layering engine of the multi-level hybrid model to process the blood image to be analyzed to obtain a corresponding layering result. In a third aspect, a smart device is provided, the smart device including at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores a computer program, which, when executed by the at least one processor, implements the multi-level hybrid model construction method of the first aspect or any corresponding technical solution and the blood image layering method of the second aspect.

[0013] In a fourth aspect, a computer-readable storage medium is provided, wherein a plurality of program codes are stored therein, the program codes being adapted to be loaded and run by a processor to perform the multi-level hybrid model construction method of the first aspect or any of the corresponding technical solutions above, and the blood image layering method of the second aspect above.

[0014] This application deploys an unsupervised clustering model (the first model) as the initial hierarchical engine for a multi-level hybrid model, enabling rapid hierarchical results and meeting real-time requirements. By acquiring user-corrected hierarchical data (the first dataset) and using it to train the second model, the hierarchical engine is automatically upgraded when the data volume of the first dataset and the performance of the second model reach preset standards. This application constructs a feedback loop that continuously improves model performance using corrected data, effectively overcoming the limitation of existing static models whose performance becomes fixed after deployment.

[0015] By introducing a third model, the corrected data obtained after running the second model (i.e., the second dataset) is used to train the third model. Once a preset standard is met, this third model is used as the final hierarchical engine, further improving the performance of the multi-level hybrid model. Furthermore, the third model, serving as the final hierarchical engine, continues to be trained using corrected data, achieving the goal of enabling the multi-level hybrid model to continuously learn and optimize autonomously.

[0016] This application employs layered sampling based on pixel gradients to obtain a set of pixel samples covering multiple regions in the vertical direction of the blood image. This ensures that key samples are collected from all layers, preventing omissions. By processing the blood image using color features and normalized vertical position features, color and spatial location information are combined, improving the accuracy of layering. The first layering result is obtained based on the clustering results using preset cluster correction rules and statistical methods, effectively resisting noise in the image and enhancing the accuracy and anti-interference capability of layering. Attached Figure Description

[0017] The disclosure of this application will become more readily understood with reference to the accompanying drawings. It will be readily understood by those skilled in the art that these drawings are for illustrative purposes only and are not intended to limit the scope of protection of this application. In the drawings: Figure 1 This is a schematic flowchart of the main steps of a multi-level hybrid model construction method according to an embodiment of this application; Figure 2 This is a schematic diagram of the main structure of a smart device according to an embodiment of this application. Detailed Implementation

[0018] Some embodiments of this application are described below with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are merely illustrative of the technical principles of this application and are not intended to limit the scope of protection of this application.

[0019] In the description of this application, the terms "first," "second," etc., are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. 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 necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus. The terms "installed," "connected," and "linked" should be interpreted broadly; for example, they can refer to a fixed connection, a detachable connection, or an integral connection; a mechanical connection or an electrical connection; a direct connection or an indirect connection via an intermediate medium; a connection within two elements; a wireless connection or a wired connection.

[0020] Furthermore, "module" and "processor" can include hardware, software, or a combination of both. A module can include hardware circuitry, various suitable sensors, communication ports, and memory, and may also include software components such as program code, or a combination of software and hardware. A processor can be a central processing unit, microprocessor, image processor, digital signal processor, or any other suitable processor. A processor has data and / or signal processing capabilities. A processor can be implemented in software, in hardware, or a combination of both. Computer-readable storage media includes any suitable medium capable of storing program code, such as magnetic disks, hard disks, optical disks, flash memory, read-only memory, random access memory, etc.

[0021] Current automated blood stratification methods have the following problems: (1) They cannot achieve self-optimization and learning. Once deployed, their performance cannot be improved. They cannot effectively learn from and utilize the human experience generated by manual corrections by users, leading to repeated stratification errors. (2) The stratification speed is slow and resource consumption is high. For high-resolution images, a large amount of pixel data usually needs to be processed, which cannot meet the real-time requirements of clinical or laboratory settings. (3) The stratification accuracy and robustness are insufficient. Existing methods have difficulty accurately identifying thinner layers such as the buffy coat, and they are prone to errors when faced with interference such as light spots, dirt, or uneven lighting.

[0022] To address the aforementioned problems, this application provides a method for constructing a multi-level hybrid model. (See appendix.) Figure 1 , Figure 1 This is a schematic flowchart illustrating the main steps of a multi-level hybrid model construction method according to an embodiment of this application. Figure 1 As shown, the method in this application embodiment mainly includes the following steps S1 to S4.

