Polyp segmentation method and system based on edge detection and multi-scale fusion

Abstract

The invention discloses a polyp segmentation method and system based on edge detection and multi-scale fusion, and the method comprises the steps: obtaining polyp image data, carrying out the feature extraction of the polyp image data through a backbone network, and obtaining primary features; performing convolution connection on the primary features to obtain polyp global multi-scale feature information; edge information of the polyp global multi-scale feature information is further extracted and fused, and polyp edge local feature information and final polyp edge global feature information are obtained; and performing feature interaction and fusion on the final polyp edge global feature information to obtain interaction features and a final polyp feature map. According to the method, edge detection and feature extraction are combined, and the polyp segmentation precision is improved through feature fusion and enhancement; a space and channel attention mechanism is adopted to strengthen features, details are supplemented during up-sampling, and the recognition and segmentation accuracy is improved. Compared with a baseline model, the method has higher segmentation accuracy and generalization, and has important clinical application value.

Description

Technical Field

[0001] The present invention relates to the technical field of medical image segmentation, and in particular to a polyp segmentation method and system based on edge detection and multi-scale fusion. Background Art

[0002] As a key technology in medical image processing, polyp image segmentation plays a vital role in the early diagnosis and treatment of gastrointestinal diseases. With the rapid development and widespread application of endoscopic technology, the number of images acquired by doctors during examinations has increased dramatically, placing higher demands on the rapid and accurate identification and segmentation of polyps. Automated polyp image segmentation technology has emerged as a result. It not only significantly improves the efficiency and accuracy of polyp detection, effectively reduces the workload of doctors, but also greatly increases the success rate of early cancer screening. This technology uses advanced computer vision methods such as deep learning to accurately locate and separate polyps in endoscopic images, providing strong support for clinical diagnosis and treatment.

[0003] Although polyp segmentation methods based on deep learning have made significant progress in recent years, they still face many challenges. First, the color contrast between polyps and surrounding normal tissues is low and the boundaries are blurred. In addition, the noise interference such as motion blur, uneven lighting, and instrument shadows that may exist in endoscopic images makes the polyp segmentation task particularly complicated. Secondly, when processing polyp images, existing deep learning models often find it difficult to fully capture the balance between global features and local details, resulting in problems such as holes, rough edges, or mis-segmentation in the segmentation results. These problems mainly stem from the model's insufficient adaptability to the diversity of polyp shapes and insufficient integration and utilization of global information. Therefore, how to design a more efficient and robust neural network architecture to better detect blurred edges and process complex image information has become the key to improving the accuracy of polyp segmentation. Summary of the Invention

[0004] The purpose of this section is to summarize some aspects of the embodiments of the present invention and briefly introduce some preferred embodiments. Some simplifications or omissions may be made in this section and the abstract and title of this application to avoid obscuring the purpose of this section, the abstract and the title of the invention, and such simplifications or omissions should not be used to limit the scope of the present invention.

[0005] In view of the above existing problems, the present invention is proposed. Therefore, the present invention provides a polyp segmentation method based on edge detection and multi-scale fusion to solve the technical problems existing in the existing polyp segmentation methods, such as insufficient utilization of global features, rough edges and low segmentation accuracy.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0007] In a first aspect, the present invention provides a polyp segmentation method based on edge detection and multi-scale fusion, comprising:

[0008] Acquire polyp image data, and perform feature extraction on the polyp image data using a backbone network to obtain primary features;

[0009] Performing convolution connection on the primary features to obtain global multi-scale feature information of the polyp;

[0010] Further extracting and fusing edge information of the global multi-scale feature information of the polyp to obtain local feature information of the polyp edge and final global feature information of the polyp edge;

[0011] The final polyp edge global feature information is subjected to feature interaction and fusion to obtain interaction features and a final polyp feature map.

[0012] As a preferred solution of the polyp segmentation method based on edge detection and multi-scale fusion described in the present invention, obtaining primary features includes:

[0013] Preprocessing the polyp image data with a fixed resolution of 448*448;

[0014] The preprocessed polyp image data is extracted using a backbone network to obtain primary features, which include low-level features f1, f2, f3 and high-level features f4, f5.

