Microcirculation hemodynamic lesion analysis method based on pathological image
By preprocessing and target detection of whole-slice renal pathology images, and combining image segmentation network models for lesion region segmentation and lesion quantification, a lightweight discriminator is constructed. This solves the problems of lost diagnostic information and difficult classification of glomerular lesions, and achieves fine-grained classification and accurate analysis of glomerular lesions.
Patent Information
- Application Number
- CN202511600155.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-11-04
- Publication Date
- 2026-03-20
AI Technical Summary
Existing methods for analyzing glomeruli based on pathological images cannot effectively capture and analyze glomeruli with special morphologies, resulting in the loss of diagnostic information. Furthermore, they cannot perform fine-grained classification of glomerular lesions, especially the classification between glomerular sclerosis and segmental sclerosis, which is difficult and lacks a unified discrimination standard.
By preprocessing whole-slice renal pathology images, a target detection network is used to locate the glomeruli and obtain ellipse fitting contours and structural information. The lesion area is segmented and the degree of lesion is quantified by combining an image segmentation network model. Finally, a lightweight glomerular lesion type discriminator is constructed for fine-grained classification.
It achieves complete diagnostic information capture for glomerular lesions, improves the accuracy of lesion classification and the robustness of fine-grained classification, provides a unified lesion discrimination standard, and reduces computational costs.
Smart Images
Figure CN121708583A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of medical image analysis technology, and more specifically, to a method for analyzing microcirculatory hemodynamic lesions based on pathological images. Background Technology
[0002] Kidney diseases such as diabetes and hypertensive nephropathy, leading to renal hyperperfusion and abnormal blood flow redistribution, often cause microcirculatory changes, resulting in glomerular sclerosis and even inducing crescentic lesions. Therefore, observing the glomerular lesion characteristics in renal pathology images is a crucial step in diagnosing renal microcirculatory hemodynamic disorders. Glomeruli appear oval in renal tissue sections, with random numbers and locations. The internal sclerotic and crescentic lesions are complex in shape and have blurred edges. Furthermore, as the sclerotic area increases, the lesion type gradually progresses from segmental sclerosis to glomerular sclerosis, exhibiting a continuous process. Therefore, the lesion types have high similarity, making classification challenging.
[0003] Existing glomerular analysis methods based on pathological images typically use feature extraction methods such as convolutional neural networks to directly extract features from whole-slice pathological images. Then, classification algorithms are used to perform multi-classification tasks to determine whether there are lesions in the glomeruli. However, there are no clear quantitative classification indicators for glomerular sclerosis lesions and segmental sclerosis lesions with high inter-class similarity, which can easily lead to classification errors and ambiguities.
[0004] Analysis reveals the following main shortcomings in existing technologies: 1) The inability to effectively capture and analyze glomeruli with unusual morphologies leads to the loss of diagnostic information. Whole-section pathological images contain a large amount of irrelevant tissue information, requiring the location of glomeruli to be detected and located before further detection and analysis. However, as human biological tissue, glomeruli have a rich variety of morphologies, are numerous, and have complex structural variations. Currently, it is not possible to capture glomeruli with unconventional morphologies, resulting in the loss of some diagnostic information.
[0005] 2) The classification of glomerular lesions is quite complex, and different types of lesions exhibit high visual similarity in their graphic features. Existing classification methods for glomerular lesions are usually simple multi-class networks. For the high inter-class similarity between glomerular sclerosis and segmental sclerosis, the classification labels do not have a unified measure, making it impossible to quantitatively analyze the degree of sclerosis.
[0006] 3) Existing methods cannot unify the criteria for diagnosing glomerular lesions and perform fine-grained classification.