[0023] Step S1: Continuously acquire blood images and process the blood images using the first model to obtain the first layering result.

[0024] In this embodiment, a blood image refers to a digital image obtained by acquiring images of a blood sample tube, such as a blood layering tube after centrifugation. It typically includes a plasma layer, a white blood cell layer, and a red blood cell layer. When constructing a multi-level hybrid model, blood images can be continuously acquired to train the model until it meets preset requirements.

[0025] It should be noted that the multi-level hybrid model includes at least two switchable models (or algorithms), one of which (such as an unsupervised clustering model) is deployed as the initial hierarchical engine. The hierarchical engine is the core model in the multi-level hybrid model used to process the input blood image to obtain the hierarchical results; the other (such as a supervised classification model based on machine learning) is trained on data from which the user corrects the hierarchical results, and after training, it replaces the previous model as the hierarchical engine.

[0026] As an example, the first model in the multi-level hybrid model is the K-means clustering algorithm, which is deployed as the initial hierarchical engine. It is used to process multiple blood images sequentially to obtain the first hierarchical result for each blood image. This embodiment uses an unsupervised clustering model as the initial hierarchical engine, which can start working without pre-labeled data and can quickly provide hierarchical results, meeting real-time requirements.

[0027] In one optional implementation, the blood image is processed using a first model to obtain a first layering result, specifically including the following steps S11 and S12.

[0028] Step S11: Perform layered sampling on the acquired blood image based on pixel gradient to obtain a set of pixel samples covering multiple regions in the vertical direction of the blood image.

[0029] Existing simple random sampling methods cannot guarantee that all layers are covered. The strategy of layered sampling based on pixel gradient in this embodiment is as follows: First, the image is divided into grids according to a preset step size (such as every 10 rows). Only pixels whose pixel color (or grayscale) gradient value exceeds the threshold, such as pixels with color abrupt changes, and pixels in the middle of each pixel row are retained. This achieves the purpose of retaining both the blood image boundary information and the representative colors inside the layer.

[0030] In one implementation, the average gradient magnitude of each pixel row in the vertical direction of the blood image can also be calculated, with a small amount of sampling in areas with flat gradients and an increased sampling density in areas with drastic gradient changes, i.e., potential layer boundaries.

[0031] This embodiment combines grid-based positioning with gradient threshold filtering to generate a pixel sample set that can comprehensively cover all vertical regions of the blood image while focusing on capturing the transition zones between different layers. This effectively avoids the risk of missing thin layers (such as the white film layer) due to sampling randomness, thus improving the accuracy of layer recognition.

[0032] Step S12 involves using the first model to perform unsupervised clustering on the pixel sample set to obtain the first hierarchical result. This specifically includes steps S121 to S123.

[0033] Step S121: Extract the multi-dimensional feature vector corresponding to each pixel sample in the pixel sample set, wherein the multi-dimensional feature vector includes at least color features and normalized vertical position features.

[0034] In this embodiment, a multi-dimensional feature vector is constructed for each pixel sample (pixel point). This vector mainly includes color features (such as the three channel values ​​of the pixel point in the RGB color space) and normalized vertical position features, such as the normalized vertical (Y-axis) coordinates from the bottom to the top of the test tube. For example, the normalized vertical position feature can be calculated by dividing the vertical coordinates of the pixel point by the total height of the blood image, and the range is within the interval [0,1].

[0035] In this embodiment, color features and normalized vertical position features are used simultaneously to distinguish layers with similar colors but different vertical positions (such as plasma and reflective areas), which can effectively solve the problem of different layers of the same color and further improve the accuracy of layering.

[0036] Step S122: Obtain clustering results using a batch clustering algorithm based on multidimensional feature vectors.

[0037] In this embodiment, the Mini-Batch K-Means clustering algorithm is used to perform unsupervised clustering of the above-mentioned multidimensional feature vectors. Compared with traditional algorithms that require full data iteration, the Mini-Batch K-Means clustering algorithm randomly selects a small portion of samples each time to update the centroids, which greatly improves the calculation speed and reduces memory usage. It is very suitable for embedded devices with limited computing resources or real-time systems that require fast response.

[0038] Step S123: Obtain the first stratification result based on the clustering results using preset cluster correction rules and statistical methods.

