A white blood cell detection method based on FBDM-YOLOv8s

Abstract

The present application aims to provide a kind of leukocyte detection method based on FBDM-YOLOv8s, comprising the following steps: A, construct improved YOLOv8s neural network model, the improved YOLOv8s neural network model includes: Backbond network, Neck network, Head network;The Backbond network is with multiple Fasternet modules as the main framework, and finally a group of BRA modules is set;B, the improved YOLOv8s neural network model is trained, and the trained improved YOLOv8s neural network model is obtained;C, the leukocyte image to be detected is introduced into the trained improved YOLOv8s neural network model, and is sequentially processed through Backbond network, Neck network, Head network, to obtain the image after detection.The method of the present application can improve the precision of edge detection while optimizing the model parameters, speed up the calculation speed, and better solve the problem of incomplete image cells and two cells overlapping each other.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of computer technology and medical image analysis, and particularly relates to a white blood cell detection method based on FBDM-YOLOv8s. BACKGROUND

[0002] Cells are also commonly known as immune cells, which help the body resist infectious diseases and foreign pathogens as part of the immune system. In the research of blood detection, accurate recognition of white blood cells can provide important information for blood analysis. Using computer-aided technology to effectively detect white blood cells in blood can help pathologists more accurately diagnose various blood diseases. Normal white blood cells are divided into five categories: neutrophils, eosinophils, basophils, lymphocytes and monocytes. The difference between different cells, the small number of white blood cells, and the interference of factors such as difficulty in finding make it a challenging task to accurately recognize and detect normal white blood cells.

[0003] At present, hospitals usually use blood analyzers to count white blood cells. This device mainly uses physical or chemical schemes for counting, but it has certain limitations and can only count a small number of white blood cells. For the diagnosis of major dangerous diseases, bone marrow smear cell morphology examination is usually required, in addition, blood analyzers can only be used for counting and cannot present cell images under a microscope for further judgment by clinicians. In contrast, manual counting methods are time-consuming and labor-intensive, and the results of different test physicians may differ. Therefore, it is necessary to improve the automation level of blood microscopy to better assist doctors in diagnosing major blood diseases.

[0004] With the rapid development of computer technology, deep learning has made breakthrough progress in the field of computer vision. More and more scholars use deep learning methods to classify and detect cells. Compared with traditional machine learning methods, deep learning methods can simplify the image processing and feature selection process, making them more automated. However, for white blood cell testing in blood, one of the challenges is that there are too many types of cells and the differences between different types of cells are not great, so how to accurately classify multiple types of white blood cells is of great significance. SUMMARY

[0005] The present application aims to provide a white blood cell detection method based on FBDM-YOLOv8s, which can optimize model parameters while improving edge detection accuracy, speed up computation, and better solve the problem of incomplete image cells and overlapping of two cells.

[0006] The technical solution of the present application is as follows:

[0007] The white blood cell detection method based on the FBDM-YOLOv8s comprises the following steps:

[0008] A, constructing an improved YOLOv8s neural network model, wherein the improved YOLOv8s neural network model comprises a Backbond network, a Neck network and a Head network;

[0009] The Backbond network takes a plurality of Fasternet modules as a main framework, and a group of BRA modules is arranged at the end;

[0010] B, training the improved YOLOv8s neural network model to obtain a trained improved YOLOv8s neural network model;

[0011] C, introducing a white blood cell image to be detected into the trained improved YOLOv8s neural network model, and sequentially processing the white blood cell image through the Backbond network, the Neck network and the Head network to obtain a detected image;

[0012] The processing process in the Fasternet module is as follows:

[0013] The input result is sequentially processed through a 3*3 PCpnv layer, a 1*1 convolution, a BN layer, a ReLU function and a 1*1 convolution, and then the obtained result is added to the input result to obtain an output result.

