Multi-scale cervical cell automatic detection method and system

Abstract

The invention relates to the technical field of image recognition, and provides a multi-scale cervical cell automatic detection method and system based on a Cell-YOLOv8 model obtained after YOLOv8 improvement. The method comprises the following steps of: optimizing Bottleneck in a C2F module by adopting a CAFM to form a C2F-CAFM module, and then replacing the C2F module with the C2F-CAFM module in a feature extraction network; a CARAFE up-sampling operator is introduced into a feature fusion network, and then re-parameterization design is performed on the CARAFE up-sampling operator, so that the model can better detect and classify cervical cells, and the improved model effectively overcomes the limitations of more information loss and low feature learning efficiency when YOLOv8 processes fine cell features. While the real-time processing capability is maintained, the capability of capturing cervical cell characteristics is enhanced, so that the detection precision is improved.

Description

Technical Field

[0001] The present disclosure relates to the technical field related to image recognition, and in particular, to a multi-scale cervical cell automatic detection method and system. Background Art

[0002] The statements in this section merely provide background information related to the present disclosure and do not necessarily constitute prior art.

[0003] Cervical cancer is formed by the malignant transformation of normal cells in the upper layer of the cervix. It ranks second in incidence among women aged 15 to 44 worldwide, and about 200,000 women die from cervical cancer each year. Through early screening programs, precancerous and cancerous cell changes in the cervix can be identified, so that cervical cancer can be effectively prevented. However, since a pathological image contains thousands of cells, the visual interpretation of these images is very time-consuming. Even with experienced cytologists, human bias and errors are prone to occur.

[0004] In order to help doctors improve the efficiency and accuracy of diagnosis and make it possible to automatically detect potential cancerous cells from cervical cell images, introducing deep learning technology into medical diagnostic systems has gradually become a research trend in recent years.

[0005] For the processing of cervical images, the current mainstream single-stage method has achieved a near-real-time detection speed while ensuring a certain degree of accuracy. However, it does not perform well when processing multi-scale complex morphological targets. For example, cervical cells in microscope images often have different scales, which makes it extremely difficult to extract universally applicable features from limited sample data, resulting in an inability to handle this variability, resulting in low ability to capture cervical cell features and inability to achieve accurate image recognition. Summary of the invention

[0006] In order to solve the above problems, the present disclosure proposes a multi-scale cervical cell automatic detection method and system, based on the Cell-YOLOv8 model obtained after the improvement of YOLOv8. The improved model effectively overcomes the limitations of YOLOv8 in processing fine cell features, such as large information loss and low feature learning efficiency. While maintaining real-time processing capabilities, it enhances the ability to capture cervical cell features, thereby improving detection accuracy.

[0007] In order to achieve the above objectives, the present disclosure adopts the following technical solutions:

[0008] One or more embodiments provide a multi-scale cervical cell automatic detection method, comprising the following steps:

[0009] Acquire the cervical cell image to be identified and perform preprocessing;

[0010] The preprocessed image is input into the Cell-YOLOv8 network model obtained by improving YOLOv8, and the cervical image detection result is obtained by detection;

[0011] The Cell-YOLOv8 network model is obtained by improving YOLOv8. CAFM is used to optimize the Bottleneck inside the C2F module to form the C2F-CAFM module. The C2F module is replaced with the C2F-CAFM module in the YOLOv8 backbone network. The C2F-CAFM module convolves the features in the cervical image and adjusts the weights of the channels or spaces of the feature map through the attention mechanism to increase the weights of important features and obtain deep features.

[0012] The CARAFE upsampling operator is introduced into the neck network of YOLOv8, and the CARAFE upsampling operator is reparameterized. The CARAFE upsampling operator uses the additional information of the features output by the backbone network to generate more informative upsampling features;

[0013] One or more embodiments provide a multi-scale cervical cell automatic detection system, including:

[0014] Image acquisition module: configured to acquire the cervical cell image to be identified and perform preprocessing;

[0015] Detection module: configured to input the preprocessed image into the Cell-YOLOv8 network model obtained by improving YOLOv8, and perform detection to obtain the cervical image detection result;

[0016] The Cell-YOLOv8 network model is obtained by improving YOLOv8. CAFM is used to optimize the Bottleneck inside the C2F module to form the C2F-CAFM module. The C2F module is replaced with the C2F-CAFM module in the YOLOv8 backbone network. The C2F-CAFM module convolves the features in the cervical image and adjusts the weights of the channels or spaces of the feature map through the attention mechanism to increase the weights of important features and obtain deep features.

