Virtual focus blurring judgment method and device for white blood cell image
By combining unsupervised and supervised learning methods and training the model using unlabeled and poorly labeled leukocyte image samples, the problem of inconsistent sharpness evaluation in leukocyte image blur determination was solved, achieving efficient blur determination results and reducing annotation costs.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- BEIJING XIAOYING TECH CO LTD
- Filing Date
- 2026-01-07
- Publication Date
- 2026-05-01
AI Technical Summary
In existing technologies, methods for determining blurriness in leukocyte images suffer from inconsistent sharpness evaluation metrics, making it difficult to effectively distinguish whether images of different content are blurry. Furthermore, supervised learning-based methods require significant annotation costs.
A combination of unsupervised and supervised learning methods is adopted. The model is pre-trained using unlabeled white blood cell image samples and then trained with a small number of samples labeled with blur levels to construct a blur detection model. The accuracy of the model is improved through data augmentation and preprocessing.
While saving annotation costs, it achieves effective defocus blur determination of white blood cell images, improves the determination effect, and reduces the impact of differences in brightness and cell type.
Smart Images
Figure CN121962048A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the fields of artificial intelligence and medical data processing technology, specifically to a method and apparatus for determining the out-of-focus blur of white blood cell images. Background Technology
[0002] During the scanning process of glass slides, the scanned images may be out of focus or blurry due to various reasons. In order to display only clear images of white blood cells to the user, it is necessary to determine whether the images of white blood cells to be displayed are out of focus, and then filter out the out-of-focus or blurry images. Therefore, a method for determining the out-of-focus or blurry nature of white blood cell images is needed and applied to scanning equipment.
[0003] Traditional methods for determining image blur or out of focus calculate an image sharpness index using a sharpness evaluation function. This index is then compared to a pre-defined threshold to determine if the image is out of focus. Sharpness evaluation functions include the Laplacian gradient function, gray-level variance function, and entropy function.
[0004] However, the sharpness rating index calculated using the sharpness rating function may be inconsistent across images with different content. Taking the Laplacian gradient function as an example, for a series of images with the same content but different degrees of blur, a larger sharpness rating index indicates a sharper image; however, for a series of images with different content, there may be cases where an image that is subjectively sharp has a smaller sharpness rating index than an image that is subjectively blurry. Therefore, it is difficult to use a fixed threshold to determine whether images with different content are blurry.
[0005] To address the inconsistency of sharpness evaluation metrics in traditional methods, existing technologies have introduced supervised learning-based judgment methods. Supervised learning-based image blur determination methods first train a machine learning model using a labeled dataset. Then, the model calculates the image's sharpness evaluation metric by inputting an image into the training dataset. Finally, a pre-defined threshold is used to compare this metric with the image's sharpness evaluation metric to determine whether the image is out of focus. However, supervised learning-based image blur determination methods require the construction of specially labeled datasets for specific scenes, a process that incurs significant time and manpower costs. Summary of the Invention
[0006] Therefore, embodiments of the present invention provide a method and apparatus for determining the out-of-focus blur of white blood cell images to solve at least one of the above-mentioned technical problems.
[0007] To achieve the above objectives, the embodiments of the present invention provide the following technical solutions: This invention provides a method for determining out-of-focus blur in white blood cell images, the method comprising: Acquire the target image to be identified; The target image is input into a pre-trained blur detection model to obtain the recognition result output by the blur detection model. The out-of-focus blur determination model is obtained by performing unsupervised and supervised learning using a sample dataset. The sample dataset is constructed based on a large number of collected white blood cell images. The unsupervised learning involves pre-training a pre-constructed machine learning network using unlabeled white blood cell image samples to obtain an initial model. The supervised learning involves training the initial model using white blood cell image samples labeled with blur levels to obtain the out-of-focus blur determination model.