[0014] The process of training the improved YOLOv8s neural network model in step B is as follows:

[0015] a, obtaining white blood cell images from a DML-LZWH data set, and pre-processing the obtained white blood cell images to divide them into a training set and a validation set;

[0016] b, training the improved YOLOv8s neural network model by using data of the training set to obtain a trained improved YOLOv8s neural network model;

[0017] c, testing the improved YOLOv8s neural network model by using data of the validation set to obtain a trained improved YOLOv8s neural network model;

[0018] The process of pre-processing the obtained white blood cell images in step a is as follows:

[0019] A certain number of white blood cell microscopic image data is randomly extracted from the DML-LZWH data set, and the image data is randomly divided into a training set and a validation set, and the number of image data in the training set and the validation set is in a ratio of 1:1; all the above images are from the CellaVision-DI-60 system, all the images are 360*367 pixels, are peripheral blood films after Wright-Giemsa staining technology, are photographed by a digital camera connected to a microscope, the microscope uses an achromatic lens with a magnification of 100 times, the image storage format is JPG format, and the labeling of white blood cells is completed by using labelimg under the guidance of a laboratory doctor.

[0020] The processing process in the Backbond network is as follows:

[0021] The input result is sequentially processed by an Embedding function, a Fasternet module, a Merging function, a Fasternet module, a Merging function, and a Fasternet module to obtain a first result, the first result is sequentially processed by a Merging function and a Fasternet module to obtain a second result, and the second result is sequentially processed by a Merging function, a Fasternet module, an SPPF module and a BRA module to obtain a third result; the first result, the second result and the third result are input into the Neck network.

[0022] The processing process in the BRA module is as follows:

[0023] In the first stage, the input result is first divided into S*S regions, each region contains HW / S2 feature vectors, then a query Qr and a key Kr at the region level are obtained in each region through an average pooling operation, the adjacency matrix Ar between regions is calculated using Qr and Kr, finally, the top-k operation is used to select k most relevant regions in each region from Ar to obtain the routing index matrix Ir; then the routing index matrix Ir is used to collect the corresponding key-value pairs from the original key K and value V to obtain Kg and Vg, and finally, the attention function is applied to the collected key-value pairs Kg and Vg to obtain the output result.

[0024] The processing process in the Neck network is as follows:

[0025] The third result is up-sampled, concatenated with the second result by a Concat function, and then processed by the first DSL4 module to obtain a first DSL4 processing result; the first DSL4 processing result is up-sampled, concatenated with the first result by a Concat function, and then processed by the second DSL4 module to obtain a second DSL4 processing result; the second DSL4 processing result is concatenated with the first DSL4 processing result by a Concat function, and then processed by the third DSL4 module to obtain a third DSL4 processing result; the third DSL4 processing result is concatenated with the third result by a Concat function, and then processed by a convolution layer to input a Head network;

[0026] The second DSL4 processing result is processed by a convolution layer to input a Head network.

[0027] The third DSL4 processing result is processed by a convolution layer to input a Head network.

[0028] The processing process in the DSL4 module is as follows:

[0029] The input result is processed by a convolution layer, and then divided into a left part and a right part by a Split function; the right part is sequentially processed by a DBB module and a convolution layer to obtain a first convolution result; the first convolution result is sequentially processed by a DBB module and a convolution layer to obtain a second convolution result; the left part, the right part, the first convolution result, and the second convolution result are concatenated by a Concat function, and then resized by a convolution layer to obtain an output result.

[0030] The processing process in the DBB module is as follows:

[0031] The input result is divided into four paths; the first path is sequentially processed by a 1*1 convolution, a BatchNorm function, and a K*K convolution, and a BatchNorm function to obtain a first path processing result; the second path is sequentially processed by a 1*1 convolution, a BatchNorm function, a K*K convolution, and a BatchNorm function to obtain a second path processing result; the third path is processed by a 1*1 convolution, a BatchNorm function, an average pooling layer, and a BatchNorm function to obtain a third path processing result; and the fourth path is processed by a K*K convolution and a BatchNorm function to obtain a fourth path processing result.

[0032] The first path processing result, the second path processing result, the third path processing result, and the fourth path processing result are added and fused, and then processed by a nonlinear function to obtain an output result.

[0033] The Head network is composed of three MultiSEAM modules.

