Pancreatic focus tissue segmentation method and system based on multi-target recognition
By using a multi-target recognition method, combining local texture and gradient features to extract the foreground region, and utilizing the U-Net network for independent segmentation of intratumoral and peritumoral tissues, the problem of insufficient segmentation efficiency and accuracy in existing technologies is solved, achieving more efficient and accurate segmentation of pancreatic lesions.
Patent Information
- Application Number
- CN202511380988.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-09-25
- Publication Date
- 2026-02-17
AI Technical Summary
Existing pancreatic CT image segmentation methods struggle to focus segmentation attention directly on the foreground region, making them susceptible to interference from the background region, which affects segmentation efficiency and accuracy.
A multi-target recognition-based approach is adopted, which extracts the foreground region through local texture and gradient features, and uses the U-Net network to perform independent segmentation of intratumoral and peritumoral tissues. The segmentation process is optimized by combining a classification model and a loss function.
It improves the segmentation efficiency and accuracy of intratumoral and peritumoral tissues, avoids interference from background areas, and achieves more efficient and accurate segmentation.
Smart Images
Figure CN121544879A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of medical image processing technology, specifically to a method and system for segmenting pancreatic lesions based on multi-target recognition. Background Technology
[0002] Pancreatic ductal adenocarcinoma (PDAC) is one of the most destructive malignancies, and computed tomography (CT) is the preferred imaging modality for its diagnosis and evaluation. In recent years, radiomics, as a technique for analyzing imaging features of regions of interest, has demonstrated excellent performance in predicting tumor biological complexity and intratumoral heterogeneity. Through high-throughput analysis of intratumoral CT images, radiomics has also been used to predict the biological behavior of PDAC tumors. While most research focuses on the primary tumor, recent studies have highlighted the important auxiliary role of the peritumor region in assessing cancer heterogeneity and its importance in prognostic prediction. Therefore, current radiomic analyses of pancreatic ductal adenocarcinoma often combine the analysis of the tumor region and the peritumor region.
[0003] Existing technologies typically involve directly segmenting intratumoral and peritumoral tissues on CT images to obtain the tumor region and the surrounding area. This segmentation method makes it difficult to focus segmentation attention directly on the foreground region and is easily affected by background interference during the segmentation process, thus affecting its segmentation efficiency and accuracy. Summary of the Invention
[0004] The purpose of this invention is to provide a method and system for segmenting pancreatic lesions based on multi-target recognition, so as to solve the technical problem in the prior art that it is difficult to focus the segmentation attention directly on the foreground region and that the segmentation process is easily affected by the background region, thus affecting the segmentation efficiency and accuracy.
[0005] To solve the above-mentioned technical problems, the present invention specifically provides the following technical solution: A method for segmenting pancreatic lesion tissue based on multi-target recognition includes the following steps: Obtain CT images of pancreatic tissue; The foreground region is extracted from the CT image of pancreatic tissue, and the foreground region is classified and identified using a pre-established classification model to obtain the attribute category of the foreground region; Based on the U-Net network corresponding to the attribute category of the foreground region, target segmentation is performed on the foreground region to obtain the multi-target segmentation result of pancreatic tissue.
[0006] As a preferred embodiment of the present invention, the method for extracting the foreground region includes: For each pixel (x,y) in a CT image of pancreatic tissue, define a local window of size N*N centered on it as its neighborhood space N(x,y); Calculate the gradient magnitude of all pixels in the neighborhood space N(x,y), and use it as the local gradient magnitude of pixel (x,y). ,in, , These are the gradient components of the neighborhood space N(x,y) calculated by the Sobel operator in the x and y directions, respectively. Calculate the variance of the intensity of all pixels in the neighborhood space N(x,y), and use it as the local intensity variance of pixel (x,y). , ,in, Let N(x,y) be the mean intensity of all pixels in the neighborhood space N(x,y). For pixels in the neighborhood space N(x,y) The pixel intensity at a given location, where N is the total number of pixels in each single dimension of the neighborhood space N(x,y) in the x,y dimension; Local intensity variance and local gradient magnitude Normalization and weighted combination are performed to obtain the foreground evaluation value for pixel (x,y). ,in, After normalization , After normalization , for The weight, for The weights; Prospect assessment value Compare with a preset threshold, and select the foreground evaluation value that exceeds the preset threshold. The corresponding pixel (x,y) is labeled as the foreground region K(x,y).