[0008] In some embodiments, constructing a sample dataset specifically includes: Acquire several sets of white blood cell images at different focal distances within the same field of view; Align each group of white blood cell images; For each group of aligned white blood cell images, white blood cell target detection is performed on the clearest image in each group to obtain white blood cell mini-images, and a sample dataset for unsupervised learning is constructed based on each white blood cell mini-image. We selected small images of white blood cells from certain groups and manually labeled them. Based on the labeled white blood cell images, we constructed a sample dataset for supervised learning.
[0009] In some embodiments, aligning each group of white blood cell images specifically includes: Arrange a group of images from the same field of view in ascending order of focus distance; The sharpest image in a set of images is selected using the energy gradient function of the green channel. The translation matrix required for alignment between two adjacent images is calculated with the sharper image as a reference. The translation matrix required for each image to be aligned with the sharpest image is calculated by summing the translation matrices obtained in the previous step. Align each image with the clearest image and crop all images using the smallest possible boundary range.
[0010] In some embodiments, data augmentation is performed before both supervised and unsupervised learning on the sample dataset.
[0011] In some embodiments, the unsupervised learning includes an encoding phase and a decoding phase; The encoding stage includes: The data-enhanced white blood cell image was divided into several patches and their locations were encoded. Randomly mask the location-encoded tiles; Input the unmasked patches into the encoder to obtain encoded features; The decoding stage includes: The encoding features are supplemented with masking words according to the masking position in the encoding stage; The supplemented coded features are then positionally encoded and input into the decoder to obtain the decoded image.
[0012] In some embodiments, supervised learning employs the encoder obtained from the unsupervised learning, wherein the supervised learning includes: The data-augmented image is divided into several patches and their locations are encoded. The location-encoded patches are input into the encoder to obtain the encoded features; The encoded features are input into a linear classifier to obtain the predicted scores.
[0013] In some embodiments, when using the defocusing blur determination model to determine whether a white blood cell image is defocused, the image is first preprocessed and the defocusing blur determination threshold for each category of white blood cell images is statistically determined.
[0014] The present invention also provides a device for determining the out-of-focus blur of a white blood cell image, the device comprising: An image acquisition unit is used to acquire the target image to be identified. The result generation unit is used to input the target image into a pre-trained defocus blur determination model to obtain the recognition result output by the defocus blur determination model. The out-of-focus blur determination model is obtained by performing unsupervised and supervised learning using a sample dataset. The sample dataset is constructed based on a large number of collected white blood cell images. The unsupervised learning involves pre-training a pre-constructed machine learning network using unlabeled white blood cell image samples to obtain an initial model. The supervised learning involves training the initial model using white blood cell image samples labeled with blur levels to obtain the out-of-focus blur determination model.
[0015] The present invention also provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of the method described above.
[0016] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method described above.
[0017] The present invention provides a method and apparatus for determining the out-of-focus blur of white blood cell images. By acquiring a target image to be identified and inputting the target image into a pre-trained out-of-focus blur determination model, the recognition result output by the out-of-focus blur determination model can be obtained. The out-of-focus blur determination model is obtained through unsupervised and supervised learning using a sample dataset. The sample dataset is constructed based on a large number of collected white blood cell images. The unsupervised learning involves pre-training a pre-constructed machine learning network using unlabeled white blood cell image samples to obtain an initial model. The supervised learning involves training the initial model using white blood cell image samples labeled with blur levels to obtain the out-of-focus blur determination model.
[0018] Thus, the method and apparatus provided by this invention employ a combination of unsupervised and supervised learning, using a large amount of unlabeled white blood cell data and a small amount of white blood cell data with varying degrees of blur to train the machine learning model, achieving a better effect in determining out-of-focus blur while saving labeling costs. Attached Figure Description
[0019] To more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely exemplary, and those skilled in the art can derive other embodiments based on the provided drawings without creative effort.