[0034] The second DSL4 processing result after convolution layer processing is input into the first MultiSEAM module, and a large-size target detection result is output after first Detect detection;

[0035] The third DSL4 processing result after convolution layer processing is input into the second MultiSEAM module, and a medium-size target detection result is output after second Detect detection;

[0036] The third DSL4 processing result and the third result after Concat function splicing are input into the third MultiSEAM module after convolution layer processing, and a small-size target detection result is output after third Detect detection;

[0037] The large-size target detection result, the medium-size target detection result, and the small-size target detection result are fused after multi-scale feature fusion to obtain an output result.

[0038] The processing process in the MultiSEAM module is as follows:

[0039] The input result is divided into three paths, each of which is processed by a GSMM module, and then the processing results of the three GSMM modules and the input result are fused by addition and then processed by average pooling, and then processed by two fully connected networks, and then the obtained results are fused by multiplication to obtain an output result;

[0040] The processing process in the GSMM module is as follows:

[0041] The input result is sequentially processed by a patch embedding module, a GELU function, and a BatchNorm function to obtain a first function processing result; the first function processing result is sequentially processed by a depth separable convolution, a GELU function, and a BatchNorm function to obtain a second function processing result; and the first function processing result and the second function processing result are fused by addition and then sequentially processed by a pointwise convolution, a GELU function, and a BatchNorm function to obtain an output result;

[0042] The patches of the patch embedding modules in the three GSMM modules are respectively set to 6, 7, and 8.

[0043] The full name of the BRA module is BiLevelRoutingAttention.

[0044] The full name of the DSL4 module is DBBNCSPELAN4.

[0045] The full name of the DBB module is Diverse Branch Block.

[0046] The beneficial effects of this invention are as follows:

[0047] 1. This invention utilizes a novel and improved YOLOv8 network model, which possesses the fast learning and rapid detection capabilities of YOLOv8. By replacing the main framework of YOLOv8 with a Fasternet module and introducing the BiLevelRoutingAttention (BRA) attention mechanism, edge detection accuracy can be improved while optimizing model parameters. The introduction of the DBBNCSPELAN4 module enables diverse feature extraction across multiple branches while simultaneously lightweighting the model, reducing the number of parameters, and accelerating computation. The integration of the detection head with the MultiSEAM module further addresses the issues of incomplete image cells and overlapping cells.

[0048] 2. This invention addresses the issue of incomplete white blood cell display in images and the contiguous nature of two white blood cell images by introducing a MultiSEAM module to effectively handle white blood cell detection under occlusion conditions. In contrast, the existing baseline YOLOv8s algorithm struggles to detect them separately, leading to false positives and false negatives. The reasons for false negatives are as follows: Figure 7 As shown.

[0049] 3. The method provided in this invention exhibits high performance in target detection of white blood cells on the DML-LZWH dataset, which can effectively improve detection accuracy and is helpful for the detection of blood diseases.

[0050] 4. The target detection method of the present invention can effectively detect five types of white blood cells, which can improve the work efficiency of doctors and provide a guarantee for the screening and treatment of blood diseases. Attached Figure Description

[0051] Figure 1 This is a schematic diagram of the overall structure of the FBDM-YOLOv8s network model;

[0052] Figure 2 This is a schematic diagram of the Fasternet module.

[0053] Figure 3 This is a schematic diagram of the structure of the BRA module (BiLevelRoutingAttention);

[0054] Figure 4 This is a schematic diagram of the DSL4 module (DBBNCSPELAN4);

[0055] Figure 5 This is a structural diagram of a DBB (Diverse Branch Block) module;

[0056] Figure 6Structure diagram of MultiSEAM module and GSMM module;

[0057] Figure 7 Identification error diagram of prior art baseline YOLOv8s algorithm;

[0058] Figure 8 Leukocyte detection result diagram, from left to right are YOLOv3-tiny, YOLOv5, YOLOv6, YOLOv7, YOLOv8s, YOLOv9, YOLOv10, and FBDM-YOLOv8s algorithm of the application. DETAILED DESCRIPTION

[0059] The application will be described in detail below with reference to the drawings and examples.