[0017] The CARAFE upsampling operator is introduced into the neck network of YOLOv8, and the CARAFE upsampling operator is reparameterized. The CARAFE upsampling operator uses additional information to generate more informative upsampling features for the features output by the backbone network.

[0018] An electronic device comprises a memory and a processor, and computer instructions stored in the memory and executed on the processor. When the computer instructions are executed by the processor, the steps in the above-mentioned multi-scale cervical cell automatic detection method are completed.

[0019] A computer-readable storage medium is used to store computer instructions. When the computer instructions are executed by a processor, the steps in the above-mentioned multi-scale cervical cell automatic detection method are completed.

[0020] Compared with the prior art, the present invention has the following beneficial effects:

[0021] The disclosed improved model of YOLOv8 is obtained as Cell-YOLOv8. The specific improvement is to use CAFM to optimize the Bottleneck inside the C2F module, which forms a C2F-CAFM module. After that, the C2F module is replaced by a C2F-CAFM module in the feature extraction network; the CARAFE upsampling operator is introduced into the feature fusion network, and then the CARAFE upsampling operator is re-parameterized so that the model can better detect and classify cervical cells; the improved Cell-YOLOv8 model effectively overcomes the limitations of YOLOv8 in processing fine cell features, such as more information loss and low feature learning efficiency. Compared with previous technologies, while maintaining real-time processing capabilities, the ability to capture cervical cell features is enhanced, thereby improving detection accuracy.

[0022] The advantages of the present disclosure and the advantages of additional aspects will be described in detail in the following specific embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] The accompanying drawings constituting a part of the present disclosure are used to provide a further understanding of the present disclosure. The exemplary embodiments of the present disclosure and the description thereof are used to explain the present disclosure but do not constitute a limitation of the present disclosure.

[0024] Figure 1 is a flow chart of a training method of the Cell-YOLOv8 network model of Embodiment 1 of the present disclosure;

[0025] Figure 2 It is a side view of the image structure of a portion of a cell data set constructed during the training process of Example 1 of the present disclosure;

[0026] Figure 3 is a network structure diagram of Cell-YOLOv8 of Embodiment 1 of the present disclosure;

[0027] Figure 4 is a schematic structural diagram of a C2F-CAFM module according to Embodiment 1 of the present disclosure;

[0028] Figure 5 is a schematic diagram of the structure of the CARAFE upsampling operator of Embodiment 1 of the present disclosure;

[0029] Figure 6It is a comparison chart of the prediction performance of Cell-YOLOv8 of Example 1 of the present disclosure and the traditional YOLOv8 model. DETAILED DESCRIPTION

[0030] The present disclosure is further described below in conjunction with the accompanying drawings and embodiments.

[0031] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present disclosure. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present disclosure belongs.

[0032] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit exemplary embodiments according to the present disclosure. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprising" and / or "including" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof. It should be noted that, in the absence of conflict, the various embodiments in the present disclosure and the features in the embodiments can be combined with each other. The embodiments will be described in detail below in conjunction with the accompanying drawings.

[0033] Example 1

[0034] In the technical solutions disclosed in one or more embodiments, Figures 1 to 6 As shown, a multi-scale cervical cell automatic detection method includes the following steps:

[0035] Step 1, obtaining the cervical cell image to be identified and performing preprocessing;

[0036] Step 2: Input the preprocessed image into the Cell-YOLOv8 network model obtained by improving YOLOv8 for detection to obtain the cervical image detection result;

[0037] The Cell-YOLOv8 network model is obtained by improving YOLOv8. CAFM is used to optimize the Bottleneck inside the C2F module to form the C2F-CAFM module. The C2F module is replaced with the C2F-CAFM module in the YOLOv8 backbone network. The C2F-CAFM module convolves the features in the cervical image and adjusts the weights of the channels or spaces of the feature map through the attention mechanism to increase the weights of important features and obtain deep features.