[0020] The structures, proportions, sizes, etc. illustrated in this specification are only for the purpose of assisting those skilled in the art in understanding and reading the content disclosed herein, and are not intended to limit the conditions under which the present invention can be implemented. Therefore, they have no substantial technical significance. Any modifications to the structure, changes in the proportions, or adjustments to the size, without affecting the effects and objectives that the present invention can produce, should still fall within the scope of the technical content disclosed in the present invention.
[0021] Figure 1 A flowchart of the method for determining the out-of-focus blur of white blood cell images provided by the present invention; Figure 2 This is a structural block diagram of the defocusing and blur determination device for white blood cell images provided by the present invention; Figure 3 This is a structural block diagram of a computer device provided by the present invention. Detailed Implementation
[0022] The following specific embodiments illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0023] In one specific implementation, please refer to Figure 1 The method for determining the out-of-focus blur of white blood cell images provided by this invention includes the following steps: S110: Acquire the target image to be identified; the target image is the image to be identified obtained using the slow scan mode of the scanning device; S120: Input the target image into a pre-trained defocus blur determination model to obtain the recognition result output by the defocus blur determination model; The out-of-focus blur determination model is obtained by performing unsupervised and supervised learning using a sample dataset. The sample dataset is constructed based on a large number of collected white blood cell images. The unsupervised learning involves pre-training a pre-constructed machine learning network using unlabeled white blood cell image samples to obtain an initial model. The supervised learning involves training the initial model using white blood cell image samples labeled with blur levels to obtain the out-of-focus blur determination model.
[0024] The method for constructing a white blood cell blur determination dataset provided by this invention can be roughly divided into four parts: data collection, data alignment, acquisition of white blood cell thumbnails, and manual annotation. Specifically, constructing the sample dataset includes the following steps: S210: Acquire several sets of white blood cell images at different focal distances under the same field of view; during the data collection process, use the slow scan mode of the scanning device to acquire several sets of images at different focal distances under the same field of view. When arranging a set of images in ascending order of focal distance, their clarity needs to meet the order of blurry → clear → blurry.
[0025] S220: Align each group of white blood cell images; S230: For each group of aligned white blood cell images, perform white blood cell target detection on the clearest image in each group to obtain white blood cell mini-images, and construct a sample dataset for unsupervised learning based on each white blood cell mini-image; during the acquisition of white blood cell mini-images, perform white blood cell target detection on the clearest image in each group to obtain the position information of several white blood cells in the field of view, and then use the position information of white blood cells to cut out each aligned image in the field of view to obtain several groups of aligned white blood cell mini-images with different focus distances.
[0026] S240: Select a subset of white blood cell images for manual annotation, and construct a sample dataset for supervised learning based on the annotated white blood cell images. For manual annotation, select a small number of white blood cell images from a set, using the following method: select the image that is subjectively the clearest in a set, and pair it with all other images in that set to form a "clear-blurred" image pair.
[0027] In step S220, each group of white blood cell images is aligned, specifically including: Arrange a group of images from the same field of view in ascending order of focus distance; The sharpest image in a set of images is selected using the energy gradient function of the green channel. The translation matrix required for alignment between two adjacent images is calculated with the sharper image as a reference. The translation matrix required for each image to be aligned with the sharpest image is calculated by summing the translation matrices obtained in the previous step. Align each image with the clearest image and crop all images using the smallest possible boundary range.
[0028] Specifically, during data alignment, due to slight lens shake when acquiring images at different focus distances within the same field of view, each group of images needs to be aligned. The alignment process is as follows: First, arrange the images in a group within the same field of view in ascending order of focus distance. Then, use the energy gradient function of the green channel to select the sharpest image in the group. For adjacent images, calculate the translation matrix required for alignment using the sharper image as a reference. Calculate the translation matrix required for aligning each image with the sharpest image by summing the translation matrices obtained in the previous step. Finally, align each image with the sharpest image and crop all images using the minimum boundary range. The ORB method is used to calculate and match the feature points of adjacent images.