[0060] Example 1

[0061] The leukocyte detection method based on FBDM-YOLOv8s of the example comprises the following steps:

[0062] A, constructing an improved YOLOv8s neural network model as shown in Figure 1 The improved YOLOv8s neural network model comprises a Backbond network, a Neck network, and a Head network.

[0063] The Backbond network takes a plurality of Fasternet modules as a main framework, and a set of BRA modules is arranged at the end.

[0064] B, training the improved YOLOv8s neural network model to obtain a trained improved YOLOv8s neural network model;

[0065] C, introducing a leukocyte image to be detected into the trained improved YOLOv8s neural network model, and sequentially processing the leukocyte image through the Backbond network, the Neck network, and the Head network to obtain a detected image.

[0066] As shown in Figure 2 The processing process in the Fasternet module is as follows:

[0067] After the input result sequentially passes through a 3x3 PCpnv layer, a 1x1 convolution, a BN layer, a ReLU function, and a 1x1 convolution, the obtained result is added to the input result to obtain an output result.

[0068] The process of training the improved YOLOv8s neural network model in step B is as follows:

[0069] a. Obtain white blood cell images from the DML-LZWH dataset, preprocess the obtained white blood cell images, and divide them into a training set and a validation set;

[0070] b. Train the improved YOLOv8s neural network model using the data of the training set, and obtain a trained improved YOLOv8s neural network model;

[0071] c. Test the improved YOLOv8s neural network model using the data of the validation set, and obtain a trained improved YOLOv8s neural network model;

[0072] In step a, the preprocessing process of the obtained white blood cell images is as follows:

[0073] A certain number of white blood cell microscope image data is randomly extracted from the DML-LZWH dataset, and the image data is randomly divided into a training set and a validation set. The number of image data in the training set and the validation set is in a ratio of 1:1. All the above images are from the CellaVision-DI-60 system, and all the images are 360x367 pixels. The peripheral blood films after Wright-Giemsa staining technology are photographed by a digital camera connected to a microscope. The microscope uses an achromatic lens with a magnification of 100 times. The image storage format is JPG. The labeling work of white blood cells is completed by labelimg under the guidance of a laboratory doctor.

[0074] The processing process in the Backbond network is as follows:

[0075] The input result is sequentially processed by the Embedding function, the Fasternet module, the Merging function, the Fasternet module, the Merging function, and the Fasternet module to obtain a first result. The first result is sequentially processed by the Merging function and the Fasternet module to obtain a second result. The second result is sequentially processed by the Merging function, the Fasternet module, the SPPF module, and the BRA module to obtain a third result. The first result, the second result, and the third result are respectively input into the Neck network.

[0076] As shown in Figure 3 The processing process in the BRA module is as follows:

[0077] In the first stage, the input result is first divided into SxS regions, each region containing HW / S2 feature vectors, then the query Qr and the key Kr at the region level are obtained by the average pooling operation in each region, the adjacency matrix Ar between regions is calculated using Qr and Kr, and finally the routing index matrix Ir is obtained by selecting the k most relevant regions in each region from Ar through the top-k operation; then the routing index matrix Ir is used to collect the corresponding key-value pairs from the original key K and value V to obtain Kg and Vg, and finally the attention function is applied to the collected key-value pairs Kg and Vg to obtain the output result.

[0078] The processing process in the Neck network is as follows:

[0079] The third result is up-sampled and concatenated with the second result through the Concat function, and then processed through the first DSL4 module to obtain the first DSL4 processing result; the first DSL4 processing result is up-sampled and concatenated with the first result through the Concat function, and then processed through the second DSL4 module to obtain the second DSL4 processing result; the second DSL4 processing result is concatenated with the first DSL4 processing result through the Concat function, and then processed through the third DSL4 module to obtain the third DSL4 processing result; the third DSL4 processing result is concatenated with the third result through the Concat function, and then processed through the convolution layer to input the Head network;

[0080] The second DSL4 processing result is processed through the convolution layer and input into the Head network.

[0081] The third DSL4 processing result is processed through the convolution layer and input into the Head network.