[0007] As a preferred embodiment of the present invention, the method for establishing the classification model includes: Multiple pancreatic tissue CT images were acquired, the foreground region was extracted from the pancreatic tissue CT images, and the attribute category labels of the foreground region were marked as ground truth values; The foreground region and the attribute category labels are combined into a dataset, and a classifier is trained to obtain a classification model with the foreground region as input and the attribute categories of the foreground region as output. ,in, Output foreground region for classification model Attribute categories, Foreground area For classifiers.
[0008] As a preferred embodiment of the present invention, the attribute category of the foreground region includes intratumoral tissue of pancreatic tumors. and peritumoral tissue .
[0009] As a preferred embodiment of the present invention, the U-Net network includes a U-Net network for intratumoral tissue segmentation. and the U-Net network for peritumoral tissue segmentation ; in, Output of the U-Net network for intratumoral tissue segmentation Results of intratumoral tissue segmentation; Output of the U-Net network for intratumoral tissue segmentation Results of intratumoral tissue segmentation; Foreground region with attribute category of intratumoral tissue; Foreground region with attribute category of peritumoral tissue; UNet is the U-Net network.
[0010] As a preferred embodiment of the present invention, the loss function of the classification model Including direct losses by category and classification indirect loss ; in, Output foreground region for classification model Attribute categories, Foreground area The truth value of the attribute category, Output of the U-Net network for intratumoral tissue segmentation The results of intratumoral tissue segmentation, Output of the U-Net network for intratumoral tissue segmentation The results of intratumoral tissue segmentation, for The true value of the intratumoral tissue segmentation results. for The true value of the peritumoral tissue segmentation results. This is the cross-entropy operator.
[0011] As a preferred embodiment of the present invention, the method for target segmentation of the foreground region includes: The foreground region output by the classification model The data is input into a U-Net network for intratumoral tissue segmentation to obtain the foreground region. Results of intratumoral tissue segmentation; The foreground region output by the classification model The input is fed into a U-Net network for peritumoral tissue segmentation to obtain the foreground region. The results of the peritumoral tissue segmentation.
[0012] As a preferred embodiment of the present invention, the present invention provides a pancreatic lesion tissue segmentation system based on multi-target recognition, applied to the aforementioned pancreatic lesion tissue segmentation method based on multi-target recognition, the system comprising: The data acquisition unit is used to acquire CT images of pancreatic tissue; The foreground extraction unit is used to extract the foreground region from the CT image of pancreatic tissue and classify and identify the foreground region using a pre-established classification model to obtain the attribute category of the foreground region. The tissue segmentation unit is used to segment the foreground region based on the U-Net network corresponding to the attribute category of the foreground region, so as to obtain the multi-target segmentation result of pancreatic tissue.
[0013] As a preferred embodiment of the present invention, the pre-established classification model in the foreground extraction unit is as follows: ,in, Output foreground region for classification model Attribute categories, Foreground area For classifiers.
[0014] As a preferred embodiment of the present invention, the foreground extraction unit extracts the foreground region using the following method: For each pixel (x,y) in a CT image of pancreatic tissue, define a local window of size N*N centered on it as its neighborhood space N(x,y); Calculate the gradient magnitude of all pixels in the neighborhood space N(x,y), and use it as the local gradient magnitude of pixel (x,y). ,in, , These are the gradient components of the neighborhood space N(x,y) calculated by the Sobel operator in the x and y directions, respectively. Calculate the variance of the intensity of all pixels in the neighborhood space N(x,y), and use it as the local intensity variance of pixel (x,y). , ,in, Let N(x,y) be the mean intensity of all pixels in the neighborhood space N(x,y). For pixels in the neighborhood space N(x,y) The pixel intensity at a given location, where N is the total number of pixels in each single dimension of the neighborhood space N(x,y) in the x,y dimension; Local intensity variance and local gradient magnitude Normalization and weighted combination are performed to obtain the foreground evaluation value for pixel (x,y). ,in, After normalization , After normalization , for The weight, for The weights; Prospect assessment value Compare with a preset threshold, and select the foreground evaluation value that exceeds the preset threshold. The corresponding pixel (x,y) is labeled as the foreground region K(x,y).