[0015] As a preferred solution of the polyp segmentation method based on edge detection and multi-scale fusion described in the present invention, wherein: performing convolution connection on the primary features to obtain the global multi-scale feature information of the polyp includes:

[0016] Input the f2, f3, f4 and f5 into the multi-scale residual inference module;

[0017] The multi-scale residual inference module consists of asymmetric convolution and residual connection;

[0018] The multi-scale residual inference module extracts and integrates primary features in the horizontal and vertical directions, narrows the semantic difference between the two types of features, and outputs global multi-scale feature information of polyps at four different scales.

[0019] As a preferred solution of the polyp segmentation method based on edge detection and multi-scale fusion described in the present invention, the global multi-scale feature information of polyps at four different scales includes:

[0020] The global multi-scale feature information of polyps at the four different scales is enhanced by using a combination of spatial attention and channel attention mechanisms.

[0021] As a preferred solution of the polyp segmentation method based on edge detection and multi-scale fusion described in the present invention, the local feature information of the polyp edge and the final global feature information of the polyp edge include:

[0022] Inputting the global multi-scale feature information of polyps at four different scales into a polyp edge extraction module;

[0023] The polyp edge extraction module consists of channel slicing processing, convolution, residual connection, splicing operation and deep supervision mechanism;

[0024] The polyp edge extraction module processes the global multi-scale feature information of polyps at four different scales to obtain the local feature information of the polyp edge;

[0025] The local feature information of the polyp edge is fused. The input features and output of the polyp edge extraction module are different each time, and the features obtained in the previous time need to be input into the new polyp edge extraction module again.

[0026] As a preferred solution of the polyp segmentation method based on edge detection and multi-scale fusion described in the present invention, wherein: performing feature interaction and fusion on the final polyp edge global feature information includes:

[0027] The global multi-scale feature information of polyps at four different scales processed by the spatial attention and channel attention mechanisms and the final global feature information of the polyp edge are subjected to convolution and splicing operations to fuse and locate the edge features to obtain the final feature map.

[0028] As a preferred solution of the polyp segmentation method based on edge detection and multi-scale fusion described in the present invention, the feature map includes:

[0029] The final feature map is upsampled to restore the original image size, the feature information in the feature map is transferred, and the detailed feature information is supplemented.

[0030] In a second aspect, the present invention provides a system for polyp segmentation based on edge detection and multi-scale fusion, comprising:

[0031] A data acquisition module, used for acquiring polyp image data;

[0032] A first feature extraction module is used to extract features from the polyp image data using a backbone network to obtain primary features;

[0033] The second feature extraction module is used to perform convolution connection on the primary features to obtain global multi-scale feature information of the polyp;

[0034] The third feature extraction module is used to further extract and fuse the edge information of the global multi-scale feature information of the polyp to obtain local feature information of the polyp edge and final global feature information of the polyp edge;

[0035] The feature map acquisition module is used to perform feature interaction and fusion on the final polyp edge global feature information to obtain interaction features and a final polyp feature map.

[0036] In a third aspect, the present invention provides a computing device, comprising:

[0037] memory and processor;

[0038] The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions. When the computer-executable instructions are executed by the processor, the steps of the polyp segmentation method based on edge detection and multi-scale fusion are implemented.

[0039] In a fourth aspect, the present invention provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the steps of the polyp segmentation method based on edge detection and multi-scale fusion.