[0082] As shown in Figure 4 , the processing process in the DSL4 module is as follows:

[0083] The input result is equally divided into left and right parts by the Split function after being processed through the convolution layer, and the right part is sequentially processed through the DBB module and the convolution layer to obtain the first convolution result; the first convolution result is sequentially processed through the DBB module and the convolution layer to obtain the second convolution result; the left part, the right part, the first convolution result, and the second convolution result are concatenated through the Concat function, and then resized through the convolution layer to obtain the output result.

[0084] As shown in Figure 5 , the processing process in the DBB module is as follows:

[0085] The input result is divided into four paths, the first path sequentially passes through a 1*1 convolution, a BatchNorm function processing, to obtain a first path processing result; the second path sequentially passes through a 1*1 convolution, a BatchNorm function, a K*K convolution, and a BatchNorm function processing, to obtain a second path processing result; the third path passes through a 1*1 convolution, a BatchNorm function, an average pooling layer, and a BatchNorm function processing, to obtain a third path processing result; and the fourth path passes through a K*K convolution and a BatchNorm function processing, to obtain a fourth path processing result.

[0086] After the first path processing result, the second path processing result, the third path processing result, and the fourth path processing result are added and fused, a nonlinear processing is performed, to obtain an output result.

[0087] The Head network is composed of three MultiSEAM modules.

[0088] The second DSL4 processing result after the convolution layer processing is input into the first MultiSEAM module, and a large-size target detection result is output after the first Detect detection.

[0089] The third DSL4 processing result after the convolution layer processing is input into the second MultiSEAM module, and a medium-size target detection result is output after the second Detect detection.

[0090] The third DSL4 processing result and the third result after the Concat function splicing are input into the third MultiSEAM module after the convolution layer processing, and a small-size target detection result is output after the third Detect detection.

[0091] After the large-size target detection result, the medium-size target detection result, and the small-size target detection result are fused by multi-scale features, an output result is obtained.

[0092] As shown in Figure 6 , the processing process in the MultiSEAM module is as follows:

[0093] The input result is divided into three paths, each path is processed by a GSMM module, then the processing results of the three GSMM modules and the input result are added and fused, and then an average pooling processing is performed, then two fully connected network processing is performed, and finally the obtained result is multiplied and fused, to obtain an output result.

[0094] The processing process in the GSMM module is as follows:

[0095] The input result is sequentially processed by a patch embedding module, a GELU function, and a BatchNorm function to obtain a first function processing result; the first function processing result is sequentially processed by a deep separable convolution, a GELU function, and a BatchNorm function to obtain a second function processing result; and the first function processing result and the second function processing result are added and fused, and then sequentially processed by a point-wise convolution, a GELU function, and a BatchNorm function to obtain an output result.

[0096] In the three GSMM modules, the patches of the patch embedding modules are respectively set to 6, 7, and 8.

[0097] Embodiment 2

[0098] The white blood cell target detection algorithm studied is mainly applied to the field of medical influence and detection, and has high requirements for the real-time performance and accuracy of the algorithm. Therefore, the performance of a one-stage algorithm is compared with that of the algorithm FBDM-YOLOv8s of the present application under the same conditions. The one-stage algorithm mainly uses the YOLO series algorithm, and the algorithm selected for comparison in the present application is the detection network of each version of the YOLO series.

[0099] In this embodiment, the YOLOv3-tiny, YOLOv5s, YOLOv6, YOLOv7, YOLOv9, and YOLOv10 algorithms are compared with the algorithm FBDM-YOLOv8s of the present application using the same data set. The specific comparison is shown in Table 1.