[0038] The CARAFE upsampling operator is introduced into the neck network of YOLOv8, and the CARAFE upsampling operator is reparameterized; the CARAFE upsampling operator uses additional information to generate more informative upsampling features for the features output by the backbone network; the additional information includes the encoder kernel size and the upsampling kernel size.

[0039] In this embodiment, the improved model of YOLOv8 is obtained to obtain Cell-YOLOv8. The specific improvement is to use CAFM to optimize the Bottleneck inside the C2F module, which forms a C2F-CAFM module. Then, the C2F module is replaced with a C2F-CAFM module in the feature extraction network; the CARAFE upsampling operator is introduced into the feature fusion network, and then the CARAFE upsampling operator is re-parameterized to enable the model to better detect and classify cervical cells;

[0040] The improved Cell-YOLOv8 model effectively overcomes the limitations of YOLOv8, such as high information loss and low feature learning efficiency when processing fine cell features. Compared with previous technologies, it enhances the ability to capture cervical cell features while maintaining real-time processing capabilities, thereby improving detection accuracy.

[0041] In step 2, improve the YOLOv8 model structure. The specific steps are as follows:

[0042] In the C2F module, the convolution and attention fusion module (CAFM) is used to optimize the neck structure inside the C2F module to form a C2F-CAFM module.

[0043] like Figure 3 As shown in the figure, the neck structure of C2F is optimized. The size and number of channels of the feature map in the input of the module are determined by the convolution operations of the previous layers. The output is the deep features obtained by the C2F-CAFM module after convolution and attention mechanism processing of the features in the cervical image, which are used for subsequent detection tasks.

[0044] In general, the input and output sizes of the improved C2F-CAFM module are the same as those of the C2F module in the original model. What changes is the weight adjustment made by the CAFM module in C2F-CAFM on the channels or spaces of the feature map. Its role is to enhance the model's attention to important features rather than modifying the size of the input feature map.

[0045] As a further technical solution, the C2F-CAFM module includes parallel local and global branches: the self-attention mechanism is used in the global branch to capture a wider range of high-frequency data information; the local branch is used to extract local features for comprehensive noise suppression;

[0046] Optional, such as Figure 4 As shown, the local branch includes a 1×1 convolutional layer, a channel transformation layer (Split), multiple cascaded C2F-CFAM_Bottleneck modules, a Concat module, and a 3×3 convolution; the C2F-CFAM_Bottleneck module includes a CAFM module and a convolutional layer connected in sequence.

[0047] 1×1 convolutional layer to adjust the channel size;

[0048] C2F-CFAM_Bottleneck module, which performs depth-wise separable convolution to reorganize channels;

[0049] The Concat module is used to perform the concatenation operation of the output tensors along the channel dimension;

[0050] Specifically, in the local branch, 1×1 convolution is first used to adjust the channel size. After that, a channel transformation operation is performed to further mix the channel information; a depth-separable convolution is used to reorganize the channel within each group of the channel-changed information; then, the output tensors of each group are connected along the channel dimension to generate a new output tensor; then, 3×3 convolution is used to extract features.

[0051] The local branch output is:

[0052] F conv =W 3×3×3 (CS(W 1×1 (Y)))

[0053] Optionally, the global branch includes a 1×1 convolutional layer, a 3×3 depth convolutional layer, and a CAFM module connected in sequence;

[0054] In the global branch, 1×1 convolution and 3×3 depth convolution are first performed to obtain three tensors of shape H×W×C. Then, the attention map is calculated through the interaction of Q and K of the CAFM module. The output of the global branch is:

[0055]

[0056] Among them, the local branch and the global branch are processed in parallel, and the final output features are:

[0057] F 输出 =F conv +F att

[0058] In some embodiments, a CARAFE upsampling operator is introduced into the feature fusion network of YOLOv8, and the CARAFE upsampling operator is reparameterized. Specifically, as shown in Table 1 "Experimental results of different kup and kencoder values ​​of CARAFE", experiments have shown that different parameters have different effects on the detection of cervical cells, and different parameters are selected for the model based on the experimental results.