[0040] Compared with the existing technology, the present invention has the following beneficial effects: The present invention combines multiple detection and processing strategies to improve the capture of polyp edge features while maintaining accurate recognition of the overall morphology, achieving efficient segmentation and identification of polyps. The proposed method comprehensively captures the edge and internal features of polyps by combining an edge detection module with a feature extraction module. At the same time, to address the problem of edge information loss, the edge feature information output by the edge detection module is fused with the global feature information output by the feature extraction module to achieve spatial, channel, and semantic information alignment, further enhancing the expressive power of the features. In addition, through the feature enhancement module, spatial and attention mechanisms are used to enhance the features, improving their sensitivity and resolution to polyps. Finally, a skip connection mechanism combined with an attention gate is used in the upsampling module to supplement detailed information, improving the network's recognition accuracy and segmentation accuracy for polyps. Compared with existing mainstream baseline models, the polyp segmentation system based on the edge-assisted detection strategy proposed in the present invention has higher segmentation accuracy and generalization, and has important clinical application value. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. Those skilled in the art can also derive other drawings based on these drawings without inventive effort. Among them:

[0042] Figure 1 This is a schematic diagram of the overall process of a polyp segmentation method based on edge detection and multi-scale fusion according to an embodiment of the present invention;

[0043] Figure 2 Schematic diagram of the overall framework of a polyp segmentation method based on edge detection and multi-scale fusion according to an embodiment of the present invention;

[0044] Figure 3 A schematic diagram of a multi-scale residual inference module of a polyp segmentation method based on edge detection and multi-scale fusion according to an embodiment of the present invention;

[0045] Figure 4 This is a schematic diagram of an edge extraction module of a polyp segmentation method based on edge detection and multi-scale fusion according to an embodiment of the present invention;

[0046] Figure 5 Schematic diagram of feature interaction and fusion of a polyp segmentation method based on edge detection and multi-scale fusion according to an embodiment of the present invention. DETAILED DESCRIPTION

[0047] To make the above-mentioned objects, features, and advantages of the present invention more clearly understood, the following detailed description of the specific embodiments of the present invention is given in conjunction with the accompanying drawings. It is obvious that the described embodiments are only part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in this field without creative work should fall within the scope of protection of the present invention.

[0048] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0049] Secondly, the term "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in various places throughout this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive of other embodiments.

[0050] The present invention is described in detail with reference to schematic diagrams. For ease of illustration, cross-sectional views of device structures may be partially enlarged and not to scale when describing embodiments of the present invention. Furthermore, the schematic diagrams are merely illustrative and should not limit the scope of the present invention. Furthermore, in actual production, the three-dimensional dimensions of length, width, and depth should be included.

[0051] Furthermore, in the description of the present invention, it should be noted that the terms "upper, lower, inner, and outer" and other references to orientations or positional relationships are based on the orientations or positional relationships shown in the accompanying drawings and are intended solely to facilitate and simplify the description of the present invention. They are not intended to indicate or imply that the systems or components referred to must have, be constructed, or operate in a specific orientation, and therefore should not be construed as limitations on the present invention. Furthermore, the terms "first, second, or third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0052] In this disclosure, unless otherwise specified or limited, the terms "mounted," "connected," and "connected" should be interpreted broadly. For example, they may refer to fixed, removable, or integral connections. They may also refer to mechanical, electrical, or direct connections, indirect connections through an intermediary, or internal communication between two components. Those skilled in the art will understand the specific meanings of these terms in this disclosure.

[0053] Example 1

[0054] Reference Figure 1 , as an embodiment of the present invention, provides a polyp segmentation method based on edge detection and multi-scale fusion, comprising:

[0055] S100: Acquire polyp image data, and perform feature extraction on the polyp image data using a backbone network to obtain primary features;

[0056] Preferably, the polyp image data is preprocessed with a fixed resolution of 448*448;

[0057] Preferably, the pre-processed polyp image data is extracted using a backbone network to obtain primary features, the primary features including low-level features f1, f2, f3 and high-level features f4, f5;

[0058] In the embodiment of the present application, the backbone network is constructed using Res2Net50, which has the advantages of global feature extraction, multi-scale feature output, and efficient computing;

[0059] Specifically, the process of using the Res2Net50 network to obtain low-level features is as follows: the endoscopic image training set data is input into the backbone network. To capture the global feature information of polyps, the backbone network is constructed using Res2Net50, which has the advantages of global feature extraction, multi-scale feature output, and efficient computing. Res2Net50 further divides the feature channels of each residual block into multiple sub-channel groups and extracts features of different scales for each sub-channel group, so that the same layer can capture information of multiple scales simultaneously.