[0039] In this embodiment, the preset cluster correction rules include, but are not limited to, minimizing the intra-cluster distance and maximizing the inter-cluster distance. As an example, the preset cluster correction rules are used to correct the number of layers. For instance, when attempting clustering with K=2, 3, and 4, if the centroids of two clusters are found to be too close vertically, they are merged to determine the actual number of physical layers.

[0040] Optionally, statistical methods include quantile analysis, such as using the median or upper and lower quartiles to perform vertical stratification based on clustering results. As an example, layer 1 and layer 2 are obtained through clustering. The median of the vertical position coordinates of all pixels in layer 1 and layer 2 is calculated respectively, and the average of the two medians is used to determine the boundaries of layer 1 and layer 2. Compared to using the maximum and minimum values ​​to determine layer boundaries, using the median has the advantages of filtering noise and improving anti-interference capabilities.

[0041] It should be noted that the execution order of determining layer boundaries and correcting the number of layers can be selected according to actual needs, and this embodiment does not impose specific limitations on this.

[0042] Step S2: Obtain the first dataset generated after the user corrects the first stratification result.

[0043] In this embodiment, the first dataset is a collection of data generated after the user corrects the first layering result obtained in step S1, such as by manually modifying the layer boundaries in the interactive interface. For example, the first dataset may be multiple feature vectors generated after the user corrects the first layering result, wherein any feature vector includes color features, normalized vertical position features, and neighborhood statistical features.

[0044] Step S3: Train the second model using the first dataset.

[0045] In this embodiment, the second model is a supervised classification model based on machine learning, such as a random forest or a gradient boosting decision tree. The model is trained using the first dataset obtained in step S2. The input is a multi-dimensional feature vector, and the output is the corresponding layer label or layer label probability.

[0046] Step S4: When the amount of data in the generated first dataset and the performance of the trained second model meet the first preset standard, the trained second model is used to replace the first model as the hierarchical engine of the multi-level hybrid model.

[0047] In this embodiment, the first preset standard includes a first quantity threshold and a first performance threshold. Specifically, when the amount of data in the first dataset accumulates to the first quantity threshold, a background test is triggered. A validation set is generated using historical data to test the second model. If the performance of the second model meets the first performance threshold, such as if the accuracy of the second model on the validation set exceeds the first performance threshold, the trained second model is used to replace the first model as the hierarchical engine of the multi-level hybrid model.

[0048] In an optional implementation, the multi-level hybrid model further includes a third model, and after step S4 above, it may further include steps S5 to S8 as follows.

[0049] Step S5: Process the blood image using the second model to obtain the second layering result.

[0050] In this embodiment, the blood image is input into a second model to obtain the corresponding layered results. For example, for the input blood image, a multi-dimensional feature vector is extracted for each sampling point according to a preset grid step size. This vector contains at least color features and normalized vertical position features. Subsequently, this vector is input into the trained second model to obtain the classification label of its layer (e.g., "1" represents the plasma layer, "2" represents the white blood cell layer, and "3" represents the red blood cell layer). After processing all sampling points, a complete layered mask is generated, which identifies the blood component layer to which each pixel belongs using different colors or numerical values.

[0051] Step S6: Obtain the second dataset generated after the user corrects the second stratification results.

[0052] In this embodiment, the second dataset is a dataset generated after the user corrects the second hierarchical result obtained in step S5.

[0053] Step S7: Train the third model using the second dataset, where the third model is a deep neural network model.

[0054] In this embodiment, the third model is a deep neural network model, such as an end-to-end model like Transformer or U-Net, which can directly learn the mapping relationship from the original blood image to the layered mask map without relying on manually designed features.

[0055] Step S8: When the amount of data in the generated second dataset and the performance of the trained third model meet the second preset standard, the trained third model is used to replace the second model as the hierarchical engine of the multi-level hybrid model.

[0056] In this embodiment, the second preset standard includes a second quantity threshold and a second performance threshold. Specifically, when the amount of data in the second dataset accumulates to the second quantity threshold, a background test is triggered. A validation set is generated using historical data to test the third model. If the accuracy of the third model on the validation set exceeds the second performance threshold, the trained third model is used to replace the second model as the hierarchical engine of the multi-level hybrid model.

[0057] In an optional implementation, after step S8, steps S9 to S11 may be further included.

[0058] Step S9: Process the blood image using the third model to obtain the third layering result.

