Multi-object separation method and device based on split cycle strategy
By employing a vectorization transformation and iterative splitting method based on a splitting loop strategy, the problem of dense target adhesion in image semantic segmentation is solved, achieving high-precision and efficient multi-target separation, which is suitable for complex adhesion scenarios.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- PEKING UNIV
- Filing Date
- 2026-01-09
- Publication Date
- 2026-06-05
AI Technical Summary
In existing image semantic segmentation techniques, densely distributed targets are prone to sticking together, leading to morphological interpretation bias and counting errors. Existing methods, such as morphological opening operations, have low separation accuracy and efficiency when dealing with complex or multi-scale targets.
A method based on a splitting loop strategy is adopted. The binarized raster mask image output by the semantic segmentation model is obtained, and vectorized transformation is performed to generate an initial vector feature queue. The initial vector feature queue is updated through an iterative buffer splitting mechanism to gradually separate the sticky targets.
It improves the separation accuracy and efficiency of densely connected targets in image semantic segmentation results, ensures topological and boundary accuracy, and is suitable for automated separation of complex connected scenes.
Smart Images
Figure CN121505618B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of image processing technology, and in particular to a multi-target separation method and apparatus based on a splitting loop strategy. Background Technology
[0002] In semantic segmentation of images, existing semantic segmentation models typically take the image as input and output a pixel-level raster mask to label target features. However, in practical applications, target features are prone to clustering due to their dense distribution (i.e., appearing as undesirable connectivity of adjacent targets at the raster level), leading to morphological interpretation bias and counting errors, significantly impacting downstream tasks. For example, in the recognition of high-density targets such as impact craters or cells, multiple targets are often misidentified as a single individual, thus interfering with counting and morphological statistics.
[0003] In existing technologies, morphological opening operations are mainly used to break undesirable connections between adjacent targets. However, morphological opening operations have the following drawbacks: First, opening operations rely on structural elements of fixed shapes (such as square or circular templates), which lacks flexibility when dealing with targets with complex shapes, irregular boundaries, or variable orientations. This leads to the topological structure being easily destroyed, resulting in errors such as target adhesion or breakage, and thus low separation accuracy. Second, existing methods usually perform an opening operation only once, which cannot effectively separate multiple complexly adhered or clustered targets. The segmentation effect is limited to the breakage between two targets. Therefore, it cannot achieve complete separation in a single operation for high-density or multi-scale targets (such as geomorphic units, lithological bodies, or spore groups), resulting in low separation efficiency. Summary of the Invention
[0004] This invention provides a multi-target separation method and apparatus based on a splitting loop strategy, which solves the technical problem of low separation accuracy and efficiency of densely connected targets in image semantic segmentation results in the prior art.
[0005] This invention provides a multi-objective separation method based on a split-loop strategy, comprising the following steps:
[0006] Obtain the binary raster mask image output by the semantic segmentation model; wherein, the semantic segmentation model is used to perform semantic segmentation on any image;
[0007] The connected regions in the binarized raster mask image are vectorized to generate an initial vector feature queue;
[0008] The initial vector feature queue is updated to obtain the target vector feature queue.
[0009] According to the present invention, a multi-target separation method based on a split-loop strategy is provided, wherein the connected regions in the binarized raster mask image are vectorized to generate an initial vector feature queue, comprising:
[0010] Each of the multiple connected regions in the binarized raster mask image is converted into multiple vector polygon objects; each connected region corresponds one-to-one with a vector polygon object.
[0011] Delete pixels with a value of 0 from each of the vector polygon objects to obtain the corresponding initial vector features;
[0012] Based on the initial vector features, the initial vector feature queue is generated.
[0013] According to a multi-target separation method based on a split-loop strategy provided by the present invention, updating the initial vector feature queue to obtain a target vector feature queue includes:
[0014] Repeatedly execute the preset iterative steps to split each initial vector element in the initial vector element queue until the area of all initial vector elements is less than or equal to a preset threshold, and use the first intermediate vector element queue obtained in the last iteration as the target vector element queue.
[0015] According to the present invention, a multi-objective separation method based on a split-loop strategy is provided, wherein the iterative steps include:
[0016] Negative buffer erosion is applied to the intermediate vector elements after the previous iteration to obtain the intermediate vector elements after the current iteration.
[0017] Based on the area and topology type of the intermediate vector elements after this round of iteration, the first intermediate vector element queue and the second intermediate vector element queue after this round of iteration are obtained.