[0060] Res2Net50 also adopts a hierarchical group convolution strategy, which divides the input features into multiple groups. Each group undergoes an independent convolution operation. This method effectively enhances the feature expression capability while keeping the number of parameters unchanged. Each Res2Net module consists of multiple parallel convolution branches, each branch processes features of different scales, and the features of these branches are fused and reused within the module.

[0061] It should be noted that while maintaining the basic architecture of ResNet50, Res2Net50 introduces the Res2Net module for multi-scale feature representation, making the network perform better in capturing detailed features and global features. This improvement improves the network's feature representation ability and classification performance without significantly increasing the computational complexity. For the polyp segmentation task, the Res2Net50 network structure outputs global feature information at multiple scales for subsequent feature processing to adapt to polyps of different appearances and shapes, as well as the surrounding features of polyps with different tissue structures and vascular distributions, thereby improving the robustness of the network. By introducing multi-scale feature representation, Res2Net50 can capture the detailed features and global features of polyps while ensuring efficient calculation, enhance the model's adaptability to various complex situations, and improve the accuracy and stability of segmentation.

[0062] S102: Perform convolution connection on the primary features to obtain global multi-scale feature information of the polyp;

[0063] Preferably, f2, f3, f4 and f5 are input into the multi-scale residual inference module;

[0064] Preferably, the multi-scale residual inference module consists of asymmetric convolution and residual connection;

[0065] Preferably, the multi-scale residual inference module extracts and integrates primary features in the horizontal and vertical directions to reduce the semantic difference between the two types of features and outputs global multi-scale feature information of polyps at four different scales;

[0066] In the embodiment of the present application, the input features are divided into two steps: first, asymmetric convolution, and then divided into residual convolution and jump connection through 1x1 convolution operation; the residual convolution part features are extracted through a residual block (ResBlock) composed of 1x3, 3x1, 3x3 convolutions and residual connections, and then spliced ​​with the jump connection part features after adjusting the channel dimension and output features through 3x3 convolution; the residual module body divides the gradient information through the cross-stage structure merging, and combines the residual connection to achieve feature fusion under the same semantics, reducing polyp information loss and improving network calculation efficiency. In order to adapt to the changes in polyp appearance, the multi-scale residual inference module is used to output four different scales of polyp global feature information for subsequent feature processing;

[0067] It should be noted that by introducing the multi-scale residual reasoning module, the multi-scale residual reasoning module is as follows Figure 3 As shown in the figure, this module cleverly combines asymmetric convolution, residual connection and cross-stage merging structure, which can not only deeply mine the primary features of polyps in the horizontal and vertical directions, but also significantly reduce the semantic differences between features by integrating feature information of different dimensions. In this process, the parallel processing of residual convolution and jump connection, combined with the carefully designed residual block, not only enhances the network's ability to extract polyp features, but also effectively reduces the loss of key information and improves computational efficiency through cross-stage gradient segmentation and fusion; finally, the multi-scale residual inference module outputs global feature information of polyps at four different scales. These rich feature hierarchies provide a solid data foundation for subsequent polyp recognition and classification.

[0068] S104: further extracting and fusing edge information of the global multi-scale feature information of the polyp to obtain local feature information of the polyp edge and final global feature information of the polyp edge;

[0069] Preferably, the global multi-scale feature information of polyps at four different scales is input into the polyp edge extraction module;

[0070] Preferably, the polyp edge extraction module consists of channel slicing processing, convolution, residual connection, splicing operation and deep supervision mechanism;

[0071] Preferably, the polyp edge extraction module processes the global multi-scale feature information of polyps at four different scales to obtain local feature information of the polyp edge;

[0072] Preferably, the local feature information of the polyp edge is fused, the input features and output of the polyp edge extraction module are different each time, and the features obtained in the previous time need to be input into the new polyp edge extraction module again;

[0073] Specifically, the global multi-scale feature information of polyps is processed by the edge extraction module. The edge extraction module is as follows: Figure 4 As shown, this module is divided into three stages, using different receptive fields in the spatial dimension to match the edge areas of polyps of different sizes and shapes;

[0074] In the first stage, the global feature information of adjacent polyps output in step 2 is fused using dot products to reduce the semantic differences between features through interaction, so that the obtained local features and global features are in the same semantic space;

[0075] In the second stage, the fused features are split according to the channel dimension to form four parallel branches. The first three branches use dilated convolution with dilation rates of 3, 5, and 7, respectively, to enhance the receptive field of the global features and thus obtain detailed local boundary detail features.