[0029] In some embodiments, data augmentation is performed before both supervised and unsupervised learning on the sample dataset. Both supervised and unsupervised learning methods augment the input data to improve the model's generalization ability. The data augmentation process is the same for both supervised and unsupervised learning, specifically: first, the image is randomly cropped and restored to its original size; then, the image's hue, saturation, and brightness are randomly changed; finally, the image is normalized and resized to a specified size.
[0030] Furthermore, the model training of this invention is divided into two parts: unsupervised and supervised learning. The unsupervised learning part uses a large amount of unlabeled white blood cell data for large-scale pre-training to obtain a good initial model weight, which is beneficial for improving the performance of subsequent supervised learning. The supervised learning part, based on the unsupervised learning model, uses white blood cell data with varying degrees of blur to train the model, enabling it to achieve better performance on the specific task of determining the blurriness of white blood cell images. The training data for unsupervised training is all data in the dataset, while the training data for supervised learning consists of manually labeled "sharp-blurred" image pairs from the dataset.
[0031] In some embodiments, the unsupervised learning includes an encoding phase and a decoding phase; wherein the encoding phase includes: The data-enhanced white blood cell image was divided into several patches and their locations were encoded. Randomly mask the location-encoded tiles; Input the unmasked patches into the encoder to obtain encoded features; The decoding stage includes: The encoding features are supplemented with masking words according to the masking position in the encoding stage; The supplemented coded features are then positionally encoded and input into the decoder to obtain the decoded image.
[0032] Specifically, the unsupervised learning part adopts a mask autoencoder architecture, which consists of two stages: encoding and decoding. In the encoding stage, the augmented image is first divided into several patches and positionally encoded. Then, the positionally encoded patches are randomly masked. Finally, the unmasked patches are input into the encoder to obtain encoded features. In the decoding stage, the encoded features are supplemented with mask terms according to the previously masked positions. Then, the supplemented encoded features are positionally encoded and input into the decoder to obtain the decoded image. During training, the mask autoencoder is trained unsupervised by reducing the error between the input image and the decoded image output by the mask autoencoder.
[0033] In some embodiments, supervised learning employs the encoder obtained from the unsupervised learning, wherein the supervised learning includes: The data-augmented image is divided into several patches and their locations are encoded. The location-encoded patches are input into the encoder to obtain the encoded features; The encoded features are input into a linear classifier to obtain the predicted scores.
[0034] Specifically, the supervised learning part directly uses the encoder trained in the previous step. First, the data-augmented image is divided into several patches and their positions are encoded. Then, the position-encoded patches are input into the encoder to obtain encoded features. Finally, the encoded features are input into the linear classifier to obtain prediction scores. During training, the classification loss is calculated using the cross-entropy function, and the encoder and linear classifier are trained in a supervised manner by reducing the classification loss.
[0035] Furthermore, when using the defocus blur determination model to determine whether a white blood cell image is defocus blurry, the image is first preprocessed, and the determination threshold for defocus blurry of each type of white blood cell image is statistically determined.
[0036] Specifically, supervised training models for determining blurriness can largely solve the problem of inconsistent image sharpness evaluation metrics for different content. However, two major influencing factors remain: image brightness and cell type. For images with the same content, the higher the brightness value within a certain range, the higher the sharpness evaluation metric predicted by the model. Furthermore, the thresholds for determining whether white blood cell images are out of focus are not the same for different types of images. Therefore, when using a blurriness determination model to determine whether a white blood cell image is out of focus, it is necessary to first perform certain preprocessing on the image and statistically determine the blurriness determination thresholds for each type of white blood cell image.