[0015] Compared with the prior art, the present invention has the following advantages: This invention extracts and classifies foreground regions, enabling the segmentation of intratumoral and peritumoral tissues to be performed on foreground regions of the corresponding categories, avoiding interference from other regions and improving the segmentation efficiency and accuracy of intratumoral and peritumoral regions. Attached Figure Description
[0016] 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.
[0017] Figure 1 A flowchart of a pancreatic lesion tissue segmentation method based on multi-target recognition provided in an embodiment of the present invention; Figure 2 A block diagram of a pancreatic lesion tissue segmentation system based on multi-target recognition provided in an embodiment of the present invention. Detailed Implementation
[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. 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.
[0019] like Figure 1As shown, this invention provides a method for segmenting pancreatic lesion tissue based on multi-target recognition, comprising the following steps: Obtain CT images of pancreatic tissue; The foreground region is extracted from the CT image of pancreatic tissue, and the foreground region is classified and identified using a pre-established classification model to obtain the attribute category of the foreground region; Based on the U-Net network corresponding to the attribute category of the foreground region, target segmentation is performed on the foreground region to obtain the multi-target segmentation result of pancreatic tissue.
[0020] In order to eliminate background interference, this invention uses features based on local texture and gradient to initially screen all foreground regions that may be intratumoral and peritumoral tissues from the original CT images, excluding large areas of uniform background (such as blank areas, uniform fat, etc.).
[0021] To further eliminate interference from peritumoral tissue in intratumoral tissue segmentation and vice versa, this invention trains a classifier that further subdivides all previously extracted foreground regions into two categories: intratumoral and peritumoral. The foreground regions of the intratumoral category are then segmented using a U-Net network specifically designed for intratumoral tissue segmentation, and the foreground regions of the peritumoral category are also segmented using a U-Net network specifically designed for peritumoral tissue segmentation. This ensures that intratumoral tissue segmentation is performed only on the foreground regions of the intratumoral category, resulting in highly targeted segmentation without requiring the involvement of the entire image, thus improving segmentation efficiency. Simultaneously, it avoids interference from other tissue regions (such as background regions or peritumoral tissue regions), thereby improving segmentation accuracy.
[0022] Similarly, this allows for segmentation of peritumoral tissue only within the foreground region of the peritumoral category, resulting in highly targeted segmentation without the need for the entire image, thus improving segmentation efficiency. At the same time, it avoids interference from other regions (such as background regions or intratumoral tissue regions), thereby improving segmentation accuracy.
[0023] Methods for extracting foreground regions include: For each pixel (x,y) in a CT image of pancreatic tissue, define a local window of size N*N centered on it as its neighborhood space N(x,y); Calculate the gradient magnitude of all pixels in the neighborhood space N(x,y), and use it as the local gradient magnitude of pixel (x,y). ,in, , These are the gradient components of the neighborhood space N(x,y) calculated by the Sobel operator in the x and y directions, respectively. Calculate the variance of the intensity of all pixels in the neighborhood space N(x,y), and use it as the local intensity variance of pixel (x,y). , ,in, Let N(x,y) be the mean intensity of all pixels in the neighborhood space N(x,y). For pixels in the neighborhood space N(x,y) The pixel intensity at a given location, where N is the total number of pixels in each single dimension of the neighborhood space N(x,y) in the x,y dimension; Local intensity variance and local gradient magnitude Normalization and weighted combination are performed to obtain the foreground evaluation value for pixel (x,y). ,in, After normalization , After normalization , for The weight, for The weights; Prospect assessment value Compare with a preset threshold, and select the foreground evaluation value that exceeds the preset threshold. The corresponding pixel (x,y) is labeled as the foreground region K(x,y).
[0024] This invention combines local intensity variance and local gradient magnitude in foreground extraction. Local intensity variance effectively captures texture complexity. For example, tumor tissue and its surrounding infiltrated tissue typically have uneven density and complex texture, resulting in higher variance values; while uniform background, fluid, or fat regions have lower variance values. Local gradient magnitude effectively captures boundary information. For instance, high gradient values are generated at the boundaries between tumor tissue and normal tissue within and around the tumor.
[0025] This invention combines both weighted and edge information, making it more reliable than using either alone and better able to extract potential tissue regions with weak boundaries or atypical textures. This improves the recall rate of foreground region extraction, ensuring that as many real tissue regions as possible are selected, providing sufficient candidate samples for subsequent classification.