[0076] In the third stage, features are aggregated using dot product plus residual connection to obtain edge features that enhance global information. To explore potential edge areas, the fourth branch is fed into a 3x3 convolution block and a 1x1 convolution to increase the number of channels. The features generated after interaction in the second stage are then fused through spatial and channel alignment. The resulting new polyp edge features are bidirectionally aligned in space, channels, and semantics.

[0077] It should be noted that, based on the extraction of global multi-scale features of polyps, the present invention further deepens the processing and fusion of polyp edge details. The edge extraction module uses channel slicing, convolution, residual connection, splicing operation and deep supervision mechanism to effectively improve the extraction accuracy and comprehensiveness of local feature information of polyp edges. Through three-stage progressive processing, the module first uses dot product fusion technology to narrow the semantic differences between global features and ensure the consistency of local features with global features. Subsequently, through the dilated convolution branches with different dilation rates, the module can capture subtle changes in polyp edges under different receptive fields, enhance the receptive field of global features, and make edge details richer. Finally, through the aggregation method of dot product plus residual connection, feature enhancement and fusion are achieved, and an additional branch is innovatively introduced to explore potential edge areas. Through bidirectional alignment in space, channel and semantics, more accurate and comprehensive global feature information of polyp edges is generated.

[0078] S106: Perform feature interaction and fusion on the final global feature information of the polyp edge to obtain the interaction feature and the final polyp feature map; feature interaction and fusion are as follows: Figure 5 As shown;

[0079] Preferably, a combination of spatial attention and channel attention mechanism is used to enhance the global multi-scale feature information of polyps at four different scales in step S102;

[0080] Preferably, the global multi-scale feature information of polyps at four different scales processed by the spatial attention and channel attention mechanisms and the final global feature information of the polyp edge are subjected to convolution and splicing operations to fuse and locate the edge features to obtain the final feature map;

[0081] Preferably, the final feature map is upsampled to restore the original image size, the feature information in the feature map is transferred, and the detailed feature information is supplemented;

[0082] Specifically, a dual-branch self-attention mechanism is used to enhance the features of spatial and channel attention.

[0083] In the first step, global feature information is aggregated in the channel dimension to generate two different spatial feature maps. Then, splicing and convolution operations are performed to obtain the final spatial feature map, which is then fused with global feature information of different scales through convolution operations.

[0084] In the spatial dimension, global pooling is used to extract global channel information. The two resulting channel feature vectors are then additively fused through a fully connected network with shared weights to generate the final channel attention map. This is then fused with the global feature information of polyps at different scales through a convolutional dot product operation to obtain a more comprehensive and accurate feature representation of the polyps.

[0085] The polyp edge feature information is fused with the global feature using convolution and splicing residual operations to strengthen the key feature polyp information and extract more detailed global feature information. The fused feature is reduced in dimension through convolution operation to obtain the final output feature.

[0086] The second step is to introduce a deep supervision mechanism to promote the effective fusion of global polyp features and edge features. This mechanism is introduced in the feature interaction and fusion module and the edge extraction module. Combined with a joint loss function consisting of weighted BEC loss, weighted IOU loss, and weighted DICE loss, the obtained features are trained and optimized through supervision to improve the stability and robustness of the model fusion.

[0087] The final feature map is upsampled and restored to the original image size to achieve the transmission of feature information, supplement the detailed feature information, improve the segmentation accuracy, and achieve segmentation.