[0100] Table 1 Comparison results of the DMC-Unet network model of the present application and multiple related network models

[0101]

[0102] As can be seen from the data in the table, the YOLOv7 algorithm has poor performance in white blood cell target detection, with an mAP of only 0.951, which cannot meet the actual application requirements in the medical field. The detection accuracy of the YOLOv5 and YOLOv6 algorithms is lower than that of the algorithm of the present application, and the network structure of the YOLOv5 algorithm is simpler, so the detection speed is faster. Compared with the advanced YOLOv9 and YOLOv10 algorithms, although the number of parameters is greatly reduced, the computational amount is greatly increased compared with the algorithm of the present application, which increases the calculation time, and the mAP is lower than that of the algorithm of the present application. Through comparison, the algorithm of the present application performs best in terms of Precision, Recall, mAP, and F1 score, and the mAP is improved by 2.1% compared with the basic YOLOv8s. Although the number of parameters is increased, the computational amount is greatly reduced, so the detection speed is improved.

[0103] (3) From Figure 8As can be seen, the data set of five types of normal white blood cells is evaluated YOLOv3-tiny, YOLOv5s, YOLOv6, YOLOv7, YOLOv8s, YOLOv9, YOLOv10 algorithm and the detection result of the FBDM-YOLOv8s algorithm of the application, using the same parameters for testing, the comparative algorithm has false detection, missed detection, low precision and the like, the FBDM-YOLOv8s algorithm of the application improves the detection precision, and also accurately detects some cells with shielding.

[0104] The above description is a detailed description of the preferred embodiments of the application, but the embodiments are not intended to limit the scope of the patent application of the application, and any equivalent changes or modifications made under the technical spirit of the application should belong to the patent scope covered by the application.

Claims

1. A white blood cell detection method based on FBDM-YOLOv8s, characterized in that, It comprises the following steps: A, constructing an improved YOLOv8s neural network model, wherein the improved YOLOv8s neural network model comprises a Backbond network, a Neck network and a Head network; The Backbond network uses a plurality of Fasternet modules as a main framework, and a set of BRA modules is arranged at the end; B, training the improved YOLOv8s neural network model to obtain a trained improved YOLOv8s neural network model; C, introducing the white blood cell image to be detected into the trained improved YOLOv8s neural network model, and sequentially processing the white blood cell image through the Backbond network, the Neck network and the Head network to obtain a detected image; The processing process in the Fasternet module is as follows: The input result is sequentially processed through a 3*3 PCpnv layer, a 1*1 convolution, a BN layer, a ReLU function, and a 1*1 convolution, and then the obtained result is added and fused with the input result to obtain an output result; The Head network comprises three MultiSEAM modules; The result after the second DSL4 processing result is processed through a convolution layer is input into the first MultiSEAM module, and a large-size target detection result is output through the first Detect detection; The result after the third DSL4 processing result is processed through a convolution layer is input into the second MultiSEAM module, and a medium-size target detection result is output through the second Detect detection; The result after the third DSL4 processing result and the third result are spliced through a Concat function and then processed through a convolution layer is input into the third MultiSEAM module, and a small-size target detection result is output through the third Detect detection; After the large-size target detection result, the medium-size target detection result and the small-size target detection result are fused through multi-scale feature fusion, an output result is obtained; The processing process in the MultiSEAM module is as follows: The input result is divided into three paths, each path is processed through a GSMM module, then the processing results of the three GSMM modules are added and fused with the input result, and then the result is processed through average pooling, and then processed through two fully connected networks, and then the obtained result is multiplied and fused to obtain an output result; The processing process in the GSMM module is as follows: The input result is sequentially processed through a patch embedding module, a GELU function and a BatchNorm function to obtain a first function processing result; the first function processing result is sequentially processed through a depth separable convolution, a GELU function and a BatchNorm function to obtain a second function processing result; the first function processing result and the second function processing result are added and fused, and then sequentially processed through a pointwise convolution, a GELU function and a BatchNorm function to obtain an output result; The patches of the patch embedding modules in the three GSMM modules are respectively 6, 7 and 8.

2. The white blood cell detection method based on the FBDM-YOLOv8s according to claim 1, wherein: The process of training the improved YOLOv8s neural network model in step B is as follows: a. Obtain white blood cell images from the DML-LZWH dataset, and pre-process the obtained white blood cell images to divide them into a training set and a validation set; b. Train the improved YOLOv8s neural network model using the data in the training set to obtain a trained improved YOLOv8s neural network model; c. Test the improved YOLOv8s neural network model using the data in the validation set to obtain a trained improved YOLOv8s neural network model.