[0007] In summary, further effective microcirculatory hemodynamic lesion analysis schemes based on pathological images are needed to address the difficulty in finely distinguishing the types of glomerular lesions at the microcirculatory end. Summary of the Invention
[0008] The purpose of this invention is to overcome the shortcomings of the prior art and provide a method for analyzing microcirculatory hemodynamic lesions based on pathological images. This method includes the following steps: Preprocess the whole-section renal pathology images to obtain cut images containing glomerular targets; The cut image is input into a target detection network to obtain the regional location and structural information of the glomerular target. For the location and structural information of the target glomerulus, an image segmentation network model is used to segment the lesion area and quantify the degree of lesion inside the glomerulus, obtain the lesion area quantification result, and obtain the identification result of whether there is sclerotic lesion and crescent lesion. Based on the quantification results of the lesion area and the identification results, a discriminator is used to obtain fine-grained classification results of glomerular blood flow tissue lesions.
[0009] Compared with existing technologies, the advantages of this invention are as follows: The microcirculation hemodynamics lesion analysis method based on pathological images first preprocesses the whole-slice renal pathological images, designs a target detection network to locate the glomeruli and obtain the glomerular elliptical fitting contour and structural information, thus solving the problem of excessive irrelevant noise in whole-slice renal pathological images. Next, a glomerular lesion region image segmentation network model is constructed, embedding the spectral and texture information unique to the lesion region into the feature extraction process based on a large image segmentation model. After model training, the lesion region inside the glomerulus is segmented and its area calculated, solving the problem of quantitative description of continuous lesion processes. Finally, a lightweight glomerular lesion type discriminator is established, realizing fine-grained classification, identification, and analysis of glomerular blood flow tissue lesion types based on the lesion region quantification results.
[0010] Other features and advantages of the invention will become clear from the following detailed description of exemplary embodiments of the invention with reference to the accompanying drawings. Attached Figure Description
[0011] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments of the invention and, together with their description, serve to explain the principles of the invention.
[0012] Figure 1 This is a flowchart of a microcirculation hemodynamic lesion analysis method based on pathological images according to an embodiment of the present invention; Figure 2 This is a network framework diagram for glomerular lesion region image segmentation according to an embodiment of the present invention; Figure 3 This is a flowchart of a glomerular lesion type discriminator according to an embodiment of the present invention. Detailed Implementation
[0013] Various exemplary embodiments of the present invention will now be described in detail with reference to the accompanying drawings. It should be noted that, unless otherwise specifically stated, the relative arrangement, numerical expressions, and values of the components and steps set forth in these embodiments do not limit the scope of the invention.
[0014] The following description of at least one exemplary embodiment is merely illustrative and is in no way intended to limit the invention or its application or use.
[0015] Techniques, methods, and equipment known to those skilled in the art may not be discussed in detail, but where appropriate, such techniques, methods, and equipment should be considered part of the specification.
[0016] In all the examples shown and discussed herein, any specific values should be interpreted as merely exemplary and not as limitations. Therefore, other examples of exemplary embodiments may have different values.
[0017] It should be noted that similar labels and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be discussed further in subsequent figures.
[0018] In summary, the microcirculatory hemodynamic lesion analysis method based on pathological images provided by this invention includes: preprocessing whole-slice renal pathological images of the microcirculatory end; locating the glomerulus position and obtaining the glomerular elliptical fitting contour and structural information through a target detection network; constructing an image segmentation network model for the glomerular lesion region; embedding the spectral and texture information unique to the lesion region into the feature extraction process based on a large image segmentation model; and, after model training, effectively segmenting and quantifying the area of the lesion region inside the glomerulus; and constructing a lightweight glomerular lesion type discriminator to achieve fine-grained classification, identification, and analysis of glomerular blood flow tissue lesion types based on the lesion area quantification results.
[0019] Specifically, see Figure 1 As shown, the provided method for microcirculatory hemodynamic lesion analysis based on pathological images includes the following steps: Step S1: Preprocess the whole-section pathological image, use a target detection network to locate the glomerulus, and obtain the glomerular elliptical fitting contour and structural information.