[0037] The goal of data preprocessing is to ensure consistent brightness across images input to the blur / out detection model, thus avoiding inconsistencies in sharpness evaluation metrics caused by varying brightness. This is achieved by first converting the image to HSV representation and calculating the mean value v of the brightness channel. Then, (140-v) is subtracted from each pixel value in the brightness channel. Finally, the image is converted from HSV to BGR representation. Furthermore, before inputting the image into the model, it needs to be normalized and resized to a specified size to maintain consistency with the training input.
[0038] The statistical threshold is determined by collecting a certain number of test images for each type of white blood cell. These pre-processed images are then input into a blur / defocus judgment model to obtain a sharpness evaluation index. The blur / defocus judgment threshold for that type of white blood cell is determined based on subjective judgment of sharpness. It is important to note that since the sharpness evaluation index output by the model is largely consistent in subjective judgment of sharpness, determining the judgment threshold is significantly easier compared to methods based on sharpness evaluation functions.
[0039] In the above specific embodiments, the method for determining the out-of-focus blur of white blood cell images provided by the present invention obtains the target image to be identified and inputs the target image into a pre-trained out-of-focus blur determination model to obtain the recognition result output by the out-of-focus blur determination model. The out-of-focus blur determination model is obtained through unsupervised and supervised learning using a sample dataset. The sample dataset is constructed based on a large number of collected white blood cell images. The unsupervised learning involves pre-training a pre-constructed machine learning network using unlabeled white blood cell image samples to obtain an initial model. The supervised learning involves training the initial model using white blood cell image samples labeled with blur levels to obtain the out-of-focus blur determination model.
[0040] Thus, the method provided by this invention adopts a combination of unsupervised and supervised learning, using a large amount of unlabeled white blood cell data and a small amount of white blood cell data with varying degrees of blur to train the machine learning model, achieving a better effect in determining out-of-focus blur while saving labeling costs.
[0041] In addition to the methods described above, this invention also provides a device for determining the out-of-focus blur of white blood cell images, such as... Figure 2 As shown, the device includes: Image acquisition unit 210 is used to acquire the target image to be identified; The result generation unit 220 is used to input the target image into a pre-trained defocus blur determination model to obtain the recognition result output by the defocus blur determination model. The out-of-focus blur determination model is obtained by performing unsupervised and supervised learning using a sample dataset. The sample dataset is constructed based on a large number of collected white blood cell images. The unsupervised learning involves pre-training a pre-constructed machine learning network using unlabeled white blood cell image samples to obtain an initial model. The supervised learning involves training the initial model using white blood cell image samples labeled with blur levels to obtain the out-of-focus blur determination model.
[0042] In some embodiments, constructing a sample dataset specifically includes: Acquire several sets of white blood cell images at different focal distances within the same field of view; Align each group of white blood cell images; For each group of aligned white blood cell images, white blood cell target detection is performed on the clearest image in each group to obtain white blood cell mini-images, and a sample dataset for unsupervised learning is constructed based on each white blood cell mini-image. We selected small images of white blood cells from certain groups and manually labeled them. Based on the labeled white blood cell images, we constructed a sample dataset for supervised learning.
[0043] In some embodiments, aligning each group of white blood cell images specifically includes: Arrange a group of images from the same field of view in ascending order of focus distance; The sharpest image in a set of images is selected using the energy gradient function of the green channel. The translation matrix required for alignment between two adjacent images is calculated with the sharper image as a reference. The translation matrix required for each image to be aligned with the sharpest image is calculated by summing the translation matrices obtained in the previous step. Align each image with the clearest image and crop all images using the smallest possible boundary range.
[0044] In some embodiments, data augmentation is performed before both supervised and unsupervised learning on the sample dataset.
[0045] In some embodiments, the unsupervised learning includes an encoding phase and a decoding phase; The encoding stage includes: The data-enhanced white blood cell image was divided into several patches and their locations were encoded. Randomly mask the location-encoded tiles; Input the unmasked patches into the encoder to obtain encoded features; The decoding stage includes: The encoding features are supplemented with masking words according to the masking position in the encoding stage; The supplemented coded features are then positionally encoded and input into the decoder to obtain the decoded image.