[0018] The intermediate vector elements used in the first round of iterations are the initial vector elements.
[0019] According to the present invention, a multi-objective separation method based on a split-loop strategy is provided, wherein obtaining a first intermediate vector feature queue and a second intermediate vector feature queue based on the area and topology type of the intermediate vector features after the current iteration includes:
[0020] If the area of the intermediate vector element after the current iteration is less than or equal to a preset threshold, the intermediate vector element after the current iteration is added to the first intermediate vector element queue after the previous iteration to obtain the first intermediate vector element queue after the current iteration; and the intermediate vector element after the previous iteration is removed from the second intermediate vector element queue after the previous iteration to obtain the second intermediate vector element queue after the current iteration.
[0021] If the topological type of the intermediate vector features after this iteration is multiple independent polygons, the intermediate vector features in the second intermediate vector feature queue after the previous iteration are replaced with the intermediate vector features after this iteration to obtain the second intermediate vector feature queue after this iteration.
[0022] According to a multi-objective separation method based on a split-loop strategy provided by the present invention, before adding the intermediate vector features after the current iteration to the first intermediate vector feature queue after the previous iteration, the method further includes:
[0023] The intermediate vector features after this iteration are subjected to positive buffer dilation; the positive buffer dilation is used to restore the scale of the intermediate vector features after the negative buffer erosion.
[0024] The present invention also provides a multi-target separation device based on a split-loop strategy, comprising the following modules:
[0025] An extraction module is used to obtain a binary raster mask image output by a semantic segmentation model; wherein, the semantic segmentation model is used to perform semantic segmentation on any image;
[0026] The conversion module is used to perform vectorization conversion on the connected regions in the binarized raster mask image to generate an initial vector feature queue.
[0027] The update module is used to update the initial vector feature queue to obtain the target vector feature queue.
[0028] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor executes the computer program to implement the multi-objective separation method based on the split loop strategy as described above.
[0029] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the multi-objective separation method based on the split-loop strategy as described above.
[0030] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the multi-objective separation method based on the splitting loop strategy described above.
[0031] The present invention provides a multi-target separation method and apparatus based on a splitting loop strategy. This method acquires a binary raster mask image output by a semantic segmentation model. The semantic segmentation model is used to perform semantic segmentation on any image. An initial vector feature queue is generated by vectorizing connected regions in the binary raster mask image, thereby converting discrete raster data into a continuous vector representation. This effectively preserves the topological structure and boundary accuracy of adhered targets, improving the subsequent separation accuracy of adhered targets. The initial vector feature queue is updated to obtain a target vector feature queue. Adhesive targets are then gradually separated through iterative splitting, generating an independent polygon queue, which improves the separation accuracy and efficiency of densely adhered targets in the image semantic segmentation results. Attached Figure Description
[0032] To more clearly illustrate the technical solutions in this invention or 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 some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.
[0033] Figure 1 This is a flowchart illustrating the multi-objective separation method based on a splitting loop strategy provided by the present invention.
[0034] Figure 2 This is a schematic diagram of the separation results of multiple severely adhered impact craters and adjacent landforms provided by the present invention.
[0035] Figure 3 This is a schematic diagram of the iterative update process provided by the present invention.
[0036] Figure 4 This is a schematic diagram of the structure of the multi-target separation device based on the splitting cycle strategy provided by the present invention.
[0037] Figure 5 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation
[0038] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.
[0039] In current target recognition and semantic segmentation tasks, deep learning models typically take images as input and output pixel-level raster masks to label target features. However, in real-world applications such as impact crater recognition, cell detection, or lithological extraction, targets often exhibit high-density distribution and complex morphologies. Adjacent targets are prone to unintended connectivity at the raster level, forming adhered regions. If these adhered targets are not effectively separated, subsequent morphological parameter calculations, target counting, and classification statistics will be significantly affected, leading to interpretation bias or even erroneous conclusions.
[0040] This application proposes a multi-target separation method and apparatus based on a split-loop strategy to address the problems of incomplete separation, boundary distortion, and topological destruction in existing raster domain morphological opening operations when handling multiple adhered targets. This method transfers the traditional raster opening operation concept to vector space, replacing the erosion and dilation of fixed structuring elements with geometric buffer operations, and achieving layer-by-layer separation of complex adhered targets through a dynamic iterative mechanism.