[0020] For example, firstly, a full-slice image reading and processing tool is used, with a sliding window size of 1200×1200 and a sliding step size of 400 as the cutting parameters to ensure that all glomeruli in the full-slice image are captured by at least one cut image. Next, after data augmentation, the image data is input into a glomerular detection network (or target detection network) for detection and localization. This glomerular detection network is also called FSJP-Net (Foreground and Shape Joint Perception Network for Glomerulus Detection). In one embodiment, the feature extraction part of the glomerular detection network includes two functionally distinct feature extraction branches: a foreground perception branch and a shape perception branch. The foreground perception branch is used to establish the foreground-background relationship and generate a foreground vector, specifically enhancing the features of the foreground image region to improve the network's ability to recognize foreground targets. The shape perception branch is based on a deep layer aggregation (DLA) structure, incorporating image edge information and performing feature extraction at different resolutions and receptive field sizes. The two sets of features are fused and the prediction regression partial anchor-free target detection framework is used to predict the center point, major and minor axes, and deflection angle of the glomerular ellipse shape, and finally regress the position and structural information of the glomerular target.
[0021] In summary, the designed whole-slice pathological image preprocessing and glomerular detection and localization method includes: selecting an appropriate sliding window size and step size to cut the whole-slice pathological image into several smaller images; analyzing and filtering the cutting results to select the smaller images containing glomerular targets as the dataset content; and inputting the image data into the glomerular target detection network FSJP-Net. This network deeply fuses the foreground and contour information of the glomerulus extracted by two branches to capture and locate glomeruli of various shapes, ensuring the integrity of diagnostic information. Furthermore, it infers and regresses the center point, major and minor axes, deflection angle, and position information of the glomerular elliptical structure, removing irrelevant noise and reducing the computational cost of subsequent analysis.
[0022] Step S2: Construct an image segmentation network model for glomerular lesion areas, embedding the unique spectral and texture information of the lesion areas into the feature extraction process based on a large image segmentation model. After model training, the lesion areas inside the glomerulus are effectively segmented and their areas are quantified.
[0023] Step S2 involves building an image segmentation network model for glomerular lesion areas to segment lesion areas and quantify the degree of lesions within the glomerulus.
[0024] First, the glomerular region obtained in step S1 is cropped to obtain a glomerular image without noise contamination from other tissue images, and then input into the lesion region image segmentation network model.
[0025] like Figure 2 As shown, the image segmentation network model mainly consists of an image encoding module, an adapter embedding module, and an image decoder. The image encoding module uses the SAM (Segment Anything Model) encoding module as the backbone network. The adapter embedding module embeds the spectral and texture features of the glomerular lesion region after each image encoding layer. The image decoder includes a modified Transformer decoder block and a dynamic mask prediction head, thereby achieving accurate segmentation of the lesion region.
[0026] Specifically, the image coding module uses the coding module of the Segment Anything Model (SAM) as the backbone network. By keeping the weight parameters of the pre-trained SAM image encoder frozen, it can leverage the powerful contextual analysis capabilities brought by the large model trained on massive amounts of data, while reducing the network's reliance on computing power and adapting to the actual medical application environment.
[0027] The adapter embedding module embeds the spectral and texture features of the glomerular lesion region after each image encoding layer. The purpose of applying the adapter embedding module is to utilize image features obtained from the SAM to provide targeted cues for specific downstream image tasks, thereby improving the model's generalization ability. Especially when annotated data is limited, the pre-trained base model and task-aware cues can significantly save model storage. For example, a multi-level adapter embedding was designed based on the SAM encoder structure, adding an adapter after each image encoding layer to achieve the embedding of image-specific feature information.
[0028] Feature embedding comes from two sources: original image feature embedding and texture and spectral information embedding. Original image features are obtained by the SAM encoder. First, they are passed through a learnable linear layer. The slice image encoded by the SAM encoder is projected onto a scale-adjustable feature. ∈ Its projection formula is as follows: (1) in, The original image. The scaling factor is used to adjust the adjustable parameters. This is the coded image without scale adjustment. Spectral and texture information are obtained by modeling the high and low frequency spectra of the image and calculating the gray-level co-occurrence matrix, respectively, and are also processed... The linear layer projects the two types of feature information into spectral features. and texture features Next, the adapter layer containing glomerular spectral and texture feature information is embedded into the feature extraction backbone network in multiple layers, as shown in the following embedding formula: (2) in, For activation function, It is a linear layer used to generate different prompts in each adapter. It is an upper projection layer shared across all adapters. Outputting feature cues enables the network to focus on specific information related to the glomerulus during feature extraction.