[0026] This invention can also perform artifact removal processing on the foreground region after it has been extracted.
[0027] Methods for building classification models include: Multiple pancreatic tissue CT images were acquired, the foreground region was extracted from the pancreatic tissue CT images, and the attribute category labels of the foreground region were marked as ground truth values; The foreground region and attribute category labels are combined into a dataset, and a classifier is trained to obtain a classification model where the foreground region is the input and the attribute categories of the foreground region are the output. ,in, Output foreground region for classification model Attribute categories, Foreground area For classifiers.
[0028] The attribute categories of the foreground region include intratumoral tissue of pancreatic tumors. and peritumoral tissue .
[0029] The U-Net network includes a U-Net network for intratumoral tissue segmentation. and the U-Net network for peritumoral tissue segmentation ; in, Output of the U-Net network for intratumoral tissue segmentation Results of intratumoral tissue segmentation; Output of the U-Net network for intratumoral tissue segmentation Results of intratumoral tissue segmentation; Foreground region with attribute category of intratumoral tissue; Foreground region with attribute category of peritumoral tissue; UNet is the U-Net network.
[0030] This invention uses two independent U-Net models. One U-Net network specifically receives regions classified as intratumoral and is responsible for outputting tumor tissue segmentation maps. The other U-Net network specifically receives regions classified as peritumoral and is responsible for outputting peritumoral tissue segmentation maps. This enables independent segmentation of the two foreground regions, namely peritumoral and intratumoral, avoiding mutual interference and improving the segmentation accuracy of the two foreground regions.
[0031] Loss function of classification model Including direct losses by category and classification indirect loss ; in, Output foreground region for classification model Attribute categories, Foreground area The truth value of the attribute category, Output of the U-Net network for intratumoral tissue segmentation The results of intratumoral tissue segmentation, Output of the U-Net network for intratumoral tissue segmentation The results of intratumoral tissue segmentation, for The true value of the intratumoral tissue segmentation results. for The true value of the peritumoral tissue segmentation results. This is the cross-entropy operator.
[0032] This invention employs a two-part classification loss when training a classification model for classifying foreground regions. The first part is the direct classification loss, which is the error between the classification model output and the true value. This loss is used to ensure the accuracy of the classification model's classification performance.
[0033] The second part is the classification indirect loss, which is the classification effect produced by the classification model output. This loss ensures that the classification model's output enables the U-Net network to achieve the best segmentation accuracy, forcing the classifier to align the feature representation learned with the optimal feature representation required for the segmentation task. This teaches the classifier to more accurately classify regions with blurred boundaries and difficult segmentation, avoiding other misclassifications. This leads to a large segmentation error in the downstream U-Net, which in turn penalizes the classifier, achieving a synergistic effect between foreground region classification and foreground region segmentation.
[0034] Therefore, by combining the two loss functions of the classification model of this invention, a high-precision classification result with the best-fit segmentation network can be obtained.
[0035] Methods for object segmentation in the foreground region include: The foreground region output by the classification model The data is input into a U-Net network for intratumoral tissue segmentation to obtain the foreground region. Results of intratumoral tissue segmentation; The foreground region output by the classification model The input is fed into a U-Net network for peritumoral tissue segmentation to obtain the foreground region. The results of the peritumoral tissue segmentation.
[0036] For the segmentation network (U-Net), the input of this invention is no longer a complete image containing various noises and irrelevant structures, but a semantically singular region (either within the tumor or around the tumor) that has been preliminarily identified. This greatly simplifies the target task that the segmentation network needs to learn, allowing it to focus more on refining the boundaries of a single category and reducing the complexity of the segmentation task.
[0037] The U-Net in this invention, which processes intratumoral regions, can fully learn the heterogeneity within the tumor (such as necrosis and cystic degeneration); while the U-Net processing peritumoral regions can focus on learning subtle features such as blurred fat spaces and vascular encapsulation caused by tumor infiltration. This targeted, independent division of labor is more effective than a single segmentation network simultaneously learning two vastly different peritumoral and intratumoral features on the entire CT image.