[0059] The CARAFE upsampling operator involves two key parameters: kencoder and kup, which represent the encoder kernel size and the upsampling kernel size respectively. Kencoder is mainly responsible for improving the efficiency and accuracy of feature encoding, while k up The key lies in the ability to reconstruct detail features. It is crucial to properly select these parameters, because parameters that are too large or too small may have a negative impact on the efficiency and accuracy of the model. A series of experiments were performed in this embodiment. Keeping other training parameters unchanged, explore the impact of different kencoder values ​​and kup values ​​on the detection effect. The experimental results are shown in Table 1. Analysis of the data found that although a smaller kup reduces the model parameters, its upsampling operator has insufficient detail reconstruction capabilities, resulting in reduced accuracy. A smaller kencoder limits the encoder's perceptual field and affects the model's ability to capture sufficient contextual information. On the contrary, too large kup and kencoder will significantly increase the computational cost and reduce performance.

[0060] Preferably, kup is 5 and kencoder is 7; this setting not only ensures the high accuracy of the model, but also effectively controls the computational complexity.

[0061] The key difference between the CARAFE operator and the original sampling operator is not the input / output shape (the input and output sizes are the same), but the quality and characteristics of the output. CARAFE aims to generate more informative upsampled features by leveraging additional information, while traditional upsamples use simpler interpolation methods.

[0062] In a further technical solution, CARAFE includes a Kernel Prediction Module and a Content-aware Reassembly Module;

[0063] A content-aware reassembly module, including a channel compression module, a dimension expansion module, and a sigmoid normalization module, performs a content-aware reassembly process including sequentially determining corresponding regions of feature maps, performing channel compression, spatial dimension expansion, and normalization on feature maps;

[0064] The kernel prediction module includes a region demarcation module, a position correspondence module, and a channel normalization module. The kernel prediction module performs the following operations: determining the corresponding positions of the regions and the upsampling kernels in turn, performing normalization processing, and obtaining the focus point information X of the regions. (i+n,j+m) ;

[0065] like Figure 5 As shown in formula (1), in the content-aware reorganization module, N(X l ,k) represents the k×k neighborhood of the neighborhood X of the feature map centered at l. The kernel prediction module first compresses the input feature map channel to reduce the subsequent computational load, then expands the channel dimension in the spatial dimension and normalizes it using the sigmoid function;

[0066] The content-aware reconstruction process is shown in formula (2), where The kernel prediction module determines the focus point information. For the target location l′, the content-aware reconstruction module first finds its corresponding region N(X l ,k up ), and use the corresponding position of the region and the upsampling kernel to better focus on the information of relevant points in the local area and generate a feature map with better semantic features;

[0067] ω l′ =Ψ(N(X l ,k encoder ))(1)

[0068]

[0069] In this embodiment, the original sampling operator is replaced, the upsampling operator CARAFE is introduced, and CARAFE is re-parameterized, so that the improved model can effectively restore image details through its unique content-aware function, reduce the information loss of tiny targets (such as cervical cells) caused by downsampling, and no additional learning parameters are required;

[0070] Furthermore, the Cell-YOLOv8 network model obtained after the improvement of YOLOv8 is as follows: Figure 3 As shown, it includes a backbone network (Backbone), a neck network (Neck) and a head network (Head) connected in sequence;

[0071] A backbone network (Backbone), comprising sequentially connected Conv modules, a plurality of Conv modules alternately connected to the first C2F-CAFM module, and an SPPF module;

[0072] A neck network (Neck), comprising a first CARAFE upsampling module, a first Concat module, a first C2F-CAFM module, a second CARAFE upsampling module, a second Concat module and a second C2F-CAFM module connected in sequence, and two groups of Conv modules, Concat modules and C2F-CAFM modules connected in sequence;

[0073] The head network (Head) includes multiple cascaded convolutional layers, which performs convolution processing on the features output by the neck network and then outputs them;

[0074] Specifically, each Conv module includes a convolutional layer (Conv2d), a batch normalization layer (BatchNorm2d), and an activation layer (Silu activation function layer) connected in sequence.