3. The white blood cell detection method based on FBDM-YOLOv8s according to claim 2, wherein: In step a, the process of pre-processing the obtained white blood cell images is as follows: A certain number of white blood cell microscope image data is randomly extracted from the DML-LZWH dataset, and the image data is randomly divided into a training set and a validation set, with a ratio of 1:

1. The image data is obtained from the CellaVision-DI-60 system, and the image data is 360x367 pixels, which is a peripheral blood film after Wright-Giemsa staining technology, and is photographed by a digital camera connected to a microscope with a magnification of 100x achromatic lens. The image storage format is JPG format, and the labeling of white blood cells is completed by labelimg under the guidance of a laboratory doctor.

4. The white blood cell detection method based on FBDM-YOLOv8s according to claim 1, wherein: The processing process in the Backbond network is as follows: The input result is processed by the Embedding function, the Fasternet module, the Merging function, the Fasternet module, the Merging function, and the Fasternet module in sequence to obtain a first result, the first result is processed by the Merging function and the Fasternet module in sequence to obtain a second result, and the second result is processed by the Merging function, the Fasternet module, the SPPF module, and the BRA module in sequence to obtain a third result. The first result, the second result, and the third result are input into the Neck network.

5. The white blood cell detection method based on FBDM-YOLOv8s according to claim 4, wherein: The processing process in the BRA module is as follows: In the first stage, the input result is first divided into S×S regions, each containing HW / S² feature vectors, then the query Qrand key Krat the region level are obtained by average pooling operation in each region, the adjacency matrix Arbetween regions is calculated using Qrand Kr, and finally the routing index matrix Iris obtained by selecting the k most relevant regions in each region from Arthrough top-k operation; then the corresponding key-value pairs are collected from the original key K and value V using the routing index matrix Irto obtain Kgand Vg, and finally the attention function is applied on the collected key-value pairs Kgand Vgto obtain the output result. 6.The FBDM-YOLOv8s-based white blood cell detection method of claim 5, wherein: The processing process in the Neck network is as follows: The third result is up-sampled and concatenated with the second result through a Concat function, and then processed through a first DSL4 module to obtain a first DSL4 processing result; The first DSL4 processing result is up-sampled and concatenated with the first result through a Concat function, and then processed through a second DSL4 module to obtain a second DSL4 processing result; The second DSL4 processing result is concatenated with the first DSL4 processing result through a Concat function, and then processed through a third DSL4 module to obtain a third DSL4 processing result; The third DSL4 processing result is concatenated with the third result through a Concat function, and then processed through a convolution layer to input the Head network; The second DSL4 processing result is processed through a convolution layer to input the Head network; The third DSL4 processing result is processed through a convolution layer to input the Head network. 7.The FBDM-YOLOv8s-based white blood cell detection method of claim 6, wherein: The processing process in the DSL4 module is as follows: The input result is equally divided into a left part and a right part through a Split function after being processed through a convolution layer, and the right part is sequentially processed through a DBB module and a convolution layer to obtain a first convolution result; the first convolution result is sequentially processed through a DBB module and a convolution layer to obtain a second convolution result; the left part, the right part, the first convolution result, and the second convolution result are concatenated through a Concat function, and then resized through a convolution layer to obtain an output result. 8.The FBDM-YOLOv8s-based white blood cell detection method of claim 7, wherein: The processing process in the DBB module is as follows: The input result is divided into four paths, the first path is sequentially processed through a 1*1 convolution and a BatchNorm function to obtain a first path processing result; the second path is sequentially processed through a 1*1 convolution, a BatchNorm function, a K*K convolution, and a BatchNorm function to obtain a second path processing result; the third path is processed through a 1*1 convolution, a BatchNorm function, an average pooling layer, and a BatchNorm function to obtain a third path processing result; and the fourth path is processed through a K*K convolution and a BatchNorm function to obtain a fourth path processing result; The first processing result, the second processing result, the third processing result, and the fourth processing result are added and fused, and then are subjected to nonlinear processing to obtain an output result.

Images