[0041] The following is combined Figures 1 to 5 This invention describes a multi-target separation method and apparatus based on a splitting loop strategy.
[0042] Figure 1 This is a flowchart illustrating the multi-objective separation method based on a split-loop strategy provided by the present invention, as shown below. Figure 1 As shown, the method includes the following steps:
[0043] Step 101: Obtain the binarized raster mask image output by the semantic segmentation model; wherein, the semantic segmentation model is used to perform semantic segmentation on any image;
[0044] Step 102: Perform vectorization transformation on the connected regions in the binarized raster mask image to generate an initial vector feature queue;
[0045] Step 103: Update the initial vector feature queue to obtain the target vector feature queue.
[0046] Specifically, the binary raster mask image output by the semantic segmentation model is first obtained; then, the raster mask is converted to the vector domain through vectorization transformation, and discrete pixels are represented as continuous geometric boundaries to generate an initial vector feature queue; finally, an iterative buffer splitting mechanism is introduced to split all the initial vector features in the initial vector feature queue to obtain the target vector feature queue, thereby breaking through the limitation of traditional methods that a single opening operation can only separate two targets, and improving the target separation accuracy and efficiency in complex and sticky multi-target scenarios.
[0047] The semantic segmentation model can be any model, such as an encoder-decoder based model (e.g., U-Net), a dilated convolution based model (e.g., DeepLabv3+ from the DeepLab series), or a visual Transformer based model (e.g., SegFormer). The semantic segmentation model is used to perform semantic segmentation on any image; specifically, it classifies each pixel in the image and outputs a binary raster mask image.
[0048] For example, a semantic segmentation model can take a color or grayscale image as input and output a segmentation mask of the same size as the input image, where the value of each pixel represents its semantic category. In a binary raster mask image, pixel values contain only two categories: foreground (usually represented by '1' or '255', representing target features such as landform units, lithological bodies, spores, or cells) and background (usually represented by '0', representing non-target areas).
[0049] In one embodiment, a schematic diagram of the separation results of multiple severely adhered impact craters and adjacent landforms provided by the present invention is shown, such as... Figure 2 As shown, the semantic segmentation model performs impact crater analysis on remote sensing images (such as...). Figure 2 The left image in the image is used for identification, and a binary raster mask is output (such as...). Figure 2 The middle image shows impact craters in the foreground and zeros in the background. However, due to the dense distribution of impact craters, multiple craters in the mask adhere together. This embodiment of the invention first converts each connected region in the mask into a vector polygon object, generating an initial vector feature queue; then, iterative updates are performed on the initial vector features (i.e., vector polygon objects) in the queue, outputting a target vector feature queue. The target vector feature queue contains independent impact crater vectors (e.g.,...). Figure 2 The right-hand image (see figure in the image) can be used for accurate counting and morphological analysis.
[0050] The multi-target separation method based on a splitting loop strategy provided by this invention obtains a binary raster mask image output by a semantic segmentation model. The semantic segmentation model is used to perform semantic segmentation on any image. An initial vector feature queue is generated by vectorizing connected regions in the binary raster mask image, thereby converting discrete raster data into a continuous vector representation. This effectively preserves the topological structure and boundary accuracy of connected targets, improving the subsequent separation accuracy of connected targets. By updating the initial vector feature queue, a target vector feature queue is obtained. Then, connected targets are gradually separated through iterative splitting, generating an independent polygon queue. This method can systematically and automatically process complex scenes containing any number of connected targets, improving the separation accuracy and efficiency of densely connected targets in image semantic segmentation results.
[0051] Further, the step of vectorizing the connected regions in the binarized raster mask image to generate an initial vector feature queue includes:
[0052] Each of the multiple connected regions in the binarized raster mask image is converted into multiple vector polygon objects; each connected region corresponds one-to-one with a vector polygon object.
[0053] Delete pixels with a value of 0 from each of the vector polygon objects to obtain the corresponding initial vector features;
[0054] Based on the initial vector features, the initial vector feature queue is generated.
[0055] Specifically, the raster-to-vector conversion algorithm from geographic information system libraries such as ArcGIS or GDAL is first used to convert multiple connected regions in the binarized raster mask image into multiple vector polygon objects. Each connected region is defined as an independent vector polygon object, and its boundary coordinates and attribute information are recorded.
[0056] Then, background pixels (i.e., non-target vector features) with attribute information of 0 (i.e., pixel value of 0) in each vector polygon object are deleted, and only the foreground target vector features are retained as initial vector features to generate an initial vector feature queue.