[0075] The SPPF module includes a Conv module, multiple maximum pooling layers, a Concat module, and a Conv module connected in sequence;

[0076] The entire improved model network includes 8 C2F-CAFM modules, 7 Conv modules, 1 SPPF module, 4 Concat modules, and 2 CARAFE upsampling modules.

[0077] Furthermore, the process of training the improved Cell-YOLOv8 network model is also included, including the following:

[0078] Step S1, acquiring cervical cell pathology images to construct a data set;

[0079] Step S1.1) Collect cervical cell pathology images and pre-process them, delete duplicate images and images that do not meet format requirements, and name and store the images in folders.

[0080] Step S1.2) dividing the cervical cell pathology images into a training set, a test set, and a validation set;

[0081] Step S1.3) Mark the cervical cell pathology image to obtain a marked file, and store the marked file and the image file separately

[0082] Specifically, create a folder named Images, create three folders named train, test, and val in the Images folder, and put the cell image training set, test set, and validation set into the train, test, and val folders respectively; create a folder named Labels, create three folders named train, test, and val in the Labels folder, and put the corresponding annotation files into the train, test, and val folders respectively.

[0083] Specifically, the cervical cell dataset (PQCC) used in this embodiment is provided by the pathology quality control center of a hospital. The dataset covers six types of cervical cells, such as Figure 2 As shown, including low-grade squamous intraepithelial lesion (LSIL), atypical squamous cells-undetermined significance (ASC-US), high-grade squamous intraepithelial lesion (HSIL), bacterial hyperproliferation (Cluecell), atypical glandular cells (AGC-NOS) and Candida albicans infection (CAN). The dataset was annotated by three well-trained cytotechnologists and an experienced pathologist to ensure the accuracy and reliability of the diagnosis.

[0084] Step S2, input the image of the data set into the Cell-YOLOv8 network for training;

[0085] Step S2.1) Initialize the parameters of the improved Cell-YOLOv8 network structure;

[0086] Specifically, parameter initialization includes input image size, training batch size, training epoch, optimizer, maximum learning rate (Init_lr), minimum learning rate (Min_lr), etc. Load the set parameters into the configured computer environment and use the annotated images in the train folder for training.

[0087] Step S2.2) After training for several generations, the loss value (Loss) basically converges to a stable value, the iteration ends, and the trained Cell-YOLOv8 network model is obtained; the image data of the validation set is used for verification until the accuracy requirements are met, and the final trained Cell-YOLOv8 network model is obtained.

[0088] As the number of iterations of network model training increases, the gap between the Loss of the training set and the validation set gradually decreases, and the difference between the Loss of the training set and the validation set is not much different. The Loss basically converges to a stable value, indicating that the network model has achieved the expected training effect. The weight file with the smallest train Loss value is selected as the detection weight file for improving the YOLOv8 network structure and stored in the run folder.

[0089] The images in the validation set are tested using the Cell-YOLOv8 network model improved based on the trained Cell-YOLOv8 network model;

[0090] 3.1) Use the trained Cell-YOLOv8 model to detect the cervical cell images in the validation set. If the accuracy does not reach the expected value, continue training.

[0091] 3.2) Perform analytical experiments on reparameterization of the CARAFE upsampling operator to determine the optimal parameters.

[0092] In order to determine the optimal parameter configuration to improve the performance of the Cell-YOLOv8 model, this embodiment performs a series of experiments. Keeping other training parameters unchanged, explore the effects of different kencoder values ​​and kup values ​​on the detection effect. The experimental results are shown in Table 1. Analysis of the data found that although a smaller kup reduces the model parameters, its upsampling operator's detail reconstruction ability is insufficient, resulting in reduced accuracy. A smaller kencoder limits the encoder's perceptual field and affects the model's ability to capture sufficient contextual information. On the contrary, excessive kup and kencoder will significantly increase the computational cost and reduce performance. Comprehensive comparison, kup of 5 and kencoder of 7 are considered to be the optimal configuration of the CARAFE upsampling operator. This setting not only ensures the high accuracy of the model, but also effectively controls the computational complexity.