[0046] In some embodiments, supervised learning employs the encoder obtained from the unsupervised learning, wherein the supervised learning includes: The data-augmented image is divided into several patches and their locations are encoded. The location-encoded patches are input into the encoder to obtain the encoded features; The encoded features are input into a linear classifier to obtain the predicted scores.
[0047] In some embodiments, when using the defocusing blur determination model to determine whether a white blood cell image is defocused, the image is first preprocessed and the defocusing blur determination threshold for each category of white blood cell images is statistically determined.
[0048] In the above specific embodiments, the defocus blur determination device for white blood cell images provided by the present invention obtains a target image to be identified and inputs the target image into a pre-trained defocus blur determination model to obtain the recognition result output by the defocus blur determination model. The defocus blur determination model is obtained through unsupervised and supervised learning using a sample dataset. The sample dataset is constructed based on a large number of collected white blood cell images. The unsupervised learning involves pre-training a pre-constructed machine learning network using unlabeled white blood cell image samples to obtain an initial model. The supervised learning involves training the initial model using white blood cell image samples labeled with blur levels to obtain the defocus blur determination model.
[0049] Thus, the device provided by this invention adopts a combination of unsupervised and supervised learning methods, using a large amount of unlabeled white blood cell data and a small amount of white blood cell data with varying degrees of blur to train the machine learning model, achieving a better effect in determining out-of-focus blur while saving labeling costs.
[0050] In one embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 3 As shown, the computer device includes a processor, memory, and a network interface connected via a system bus. The processor provides computing and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system, computer programs, and model predictions. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The model predictions of the computer device store static and dynamic information data. The network interface of the computer device is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements the steps in the above method embodiments.
[0051] Those skilled in the art will understand that Figure 3 The structure shown is merely a block diagram of a portion of the structure related to the present invention and does not constitute a limitation on the computer device to which the present invention is applied. A specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.
[0052] Corresponding to the above embodiments, this invention also provides a computer storage medium containing one or more program instructions. These one or more program instructions are used to execute the method described above.
[0053] The present invention also provides a computer program product, the computer program product including a computer program, the computer program being stored on a non-transitory computer-readable storage medium, and the computer being able to perform the above-described method when the computer program is executed by a processor.
[0054] In this embodiment of the invention, the processor can be an integrated circuit chip with signal processing capabilities. The processor can be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.
[0055] The various methods, steps, and logic diagrams disclosed in the embodiments of this invention can be implemented or executed. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the methods disclosed in the embodiments of this invention can be directly implemented by a hardware decoding processor, or implemented by a combination of hardware and software modules in the decoding processor. The software modules can reside in random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or other mature storage media in the art. The processor reads information from the storage medium and, in conjunction with its hardware, completes the steps of the above methods.
[0056] The storage medium can be memory, such as volatile memory or non-volatile memory, or may include both volatile and non-volatile memory.
[0057] Among them, non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory.
[0058] Volatile memory can be random access memory (RAM), which is used as an external cache. By way of example, but not limitation, many forms of RAM are available, such as static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDRSDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous linked dynamic random access memory (SLDRAM), and direct memory bus RAM (DRRAM).
[0059] The storage media described in the embodiments of the present invention are intended to include, but are not limited to, these and any other suitable types of memory.
[0060] Those skilled in the art will recognize that, in one or more of the examples above, the functions described in this invention can be implemented using a combination of hardware and software. When applied as software, the corresponding functions can be stored in a computer-readable medium or transmitted as one or more instructions or code on a computer-readable medium. Computer-readable media include computer storage media and communication media, wherein communication media include any medium that facilitates the transmission of computer programs from one place to another. Storage media can be any available medium accessible to general-purpose or special-purpose computers.
[0061] The above specific embodiments further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made on the basis of the technical solution of the present invention should be included within the scope of protection of the present invention.