[0029] Finally, the SAM decoder, which includes a modified Transformer decoder block and a dynamic mask prediction head, is used for decoding, which can obtain accurate segmentation of the lesion area.
[0030] In summary, the designed image segmentation method for glomerular lesion regions includes: the image encoding module of the segmentation model uses SAM as the backbone network, keeping the weight parameters of the pre-trained SAM image encoder frozen to provide basic image context information; then, an adapter containing glomerular spectral and texture feature information is embedded in multiple layers into the feature extraction backbone network; finally, the SAM decoder is used for decoding to obtain the segmentation contour and relative area of the lesion region.
[0031] Step S3: Construct a lightweight glomerular lesion type discriminator to achieve fine-grained classification, identification and analysis of glomerular blood flow tissue lesion types based on the lesion area quantification results.
[0032] The lesion region mask generated by the upper-level lesion segmentation task is further input into a simple lesion discriminator. The overall process of the discriminator is as follows: First, the presence of sclerosis and crescent lesions is identified through multi-class segmentation of the upper-level lesions. Then, for segmental sclerosis and spherical sclerosis with extremely high inter-class feature similarity, quantitative classification is performed by calculating the area, and finally, fine-grained classification of spherical sclerosis, segmental sclerosis and crescent lesions is obtained.
[0033] See Figure 3 The discriminator workflow is as follows: First, multi-class segmentation of upper-layer lesions identifies the presence of sclerosis and crescentic lesions. For segmental sclerosis and spherical sclerosis with extremely high inter-class feature similarity, quantification is performed by calculating the area, as expressed by the formula: (3) in, This indicates the percentage of glomerular sclerosis. This indicates the area of the glomerular lesion segmentation result. This represents the area of the fitted ellipse frame for the glomerular contour obtained from the regression of the upper-level glomerular detection network. The fine-grained classification process is as follows: (4) Among them, GS indicates spherical sclerosis, SS indicates segmental sclerosis, and NoA indicates that there is currently no sclerosis index. This indicates a manually set hardening threshold. The calculated lesion area percentage is used to determine the final lesion analysis result.
[0034] In summary, the fine-grained classification method for glomerular blood flow tissue lesions generally includes: after multi-category segmentation of upper-level lesions to identify the presence of sclerosis and crescentic lesions, a discrimination result for the presence of sclerotic and crescentic lesions is obtained. Next, samples with sclerotic lesions are further fine-grained classified. For segmental sclerosis and glomerular sclerosis with extremely high inter-class feature similarity, area calculation is used for quantification. The calculated lesion area proportion is then used to determine the final lesion analysis result through threshold judgment.
[0035] In summary, compared with the prior art, the present invention has the following advantages: 1) This invention preprocesses whole-section pathological images using methods such as sliding cutting, inputting the image data into a targeted and efficient glomerular target detection network to pre-locate the glomerular positions and fit the glomerular structure, thereby capturing and locating glomeruli of various morphologies to ensure complete diagnostic information. Furthermore, it infers and regresses the center point, major and minor axes, deflection angle, and positional information of the glomerular elliptical structure, removing irrelevant noise and reducing the computational cost of subsequent analysis.
[0036] 2) This invention combines the foundation of a large image segmentation model with the embedding of glomerular lesion region features to design a targeted image segmentation model, which achieves good edge segmentation and preliminary classification of glomerular sclerosis and crescent lesions, realizes accurate quantification of lesion regions, and improves the accuracy of lesion classification.