[0093] Table 1 Experimental results of CARAFE with different kup and kencoder values

[0094]

[0095]

[0096] In the table, kencoder and kup represent the encoder kernel size and upsampling kernel size, respectively; mean average precision (mAP) is a commonly used evaluation index in target detection, which directly reflects the overall performance of the current model; the computational complexity of the model is reflected by the number of floating point operations (FLOPs). The lightweight of the model is evaluated by the number of parameters (Params).

[0097] In order to illustrate the effect of the method of this embodiment, an experiment was conducted, which is specifically described as follows:

[0098] (1) Conduct ablation experiments on Cell-YOLOv8 to verify the effectiveness of introducing the functions of each module.

[0099] To evaluate the role of each module. The experimental results are shown in Table 2. First, the C2F-CAFM module is introduced into YOLOv8, and the Precision, Recall and mAP@0.5 are increased by 1.8%, 1.5% and 4.3% respectively, which effectively improves the detection accuracy of the model. When the CARAFE module is introduced, the advantages of CARAFE in detail enhancement and feature reorganization are shown, and the Precision, Recall and mAP@0.5 are increased by 2.3%, 1.7% and 5.6% respectively. Finally, compared with the baseline model, the Cell-YOLOv8 designed in this embodiment improves Precision, Recall and mAP@0.5 by 7.8%, 5.2% and 13.9% respectively, and the detection effect is better than YOLOv8. Therefore, this embodiment effectively improves the detection performance of the model. The ablation experiment clearly demonstrates the superiority of the model of this embodiment in terms of accuracy and complexity, and proves the feasibility of the scheme designed in this embodiment for cervical cell detection.

[0100] Precision is the proportion of all predicted targets that the model correctly identifies, and recall is the proportion of all actual targets that the model correctly identifies.

[0101] Table 2 Ablation experiment results

[0102]

[0103]

[0104] (2) Comparison of visualization results: Explore the visualization results of the baseline model YOLOv8 and the improved model Cell-YOLOv8 on the self-built dataset, and analyze the detection effect of the Cell-YOLOv8 model on cervical cells of different shapes and sizes.

[0105] The Cell-YOLOv8 model has higher detection accuracy for cervical cells of different shapes and sizes, and the targets not detected by the baseline model are also improved in the improved method. Figure 6 As shown in the figure, the Cell-YOLOv8 model can effectively improve the detection ability of targets of different shapes, effectively solve the problem of unbalanced detection accuracy, and is more suitable for complex cervical cell detection. From the comparison results, the Cell-YOLOv8 model is significantly better than other models in performance and can efficiently handle complex cervical cell detection tasks.

[0106] This example is based on the established cervical cell image dataset. This example comprehensively compares Cell-YOLOv8 with existing mainstream target detection models. The experimental results show that the Cell-YOLOv8 model proposed in this study is superior to other target detection methods and provides technical support for the automatic detection and classification of cervical cells.

[0107] Example 2

[0108] Based on Example 1, this embodiment provides a multi-scale cervical cell automatic detection system, including:

[0109] Image acquisition module: configured to acquire the cervical cell image to be identified and perform preprocessing;

[0110] Detection module: configured to input the preprocessed image into the Cell-YOLOv8 network model obtained by improving YOLOv8, and perform detection to obtain the cervical image detection result;

[0111] The Cell-YOLOv8 network model is obtained by improving YOLOv8. CAFM is used to optimize the Bottleneck inside the C2F module to form the C2F-CAFM module. The C2F module is replaced with the C2F-CAFM module in the YOLOv8 backbone network. The C2F-CAFM module convolves the features in the cervical image and adjusts the weights of the channels or spaces of the feature map through the attention mechanism to increase the weights of important features and obtain deep features.