[0059] In this embodiment, the third model serves as the layering engine of the multi-level hybrid model. Based on the input blood image, it outputs the corresponding third layering result, such as a pixel-level layering mask. This mask not only identifies the partitions and boundaries of each layer, but also has better boundary smoothness, anti-interference ability, and recognition accuracy for thin layers than the first and second models.

[0060] Step S10: Obtain the third dataset generated after the user corrects the third stratification results.

[0061] In this embodiment, the third dataset is a set of data generated after the user corrects the third layering result obtained in step S9, such as the corrected layered mask image.

[0062] Step S11: Continuously train the third model using the third dataset.

[0063] In this embodiment, the third dataset obtained in step S10 is used to perform continuous incremental learning or fine-tuning on the third model.

[0064] This embodiment uses the third model as the final layering engine and continuously performs incremental learning or fine-tuning on it, thereby further improving the layering accuracy and robustness of the multi-level hybrid model.

[0065] In one alternative implementation, when the system is reset or a new type of centrifuge tube is encountered, causing the current layering engine confidence to be very low, the first model can be called back as the layering engine, and the second and third models can be retrained.

[0066] In another aspect, this application also provides a method for layering blood images, specifically including the following steps S12 and S13.

[0067] Step S12: Obtain the blood image to be analyzed and the multi-level hybrid model constructed according to any of the above embodiments.

[0068] In practical applications, this blood image layering method can be executed on embedded devices (such as embedded blood cell analyzers). First, a pre-built multi-level blending model is loaded. It should be noted that this model may be at different stages of evolution; for example, in the initial deployment phase, its layering engine is the first model; after accumulating a certain amount of data, the layering engine is upgraded to the second model; and after long-term use, the layering engine evolves into the third model. Next, one or more blood images to be analyzed are acquired through an image acquisition interface (such as a USB camera, network transmission, or file upload), and necessary preprocessing (such as format conversion and size normalization) is performed. Finally, the preprocessed blood images are input into the multi-level blending model.

[0069] Step S13: Call the hierarchical engine of the multi-level hybrid model to process the blood image to be analyzed, so as to obtain the corresponding hierarchical results.

[0070] In this embodiment, the layering engine of the loaded multi-level hybrid model is invoked to output structured layering results, such as an intuitive layered mask or a structured JSON document containing statistical information for each layer.

[0071] The blood image layering method in this embodiment can effectively reduce manual intervention time and has no complicated processing steps, thus having higher processing efficiency compared to existing methods.

[0072] It should be noted that although the steps in the above embodiments are described in a specific order, those skilled in the art will understand that in order to achieve the effect of this application, different steps do not necessarily have to be executed in such an order. They can be executed simultaneously (in parallel) or in other orders. These adjusted solutions are equivalent to the technical solutions described in this application and therefore will also fall within the protection scope of this application.

[0073] Those skilled in the art will understand that all or part of the processes in the method of the above-described embodiment can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable file, or some intermediate form. The computer-readable storage medium can include any entity or device capable of carrying the computer program code, a medium, a USB flash drive, a portable hard drive, a magnetic disk, an optical disk, a computer memory, a read-only memory, a random access memory, an electrical carrier signal, a telecommunication signal, and a software distribution medium, etc.

[0074] Another aspect of this application provides a computer-readable storage medium.

[0075] In one embodiment of a computer-readable storage medium according to this application, the computer-readable storage medium can be configured to store a program for performing the multi-level hybrid model construction method and the blood image layering method of the above-described method embodiments. This program can be loaded and run by a processor to implement the above-described multi-level hybrid model construction method and the blood image layering method. For ease of explanation, only the parts related to the embodiments of this application are shown; for specific technical details not disclosed, please refer to the method section of the embodiments of this application. The computer-readable storage medium can be a storage device comprising various electronic devices. Optionally, in the embodiments of this application, the computer-readable storage medium is a non-transitory computer-readable storage medium.

[0076] Another aspect of this application provides a smart device.

[0077] In one embodiment of a smart device according to this application, the smart device may include at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores a computer program, which, when executed by the at least one processor, implements the method described in any of the above embodiments. The smart device described in this application may be a device such as an embedded blood cell analyzer. See appendix. Figure 2 , Figure 2 The image exemplarily illustrates a communication connection between memory 11 and processor 12 via a bus.