[0037] 3) This invention utilizes the overall glomerular area obtained from the target detection network and the lesion area obtained from the lesion segmentation network to calculate the lesion proportion and quantify the lesion degree, providing a basis for lesion identification. The dynamically adjustable threshold also improves the robustness of glomerular lesion classification results from different data sources, ultimately achieving fine-grained classification of segmental glomerular sclerosis, glomerular sclerosis, and crescent glomeruli.
[0038] 4) This invention constructs a lightweight lesion type discriminator. Based on the ratio of the quantified area of the lesion region to the intact area of the glomerulus regressed by the detection network, a dynamic threshold is set, and a unified discrimination standard for glomerular sclerosis lesions is proposed.
[0039] 5) Verification has shown that the present invention can efficiently achieve fine-grained classification, identification and analysis of glomerular blood flow tissue lesions.
[0040] This invention can be a system, method, and / or computer program product. A computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for causing a processor to implement various aspects of the invention.
[0041] Computer-readable storage media can be tangible devices capable of holding and storing instructions for use by an instruction execution device. Computer-readable storage media can be, for example, but not limited to, electrical storage devices, magnetic storage devices, optical storage devices, electromagnetic storage devices, semiconductor storage devices, or any suitable combination thereof. More specific examples (a non-exhaustive list) of computer-readable storage media include: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), static random access memory (SRAM), portable compact disc read-only memory (CD-ROM), digital multifunction disc (DVD), memory sticks, floppy disks, mechanical encoding devices, such as punch cards or recessed protrusions storing instructions thereon, and any suitable combination thereof. The computer-readable storage media used herein are not to be construed as transient signals themselves, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through waveguides or other transmission media (e.g., light pulses through fiber optic cables), or electrical signals transmitted through wires.
[0042] The computer-readable program instructions described herein can be downloaded from computer-readable storage media to various computing / processing devices, or downloaded via a network, such as the Internet, local area network, wide area network, and / or wireless network, to an external computer or external storage device. The network may include copper transmission cables, fiber optic transmission, wireless transmission, routers, firewalls, switches, gateway computers, and / or edge servers. A network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards them to the computer-readable storage media in the respective computing / processing device.
[0043] The computer program instructions used to perform the operations of this invention may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, state setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages such as Smalltalk, C++, Python, etc., and conventional procedural programming languages such as "C" or similar languages. The computer-readable program instructions may be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer may be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or may be connected to an external computer (e.g., via the Internet using an Internet service provider). In some embodiments, electronic circuitry, such as programmable logic circuitry, field-programmable gate arrays (FPGAs), or programmable logic arrays (PLAs), is personalized by utilizing state information from the computer-readable program instructions. This electronic circuitry can execute the computer-readable program instructions to implement various aspects of the invention.
[0044] Various aspects of the present invention are described herein with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer-readable program instructions.
[0045] These computer-readable program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing apparatus to produce a machine such that, when executed by the processor of the computer or other programmable data processing apparatus, they create means for implementing the functions / actions specified in one or more blocks of the flowchart and / or block diagram. These computer-readable program instructions can also be stored in a computer-readable storage medium that causes a computer, programmable data processing apparatus, and / or other device to operate in a particular manner; thus, the computer-readable medium storing the instructions comprises an article of manufacture that includes instructions for implementing aspects of the functions / actions specified in one or more blocks of the flowchart and / or block diagram.
[0046] Computer-readable program instructions may also be loaded onto a computer, other programmable data processing apparatus, or other device to cause a series of operational steps to be performed on the computer, other programmable data processing apparatus, or other device to produce a computer-implemented process, thereby causing the instructions executed on the computer, other programmable data processing apparatus, or other device to perform the functions / actions specified in one or more boxes of a flowchart and / or block diagram.
[0047] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present invention. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of an instruction containing one or more executable instructions for implementing a specified logical function. In some alternative implementations, the functions marked in the blocks may occur in a different order than those marked in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions. It will be known to those skilled in the art that implementation in hardware, implementation in software, and implementation using a combination of software and hardware are equivalent.