[0112] The CARAFE upsampling operator is introduced into the neck network of YOLOv8, and the CARAFE upsampling operator is reparameterized; the CARAFE upsampling operator uses additional information to generate more informative upsampling features for the features output by the backbone network.

[0113] It should be noted here that the various modules in this embodiment correspond one-to-one to the various steps in Example 1, and the specific implementation process is the same, which will not be repeated here.

[0114] Example 3

[0115] This embodiment provides an electronic device, including a memory and a processor, and computer instructions stored in the memory and running on the processor. When the computer instructions are run by the processor, the steps of a multi-scale cervical cell automatic detection method in Example 1 are completed.

[0116] Example 4

[0117] This embodiment provides a computer-readable storage medium for storing computer instructions. When the computer instructions are executed by a processor, the steps of the multi-scale cervical cell automatic detection method in Embodiment 1 are completed.

[0118] The electronic devices proposed in the present disclosure can be mobile terminals and non-mobile terminals. Non-mobile terminals include desktop computers, and mobile terminals include smart phones (such as Android phones, IOS phones, etc.), smart glasses, smart watches, smart bracelets, tablet computers, laptops, personal digital assistants, and other mobile Internet devices that can communicate wirelessly.

[0119] It should be understood that in the present disclosure, the processor may be a central processing unit CPU, and the processor may also be other general-purpose processors, digital signal processors DSP, application-specific integrated circuits ASIC, off-the-shelf programmable gate arrays FPGA or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0120] The memory may include a read-only memory and a random access memory, and provide instructions and data to the processor, and a portion of the memory may also include a non-volatile random access memory. For example, the memory may also store information about the device type.

[0121] In the implementation process, each step of the above method can be completed by an integrated logic circuit of hardware in a processor or an instruction in the form of software. The steps of the method disclosed in the present disclosure can be directly embodied as being executed by a hardware processor, or being executed by a combination of hardware and software modules in a processor. The software module can be located in a mature storage medium in the art such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory, or an electrically erasable programmable memory, a register, etc. The storage medium is located in a memory, and the processor reads the information in the memory and completes the steps of the above method in combination with its hardware. To avoid repetition, it is not described in detail here. A person of ordinary skill in the art can appreciate that the units and algorithm steps of each example described in combination with the embodiments disclosed herein can be implemented in electronic hardware or a combination of computer software and electronic hardware. Whether these functions are executed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of the present disclosure.

[0122] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0123] In the several embodiments provided in the present disclosure, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only a division of logical functions. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, and the indirect coupling or communication connection of devices or units can be electrical, mechanical or other forms.

[0124] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present disclosure, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present disclosure. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0125] The above description is only a preferred embodiment of the present disclosure and is not intended to limit the present disclosure. For those skilled in the art, the present disclosure may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present disclosure shall be included in the protection scope of the present disclosure.

[0126] Although the above describes the specific implementation methods of the present disclosure in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present disclosure. Technical personnel in the relevant field should understand that on the basis of the technical solution of the present disclosure, various modifications or variations that can be made by those skilled in the art without creative work are still within the scope of protection of the present disclosure.

Claims

1. A multi-scale cervical cell automatic detection method, characterized in that: The steps include: Acquire the cervical cell image to be identified and perform preprocessing; The preprocessed image is input into the Cell-YOLOv8 network model obtained by improving YOLOv8, and the cervical image detection result is obtained by detection; The Cell-YOLOv8 network model is obtained by improving YOLOv8. CAFM is used to optimize the Bottleneck inside the C2F module to form the C2F-CAFM module. The C2F module is replaced with the C2F-CAFM module in the YOLOv8 backbone network. The C2F-CAFM module convolves the features in the cervical image and adjusts the weights of the channels or spaces of the feature map through the attention mechanism to increase the weights of important features and obtain deep features. The CARAFE upsampling operator is introduced into the neck network of YOLOv8, and the CARAFE upsampling operator is reparameterized; the CARAFE upsampling operator uses additional information to generate more informative upsampling features for the features output by the backbone network.

2. A multi-scale cervical cell automatic detection method as claimed in claim 1, characterized in that: The C2F-CAFM module consists of a local branch and a global branch. The global branch uses a self-attention mechanism to capture high-frequency data information; the local branch is used to extract local features for noise suppression.