[0038] like Figure 2 As shown, this invention provides a pancreatic lesion tissue segmentation system based on multi-target recognition, applied to a pancreatic lesion tissue segmentation method based on multi-target recognition. The system includes: The data acquisition unit is used to acquire CT images of pancreatic tissue; The foreground extraction unit is used to extract the foreground region from the CT image of pancreatic tissue and classify and identify the foreground region using a pre-established classification model to obtain the attribute category of the foreground region. The tissue segmentation unit is used to segment the foreground region based on the U-Net network corresponding to the attribute category of the foreground region, so as to obtain the multi-target segmentation result of pancreatic tissue.
[0039] The pre-built classification model in the foreground extraction unit is ,in, Output foreground region for classification model Attribute categories, Foreground area For classifiers.
[0040] The foreground extraction unit uses the following methods to extract the foreground region: For each pixel (x,y) in a CT image of pancreatic tissue, define a local window of size N*N centered on it as its neighborhood space N(x,y); Calculate the gradient magnitude of all pixels in the neighborhood space N(x,y), and use it as the local gradient magnitude of pixel (x,y). ,in, , These are the gradient components of the neighborhood space N(x,y) calculated by the Sobel operator in the x and y directions, respectively. Calculate the variance of the intensity of all pixels in the neighborhood space N(x,y), and use it as the local intensity variance of pixel (x,y). , ,in, Let N(x,y) be the mean intensity of all pixels in the neighborhood space N(x,y). For pixels in the neighborhood space N(x,y) The pixel intensity at a given location, where N is the total number of pixels in each single dimension of the neighborhood space N(x,y) in the x,y dimension; Local intensity variance and local gradient magnitude Normalization and weighted combination are performed to obtain the foreground evaluation value for pixel (x,y). ,in, After normalization , After normalization , for The weight, for The weights; Prospect assessment value Compare with a preset threshold, and select the foreground evaluation value that exceeds the preset threshold. The corresponding pixel (x,y) is labeled as the foreground region K(x,y).
[0041] This invention extracts and classifies foreground regions, enabling the segmentation of intratumoral and peritumoral tissues to be performed on foreground regions of the corresponding categories, avoiding interference from other regions and improving the segmentation efficiency and accuracy of intratumoral and peritumoral regions.
[0042] The above embodiments are merely exemplary embodiments of this application and are not intended to limit this application. The scope of protection of this application is defined by the claims. Those skilled in the art can make various modifications or equivalent substitutions to this application within its substance and scope of protection, and such modifications or equivalent substitutions should also be considered to fall within the scope of protection of this application.
Claims
1. A method for segmenting pancreatic lesion tissue based on multi-target recognition, characterized in that, Includes the following steps: Obtain CT images of pancreatic tissue; The foreground region is extracted from the CT image of pancreatic tissue, and the foreground region is classified and identified using a pre-established classification model to obtain the attribute category of the foreground region; Based on the U-Net network corresponding to the attribute category of the foreground region, target segmentation is performed on the foreground region to obtain the multi-target segmentation result of pancreatic tissue.
2. The pancreatic lesion tissue segmentation method based on multi-target recognition according to claim 1, characterized in that: The method for extracting the foreground region includes: For each pixel (x,y) in a CT image of pancreatic tissue, define a local window of size N*N centered on it as its neighborhood space N(x,y); Calculate the gradient magnitude of all pixels in the neighborhood space N(x,y), and use it as the local gradient magnitude of pixel (x,y). ,in, , These are the gradient components of the neighborhood space N(x,y) calculated by the Sobel operator in the x and y directions, respectively. Calculate the variance of the intensity of all pixels in the neighborhood space N(x,y), and use it as the local intensity variance of pixel (x,y). , ,in, Let N(x,y) be the mean intensity of all pixels in the neighborhood space N(x,y). For pixels in the neighborhood space N(x,y) The pixel intensity at a given location, where N is the total number of pixels in each single dimension of the neighborhood space N(x,y) in the x,y dimension; Local intensity variance and local gradient magnitude Normalization and weighted combination are performed to obtain the foreground evaluation value for pixel (x,y). ,in, After normalization , After normalization , for The weight, for The weights; Prospect assessment value Compare with a preset threshold, and select the foreground evaluation value that exceeds the preset threshold. The corresponding pixel (x,y) is labeled as the foreground region K(x,y).