[0057] For example, in one embodiment, the semantic segmentation model outputs a lithological body raster mask based on seismic images, which includes multiple lithological body adhering targets. First, a boundary tracing algorithm (such as Moore's adjacency method) is used to convert each connected region into a vector polygon object, ensuring that each vector polygon object corresponds to a lithological body outline. Then, pixels with a value of 0 are deleted, generating an initial vector feature queue where each feature contains the geometric properties and coordinates of the lithological body.
[0058] For example, in another embodiment, for the detection of rice blast fungus spores in agricultural pathology, after segmenting the microscopic image using a semantic segmentation model, spore regions are found to be clustered together in the image. First, an open-source tool (such as OpenCV's `findContours` function) is used to convert each spore connected region into a vector polygon object. Then, non-spore pixels (pixel values of 0) are removed to obtain an initial vector feature queue, where each element is an independent spore polygon for subsequent iterative separation.
[0059] This invention converts discrete raster data into a continuous vector representation by vectorizing connected regions in a binary raster mask image. This effectively avoids problems such as jagged edges, local breaks, or misconnections caused by pixel discretization in raster operations, ensuring that the topological relationship of the target is not destroyed during the separation process and that the geometric shape remains realistic and smooth. Furthermore, by removing background pixels, noise interference is eliminated, providing clean input data for iterative splitting and reducing the computational complexity of subsequent iterative separation, thereby improving the accuracy and efficiency of subsequent iterative separation.
[0060] Further, updating the initial vector feature queue to obtain the target vector feature queue includes:
[0061] Repeatedly execute the preset iterative steps to split each initial vector element in the initial vector element queue until the area of all initial vector elements is less than or equal to a preset threshold, and use the first intermediate vector element queue obtained in the last iteration as the target vector element queue.
[0062] Furthermore, the iterative steps include:
[0063] Negative buffer erosion is applied to the intermediate vector elements after the previous iteration to obtain the intermediate vector elements after the current iteration.
[0064] Based on the area and topology type of the intermediate vector elements after this round of iteration, the first intermediate vector element queue and the second intermediate vector element queue after this round of iteration are obtained.
[0065] The intermediate vector elements used in the first round of iterations are the initial vector elements.
[0066] Further, obtaining the first and second intermediate vector feature queues based on the area and topology type of the intermediate vector features after the current iteration includes:
[0067] If the area of the intermediate vector element after the current iteration is less than or equal to a preset threshold, the intermediate vector element after the current iteration is added to the first intermediate vector element queue after the previous iteration to obtain the first intermediate vector element queue after the current iteration; and the intermediate vector element after the previous iteration is removed from the second intermediate vector element queue after the previous iteration to obtain the second intermediate vector element queue after the current iteration.
[0068] If the topological type of the intermediate vector features after this iteration is multiple independent polygons, the intermediate vector features in the second intermediate vector feature queue after the previous iteration are replaced with the intermediate vector features after this iteration to obtain the second intermediate vector feature queue after this iteration.
[0069] Furthermore, before adding the intermediate vector features from the current iteration to the first intermediate vector feature queue from the previous iteration, the method further includes:
[0070] The intermediate vector features after this iteration are subjected to positive buffer dilation; the positive buffer dilation is used to restore the scale of the intermediate vector features after the negative buffer erosion.
[0071] Specifically, Figure 3 This is a schematic diagram of the iterative update process provided by the present invention, such as... Figure 3 As shown, a single initial vector feature is first taken from the initial vector feature queue and negative buffer erosion (i.e., offsetting the boundary inward) is performed on it to gradually weaken the adhesion parts such as narrow necks or narrow bridges in the connected region, so that the originally connected boundaries tend to separate.
[0072] After completing negative buffer erosion, the area and topology of the currently obtained intermediate vector features are automatically detected: if the area is less than or equal to a preset threshold, the vector polygon object is considered to have been completely eroded and is deemed indivisible. In this case, positive buffer dilation of the same distance is applied to the intermediate vector features after this iteration, and they are moved to the completed list to obtain the first intermediate vector feature queue after this iteration. If the area is greater than the preset threshold, and the topology changes from a single polygon to multiple independent polygons, the split is considered successful. At this point, the intermediate vector features before the split are removed from the initial vector feature queue, and positive buffer dilation of the same distance is applied to the newly generated multiple independent polygons (i.e., the intermediate vector features after this iteration). These are then re-added to the initial vector feature queue to obtain the second intermediate vector feature queue after this iteration. Positive buffer dilation is used to restore the original scale and shape of the object, ensuring that weakened connections do not re-attach while maintaining overall geometric accuracy.