3. A multi-scale cervical cell automatic detection method as claimed in claim 2, characterized in that: The local branch includes a 1×1 convolutional layer, a channel transformation layer, multiple cascaded C2F-CFAM_Bottleneck modules, a Concat module, and a 3×3 convolution. The C2F-CFAM_Bottleneck module includes CAFM modules and convolutional layers connected in sequence; 1×1 convolutional layer to adjust the channel size; C2F-CFAM_Bottleneck module, which performs depth-wise separable convolution to reorganize channels; The Concat module is used to perform concatenation operations on the output tensors along the channel dimension.

4. A multi-scale cervical cell automatic detection method as claimed in claim 2, characterized in that: The global branch consists of a 1×1 convolutional layer, a 3×3 deep convolutional layer, and an attention fusion module that are connected sequentially.

5. The multi-scale automatic cervical cell detection method according to claim 1, characterized in that: CARAFE includes a kernel prediction module and a content-aware reconstruction module; A content-aware reorganization module includes a channel compression module, a dimension expansion module, and a sigmoid normalization module. The content-aware reorganization process includes sequentially determining corresponding regions of feature maps, performing channel compression, spatial dimension expansion, and normalization on feature maps. The kernel prediction module includes a region demarcation module, a position correspondence module and a channel normalization module, and performs the following steps: determining the corresponding positions of the regions and the up-sampled kernels in turn, and performing normalization processing to obtain the focus point information of the regions; The additional information is used to generate more informative upsampled features, where the additional information includes the encoder kernel size and the upsampled kernel size.

6. A multi-scale cervical cell automatic detection method as claimed in claim 1, characterized in that: The Cell-YOLOv8 network model includes a backbone network, a neck network, and a head network connected in sequence; A backbone network including sequentially connected Conv modules, a plurality of Conv modules alternately connected to a first C2F-CAFM module, and a SPPF module; A neck network, comprising a first CARAFE upsampling module, a first Concat module, a first C2F-CAFM module, a second CARAFE upsampling module, a second Concat module and a second C2F-CAFM module connected in sequence, and two groups of Conv modules, Concat modules and C2F-CAFM modules connected in sequence; The head network includes multiple cascaded convolutional layers, which performs convolution processing on the features output by the neck network and then outputs them.

7. A multi-scale automatic cervical cell detection method as claimed in claim 6, characterized in that: The Conv module consists of a convolutional layer, a batch normalization layer, and an activation layer connected in sequence; The SPPF module includes a Conv module, multiple maximum pooling layers, a Concat module, and a Conv module connected in sequence.

8. A multi-scale cervical cell automatic detection system, characterized in that: include: Image acquisition module: configured to acquire the cervical cell image to be identified and perform preprocessing; Detection module: configured to input the preprocessed image into the Cell-YOLOv8 network model obtained by improving YOLOv8, and perform detection to obtain the cervical image detection result; The Cell-YOLOv8 network model is obtained by improving YOLOv8. CAFM is used to optimize the Bottleneck inside the C2F module to form the C2F-CAFM module. The C2F module is replaced with the C2F-CAFM module in the YOLOv8 backbone network. The C2F-CAFM module convolves the features in the cervical image and adjusts the weights of the channels or spaces of the feature map through the attention mechanism to increase the weights of important features and obtain deep features. The CARAFE upsampling operator is introduced into the neck network of YOLOv8, and the CARAFE upsampling operator is reparameterized; the CARAFE upsampling operator uses additional information to generate more informative upsampling features for the features output by the backbone network.

9. An electronic device, characterized in that: The method comprises a memory and a processor and computer instructions stored in the memory and executed on the processor. When the computer instructions are executed by the processor, the steps of the multi-scale cervical cell automatic detection method according to any one of claims 1 to 7 are completed.

10. A computer-readable storage medium, characterized in that: Used to store computer instructions, which, when executed by a processor, complete the steps of a multi-scale cervical cell automatic detection method as described in any one of claims 1-7.

Images