[0088] It should be noted that the present invention significantly enhances the interaction and fusion capabilities of the global multi-scale feature information of polyps and the final global feature information of polyp edges by introducing a combined spatial and channel attention mechanism and a deep supervision mechanism. This process not only improves the accuracy and comprehensiveness of feature representation, but also restores the details of the original image through upsampling, ensuring the complete transmission and supplementation of feature information; combined with the supervised training of the joint loss function, the model is significantly improved in terms of fusion stability and robustness, and ultimately achieves high precision and efficiency in polyp segmentation, bringing important technological breakthroughs to the field of medical image analysis.

[0089] The above is a schematic diagram of a polyp segmentation method based on edge detection and multi-scale fusion in this embodiment. It should be noted that the technical solution of the polyp segmentation system based on edge detection and multi-scale fusion and the technical solution of the polyp segmentation method based on edge detection and multi-scale fusion described above are based on the same concept. For details not described in detail in the technical solution of the polyp segmentation system based on edge detection and multi-scale fusion in this embodiment, please refer to the description of the technical solution of the polyp segmentation method based on edge detection and multi-scale fusion described above.

[0090] The polyp segmentation system based on edge detection and multi-scale fusion in this embodiment includes:

[0091] A data acquisition module, used for acquiring polyp image data;

[0092] The first feature extraction module is used to extract features from polyp image data using the backbone network to obtain primary features;

[0093] The second feature extraction module is used to convolute the primary features to obtain the global multi-scale feature information of the polyp;

[0094] The third feature extraction module is used to further extract and fuse the edge information of the global multi-scale feature information of the polyp to obtain the local feature information of the polyp edge and the final global feature information of the polyp edge;

[0095] The feature map acquisition module is used to perform feature interaction and fusion on the final global feature information of the polyp edge to obtain the interaction features and the final polyp feature map.

[0096] This embodiment further provides a computing device suitable for polyp segmentation based on edge detection and multi-scale fusion, including:

[0097] Memory and processor; the memory is used to store computer-executable instructions, and the processor is used to execute computer-executable instructions to implement the polyp segmentation method based on edge detection and multi-scale fusion as proposed in the above embodiment.

[0098] This embodiment further provides a storage medium storing a computer program, which, when executed by a processor, implements the polyp segmentation method based on edge detection and multi-scale fusion as proposed in the above embodiment.

[0099] The storage medium proposed in this embodiment and the polyp segmentation method based on edge detection and multi-scale fusion proposed in the above embodiment belong to the same inventive concept. For technical details not described in detail in this embodiment, please refer to the above embodiment, and this embodiment has the same beneficial effects as the above embodiment.

[0100] Through the above description of the implementation methods, those skilled in the art can clearly understand that the present invention can be implemented with the help of software and necessary general-purpose hardware, and of course it can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art can be embodied in the form of a software product, which can be stored in a computer-readable storage medium, such as a computer's floppy disk, read-only memory (ROM), random access memory (RAM), flash memory (FLASH), hard disk or optical disk, etc., including a number of instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute the methods of various embodiments of the present invention.

[0101] Example 2

[0102] Referring to Table 1, an embodiment of the present invention provides a polyp segmentation method based on edge detection and multi-scale fusion. In order to verify its beneficial effects, comparison results of two schemes are provided.

[0103] In order to prove that the constructed model has good segmentation effect, this model is compared with several polyp segmentation network models such as UNet, UNet++, SFA, PraNet, TPRNet and MSNet to test the segmentation effect of the model algorithm;

[0104] The model algorithm of this embodiment was trained on the training sets of the Kvasir and CVC-ClinicDB datasets; and tested and evaluated on the test sets of ETIS and CVC-ClinicDB. Table 1 shows the evaluation results of the present invention.

[0105] Table 1 Evaluation results of our method on the ETIS and CVC-ClinicDB datasets compared with previous state-of-the-art methods

[0106]

[0107] It should be noted that in Table 1, ↑ indicates that the higher the better, and ↓ indicates that the lower the better.

[0108] As shown in Table 1, taking the ETIS test set as an example, compared with the most current method, mDice is improved by 7.8% and mIoU is improved by 5.1%. The model of the present invention has good generalization ability and has a good ability to segment small polyps.