3. The pancreatic lesion tissue segmentation method based on multi-target recognition according to claim 2, characterized in that: The method for establishing the classification model includes: Multiple pancreatic tissue CT images were acquired, the foreground region was extracted from the pancreatic tissue CT images, and the attribute category labels of the foreground region were marked as ground truth values; The foreground region and the attribute category labels are combined into a dataset, and a classifier is trained to obtain a classification model with the foreground region as input and the attribute categories of the foreground region as output. ,in, Output foreground region for classification model Attribute categories, Foreground area For classifiers.
4. The pancreatic lesion tissue segmentation method based on multi-target recognition according to claim 3, characterized in that: The attribute categories of the foreground region include intratumoral tissue of pancreatic tumors. and peritumoral tissue .
5. The pancreatic lesion tissue segmentation method based on multi-target recognition according to claim 4, characterized in that: The U-Net network includes a U-Net network for intratumoral tissue segmentation. and the U-Net network for peritumoral tissue segmentation ; in, Output of the U-Net network for intratumoral tissue segmentation Results of intratumoral tissue segmentation; Output of the U-Net network for intratumoral tissue segmentation Results of intratumoral tissue segmentation; Foreground region with attribute category of intratumoral tissue; Foreground region with attribute category of peritumoral tissue; UNet is the U-Net network.
6. The pancreatic lesion tissue segmentation method based on multi-target recognition according to claim 5, characterized in that: The loss function of the classification model Including direct losses by category and classification indirect loss ; in, Output foreground region for classification model Attribute categories, Foreground area The truth value of the attribute category, Output of the U-Net network for intratumoral tissue segmentation The results of intratumoral tissue segmentation, Output of the U-Net network for intratumoral tissue segmentation The results of intratumoral tissue segmentation, for The true value of the intratumoral tissue segmentation results. for The true value of the peritumoral tissue segmentation results. This is the cross-entropy operator.
7. The pancreatic lesion tissue segmentation method based on multi-target recognition according to claim 6, characterized in that: Methods for object segmentation in the foreground region include: The foreground region output by the classification model The data is input into a U-Net network for intratumoral tissue segmentation to obtain the foreground region. Results of intratumoral tissue segmentation; The foreground region output by the classification model The input is fed into a U-Net network for peritumoral tissue segmentation to obtain the foreground region. The results of the peritumoral tissue segmentation.
8. A pancreatic lesion tissue segmentation system based on multi-target recognition, characterized in that, The system, applicable to the pancreatic lesion tissue segmentation method based on multi-target recognition as described in any one of claims 1-7, comprises: The data acquisition unit is used to acquire CT images of pancreatic tissue; The foreground extraction unit is used to extract the foreground region from the CT image of pancreatic tissue and classify and identify the foreground region using a pre-established classification model to obtain the attribute category of the foreground region. The tissue segmentation unit is used to segment the foreground region based on the U-Net network corresponding to the attribute category of the foreground region, so as to obtain the multi-target segmentation result of pancreatic tissue.
9. A pancreatic lesion tissue segmentation system based on multi-target recognition according to claim 8, characterized in that: The pre-established classification model in the foreground extraction unit is ,in, Output foreground region for classification model Attribute categories, Foreground area For classifiers.
10. A pancreatic lesion tissue segmentation system based on multi-target recognition according to claim 8, characterized in that: The foreground extraction unit uses the following method to extract the foreground region: For each pixel (x,y) in a CT image of pancreatic tissue, define a local window of size N*N centered on it as its neighborhood space N(x,y); Calculate the gradient magnitude of all pixels in the neighborhood space N(x,y), and use it as the local gradient magnitude of pixel (x,y). ,in, , These are the gradient components of the neighborhood space N(x,y) calculated by the Sobel operator in the x and y directions, respectively. Calculate the variance of the intensity of all pixels in the neighborhood space N(x,y), and use it as the local intensity variance of pixel (x,y). , ,in, Let N(x,y) be the mean intensity of all pixels in the neighborhood space N(x,y). For pixels in the neighborhood space N(x,y) The pixel intensity at a given location, where N is the total number of pixels in each single dimension of the neighborhood space N(x,y) in the x,y dimension; Local intensity variance and local gradient magnitude Normalization and weighted combination are performed to obtain the foreground evaluation value for pixel (x,y). ,in, After normalization , After normalization , for The weight, for The weights; Prospect assessment value Compare with a preset threshold, and select the foreground evaluation value that exceeds the preset threshold. The corresponding pixel (x,y) is labeled as the foreground region K(x,y).