[0078] In some embodiments of this application, the smart device may further include at least one sensor for sensing information. The sensor is communicatively connected to any type of processor mentioned in this application. Optionally, the smart device described in this application may be, but is not limited to, a mobile phone, tablet computer, desktop computer, laptop computer, handheld computer, notebook computer, in-vehicle device, ultra-mobile personal computer (UMPC), etc., and this application embodiment does not limit this.

[0079] The technical solutions of this application have been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of this application is obviously not limited to these specific embodiments. Without departing from the principles of this application, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will all fall within the scope of protection of this application.

Claims

1. A method for constructing a multi-level hybrid model, characterized in that, The multi-level hybrid model includes a first model and a second model, wherein the first model is deployed as the initial hierarchical engine of the multi-level hybrid model, and the method includes: Blood images are continuously acquired, and the blood images are processed using the first model to obtain a first layering result, wherein the first model is an unsupervised clustering model; Obtain the first dataset generated after the user corrects the first layering result; The second model is trained using the first dataset, wherein the second model is a supervised classification model based on machine learning; When the amount of data in the first generated dataset and the performance of the trained second model meet the first preset standard, the trained second model is used to replace the first model as the hierarchical engine of the multi-level hybrid model.

2. The multi-level hybrid model construction method according to claim 1, characterized in that, The multi-level hybrid model also includes a third model, and the method further includes: The second model is used to process the blood image to obtain a second layered result; Obtain the second dataset generated after the user corrects the second stratification result; The third model is trained using the second dataset, wherein the third model is a deep neural network model; When the amount of data in the generated second dataset and the performance of the trained third model meet the second preset standard, the trained third model is used to replace the second model as the hierarchical engine of the multi-level hybrid model.

3. The multi-level hybrid model construction method according to claim 2, characterized in that, The method further includes: The blood image is processed using the third model to obtain a third layering result; Obtain the third dataset generated after the user corrects the third layer result; The third model is continuously trained using the third dataset.

4. The method for constructing a multi-level hybrid model according to any one of claims 1 to 3, characterized in that, The step of processing the blood image using the first model to obtain the first layering result includes: The acquired blood image is subjected to stratified sampling based on pixel gradient to obtain a set of pixel samples covering multiple regions in the vertical direction of the blood image; The first model is used to perform unsupervised clustering on the pixel sample set to obtain the first hierarchical result.

5. The multi-level hybrid model construction method according to claim 4, characterized in that, The first model includes a batch clustering algorithm, and the step of using the first model to perform unsupervised clustering on the pixel sample set to obtain a first hierarchical result includes: Extract the multidimensional feature vector corresponding to each pixel sample in the pixel sample set, wherein the multidimensional feature vector includes at least color features and normalized vertical position features; Based on the multidimensional feature vectors, the batch clustering algorithm is used to obtain the clustering results; The first stratification result is obtained based on the clustering results using preset cluster correction rules and statistical methods.

6. The multi-level hybrid model construction method according to claim 5, characterized in that, The statistical method includes quantile statistics, and the step of obtaining the first stratification result based on the clustering result using a preset cluster correction rule and statistical methods includes: The actual number of layers in the blood image is determined based on the clustering results using a preset cluster correction rule. The boundary of each layer in the actual number of layers is determined using the quantile statistics method based on the actual number of layers and the clustering results. The first layering result is obtained based on the actual number of layers and the boundary of each layer in the actual number of layers.

7. The multi-level hybrid model construction method according to claim 3, characterized in that, The first dataset includes at least: multiple feature vectors generated by the user after correcting the first layering result, wherein any feature vector includes color features, normalized vertical position features, and neighborhood statistical features; The second dataset includes at least: a layered mask image generated after the user corrects the second layered result.

8. A method for layering blood images, characterized in that, The method includes: Acquire the blood image to be analyzed and the multi-level hybrid model constructed according to any one of claims 1 to 7; The hierarchical engine of the multi-level hybrid model is invoked to process the blood image to be analyzed in order to obtain the corresponding hierarchical results.

9. A smart device, characterized in that, include: At least one processor; And a memory communicatively connected to the at least one processor; wherein the memory stores a computer program that, when executed by the at least one processor, implements the multi-level hybrid model construction method of any one of claims 1 to 7 and the blood image layering method of claim 8.

10. A computer-readable storage medium storing a plurality of program codes, characterized in that, The program code is adapted to be loaded and run by a processor to perform the multi-level hybrid model construction method of any one of claims 1 to 7 and the blood image layering method of claim 8.

Images