[0109] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A polyp segmentation method based on edge detection and multi-scale fusion, characterized in that: include: Acquire polyp image data, and perform feature extraction on the polyp image data using a backbone network to acquire primary features; Perform convolution connection on the primary features to obtain global multi-scale feature information of polyps; Further extracting and fusing edge information of the global multi-scale feature information of the polyp to obtain local feature information of the edge of the polyp and final global feature information of the edge of the polyp; The final polyp edge global feature information is subjected to feature interaction and fusion to obtain interaction features and a final polyp feature map.

2. The polyp segmentation method based on edge detection and multi-scale fusion as claimed in claim 1, characterized in that: Obtaining primary features includes: Preprocessing the polyp image data, fixing the resolution to 448*448; The preprocessed polyp image data is extracted using a backbone network to obtain primary features, which include low-level features f1, f2, f3 and high-level features f4, f5.

3. The polyp segmentation method based on edge detection and multi-scale fusion according to claim 1 or 2, characterized in that: The primary features are convolutionally connected to obtain the global multi-scale feature information of the polyp, including: Inputting said f2, f3, f4 and f5 into a multi-scale residual inference module; The multi-scale residual inference module consists of asymmetric convolution and residual connection; The multi-scale residual inference module extracts and integrates primary features in the horizontal and vertical directions, narrows the semantic difference between the two types of features, and outputs global multi-scale feature information of polyps at four different scales.

4. The polyp segmentation method based on edge detection and multi-scale fusion as claimed in claim 3, characterized in that: The global multi-scale feature information of polyps at four different scales includes: The global multi-scale feature information of polyps at the four different scales is enhanced by utilizing the combined mechanism of spatial attention and channel attention.

5. The polyp segmentation method based on edge detection and multi-scale fusion as claimed in claim 4, characterized in that: Obtaining the local feature information of the polyp edge and the final global feature information of the polyp edge includes: Inputting the global multi-scale feature information of polyps at four different scales into a polyp edge extraction module; The polyp edge extraction module consists of channel slicing processing, convolution, residual connection, splicing operation and deep supervision mechanism; The polyp edge extraction module processes the global multi-scale feature information of polyps at four different scales to obtain the local feature information of the polyp edge; The local feature information of the polyp edge is fused, and the input features and outputs of the polyp edge extraction module are different each time, and the features obtained last time need to enter the new polyp edge extraction module again.

6. The polyp segmentation method based on edge detection and multi-scale fusion as claimed in claim 5, characterized in that: The feature interaction and fusion of the final polyp edge global feature information includes: The global multi-scale feature information of polyps at four different scales processed by the spatial attention and channel attention mechanisms and the final global feature information of the polyp edge are subjected to convolution and splicing operations to fuse and locate the edge features to obtain the final feature map.

7. The polyp segmentation method based on edge detection and multi-scale fusion as claimed in claim 6, characterized in that: The feature maps include, The final feature map is upsampled to restore the original image size, the feature information in the feature map is transferred, and the detail feature information is supplemented.

8. A system for polyp segmentation based on edge detection and multi-scale fusion, characterized in that: include, A data acquisition module, used for acquiring polyp image data; The first feature extraction module is used to extract features from the polyp image data using a backbone network to obtain primary features; The second feature extraction module is used to perform convolution connection on the primary features to obtain global multi-scale feature information of polyps; The third feature extraction module is used to further extract and fuse the edge information of the global multi-scale feature information of the polyp to obtain the local feature information of the polyp edge and the final global feature information of the polyp edge; The feature map acquisition module is used to perform feature interaction and fusion on the final polyp edge global feature information to obtain interaction features and a final polyp feature map.

9. An electronic device, comprising: Memory and processor; The memory is used to store computer executable instructions, and the processor is used to execute the computer executable instructions. When the computer executable instructions are executed by the processor, the steps of the polyp segmentation method based on edge detection and multi-scale fusion according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the steps of the polyp segmentation method based on edge detection and multi-scale fusion as described in any one of claims 1 to 7.

Images