[0048] The various embodiments of the present invention have been described above. These descriptions are exemplary and not exhaustive, and are not limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is chosen to best explain the principles, practical application, or technical improvements to the embodiments in the market, or to enable others skilled in the art to understand the embodiments disclosed herein. The scope of the invention is defined by the appended claims.
Claims
1. A method for analyzing microcirculatory hemodynamic lesions based on pathological images, comprising the following steps: Preprocess the whole-section renal pathology images to obtain cut images containing glomerular targets; The cut image is input into a target detection network to obtain the regional location and structural information of the glomerular target. For the location and structural information of the target glomerulus, an image segmentation network model is used to segment the lesion area and quantify the degree of lesion inside the glomerulus, obtain the lesion area quantification result, and obtain the identification result of whether there is sclerotic lesion and crescent lesion. Based on the quantification results of the lesion area and the identification results, a discriminator is used to obtain fine-grained classification results of glomerular blood flow tissue lesions.
2. The method according to claim 1, characterized in that, The target detection network has a foreground perception branch and a shape perception branch. The foreground perception branch establishes the foreground-background relationship and generates the foreground vector, and performs targeted feature enhancement on the foreground image region to obtain the first set of features. The shape perception branch is based on a deep aggregation structure, incorporates image edge information, and extracts features at different resolutions and receptive field sizes to obtain the second set of features. The first and second sets of features are fused, and using anchor-free target detection, the center point, major and minor axes, and deflection angle of the glomerular ellipse are predicted respectively, and finally the regional location information and structural information of the glomerular target are regressed.
3. The method according to claim 1, characterized in that, The image segmentation network model includes an image encoding module, an adapter embedding module, and an image decoder. The image encoding module uses a SAM encoding module as the backbone network. The adapter embedding module is used to embed the spectral and texture features of the glomerular lesion region after each image encoding layer. The image decoder is used to obtain the segmentation of the glomerular lesion region.
4. The method according to claim 3, characterized in that, The adapter embedding module performs feature embedding according to the following process: Through linear layer The slice map encoded by the SAM encoding module is projected onto a scale-adjustable feature. ∈ Its projection formula is as follows: in, The original image. As a scale adjustment factor, This is a coded image without scale adjustment; use The linear layer projects the spectral and textural information of the glomerular lesion region into spectral features. and texture features Furthermore, an adapter containing glomerular spectral and texture features is multi-layered and embedded into the feature extraction backbone network. The embedding formula is as follows: in, For activation function, It is a linear layer used to generate different prompts in each adapter. It is a shared upper projection layer across all adapter embedded modules. This is a feature hint in the output.
5. The method according to claim 1, characterized in that, The fine-grained classification results of the glomerular blood flow tissue lesions are expressed as follows: in: Among them, GS indicates spherical sclerosis, SS indicates segmental sclerosis, and NoA indicates that there is currently no sclerosis index. This indicates the hardening threshold that is set. This indicates the percentage of glomerular sclerosis. This indicates the area of the glomerular lesion segmentation result. This represents the area of the fitted ellipse frame around the glomerular contour.
6. The method according to claim 4, characterized in that, The spectral and texture information of the glomerular lesion region was obtained by modeling the high and low frequency spectra of the image and calculating the gray-level co-occurrence matrix, respectively.
7. The method according to claim 1, characterized in that, The preprocessing of the whole-slice renal pathology image includes: using a whole-slice image reading and processing tool, with a set sliding window size and sliding step size as cutting parameters to perform image cutting processing, so as to ensure that all glomeruli of the whole-slice renal pathology image are captured by at least one cut image.
8. The method according to claim 7, characterized in that, The sliding window size is set to 1200×1200, and the sliding step size is set to 400.
9. A computer-readable storage medium having a computer program stored thereon, wherein, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 8.
10. A computer device comprising a memory and a processor, wherein a computer program capable of running on the processor is stored in the memory, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 8.