Claims
1. A method for determining out-of-focus blur in white blood cell images, characterized in that, The method includes: Acquire the target image to be identified; The target image is input into a pre-trained blur detection model to obtain the recognition result output by the blur detection model. The out-of-focus blur determination model is obtained by performing unsupervised and supervised learning using a sample dataset. The sample dataset is constructed based on a large number of collected white blood cell images. The unsupervised learning involves pre-training a pre-constructed machine learning network using unlabeled white blood cell image samples to obtain an initial model. The supervised learning involves training the initial model using white blood cell image samples labeled with blur levels to obtain the out-of-focus blur determination model.
2. The method for determining the out-of-focus blur of a white blood cell image according to claim 1, characterized in that, The construction of the sample dataset specifically includes: Acquire several sets of white blood cell images at different focal distances within the same field of view; Align each group of white blood cell images; For each group of aligned white blood cell images, white blood cell target detection is performed on the clearest image in each group to obtain white blood cell mini-images, and a sample dataset for unsupervised learning is constructed based on each white blood cell mini-image. We selected small images of white blood cells from certain groups and manually labeled them. Based on the labeled white blood cell images, we constructed a sample dataset for supervised learning.
3. The method for determining the out-of-focus blur of leukocyte images according to claim 2, characterized in that, Alignment was performed on each group of white blood cell images, specifically including: Arrange a group of images from the same field of view in ascending order of focus distance; The sharpest image in a set of images is selected using the energy gradient function of the green channel. The translation matrix required for alignment between two adjacent images is calculated with the sharper image as a reference. The translation matrix required for each image to be aligned with the sharpest image is calculated by summing the translation matrices obtained in the previous step. Align each image with the clearest image and crop all images using the smallest possible boundary range.
4. The method for determining the out-of-focus blur of a white blood cell image according to claim 2, characterized in that, Data augmentation was performed on the sample dataset before both supervised and unsupervised learning.
5. The method for determining the out-of-focus blur of leukocyte images according to any one of claims 1-4, characterized in that, The unsupervised learning includes an encoding phase and a decoding phase; The encoding stage includes: The data-enhanced white blood cell image was divided into several patches and their locations were encoded. Randomly mask the location-encoded tiles; Input the unmasked patches into the encoder to obtain encoded features; The decoding stage includes: The encoding features are supplemented with masking words according to the masking position in the encoding stage; The supplemented coded features are then positionally encoded and input into the decoder to obtain the decoded image.
6. The method for determining the out-of-focus blur of a white blood cell image according to claim 5, characterized in that, Supervised learning employs the encoder obtained from the unsupervised learning, wherein the supervised learning includes: The data-augmented image is divided into several patches and their locations are encoded. The location-encoded patches are input into the encoder to obtain the encoded features; The encoded features are input into a linear classifier to obtain the predicted scores.
7. The method for determining the out-of-focus blur of a white blood cell image according to claim 1, characterized in that, When using the blur-out determination model to determine whether a white blood cell image is blur-out, the image is first preprocessed, and the blur-out determination threshold for each category of white blood cell images is statistically determined.
8. A device for determining the out-of-focus blur of a white blood cell image, characterized in that, The device includes: An image acquisition unit is used to acquire the target image to be identified. The result generation unit is used to input the target image into a pre-trained defocus blur determination model to obtain the recognition result output by the defocus blur determination model. The out-of-focus blur determination model is obtained by performing unsupervised and supervised learning using a sample dataset. The sample dataset is constructed based on a large number of collected white blood cell images. The unsupervised learning involves pre-training a pre-constructed machine learning network using unlabeled white blood cell image samples to obtain an initial model. The supervised learning involves training the initial model using white blood cell image samples labeled with blur levels to obtain the out-of-focus blur determination model.
9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps of the method as described in any one of claims 1-7.
10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method as described in any one of claims 1-7.