[0073] In the iterative steps, during each negative buffer erosion, the offset distance of the inward boundary decreases by a certain distance in each iteration until the object is completely eroded or successfully split. Simultaneously, the second intermediate vector feature queue is continuously updated until the queue is empty, meaning all objects have been separated or determined to be indivisible units. The resulting first intermediate vector feature queue is the target vector feature queue.
[0074] All final individual objects will be exported from the target vector feature queue as vector format files, which can be directly used for subsequent spatial analysis, statistics, and 3D visualization.
[0075] For example, in medical image analysis, microscopic images are processed by a semantic segmentation model to output a binary raster mask image of the cell region. Cells often adhere together due to overlap. First, each adhered region is vectorized to generate an initial vector feature queue. Then, through a pre-defined iterative step, each initial vector feature in the initial vector feature queue is split. Specifically, the first iteration uses a large negative buffer distance (e.g., 5 pixels) to erode cell boundaries and break the thin necks. During iterations, the area and topology of the intermediate vector features generated in each iteration are automatically detected: if the area is less than or equal to a preset threshold (e.g., 100 pixels), the vector polygon object is considered completely eroded and indivisible. In this case, the intermediate vector features after this iteration undergo positive buffer dilation at the same distance and are moved to the completed list, obtaining the first intermediate vector feature queue after this iteration. If the topology changes from a single polygon to multiple independent polygons (e.g., one cell polygon splits into two cell polygons), the split is considered successful. At this point, the cell polygon before the split is removed from the initial vector feature queue, and the two newly generated cell polygons undergo positive buffer dilation at the same distance. They are then re-added to the initial vector feature queue, obtaining the second intermediate vector feature queue after this iteration. The above iterative steps are repeated until all cell units are separated, resulting in the target vector feature queue, which can be used for cell counting and morphological statistics.
[0076] This invention determines the degree of separation of adhered targets by the change in polygon area in the corrosion results. When the area is less than or equal to a preset threshold, it indicates that the target can no longer be further divided (e.g., it becomes the smallest identifiable unit or an indivisible unit), thus confirming complete separation and improving separation accuracy. By gradually weakening the adhered region through an iterative mechanism and dual-queue management, it achieves layer-by-layer separation of multiple complex adhered, clustered, or irregularly distributed targets, improving the integrity and automation of separation. This breaks through the limitation of traditional methods where a single opening operation can only separate two targets, improving the target separation efficiency and computational efficiency in multi-target complex adhered scenarios.
[0077] The multi-target separation device based on the split-loop strategy provided by the present invention is described below. The multi-target separation device based on the split-loop strategy described below can be referred to in correspondence with the multi-target separation method based on the split-loop strategy described above.
[0078] Figure 4 This is a schematic diagram of the structure of the multi-target separation device based on the splitting loop strategy provided by the present invention, as shown below. Figure 4 As shown. An embodiment of the present invention provides a multi-target separation device based on a splitting loop strategy, comprising an extraction module 401, a conversion module 402, and an update module 403, wherein:
[0079] Extraction module 401 is used to obtain the binary raster mask image output by the semantic segmentation model; wherein, the semantic segmentation model is used to perform semantic segmentation on any image; conversion module 402 is used to perform vectorization conversion on the connected regions in the binary raster mask image to generate an initial vector feature queue; update module 403 is used to update the initial vector feature queue to obtain the target vector feature queue.
[0080] The multi-target separation device based on a splitting loop strategy provided by this invention acquires a binary raster mask image output by a semantic segmentation model. The semantic segmentation model is used to perform semantic segmentation on any image. An initial vector feature queue is generated by vectorizing connected regions in the binary raster mask image, thereby converting discrete raster data into a continuous vector representation. This effectively preserves the topological structure and boundary accuracy of adhered targets, improving the subsequent separation accuracy of adhered targets. By updating the initial vector feature queue, a target vector feature queue is obtained. Adhesive targets are then gradually separated through iterative splitting, generating an independent polygon queue, which improves the separation accuracy and efficiency of densely adhered targets in the image semantic segmentation results.
[0081] Figure 5 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 5 As shown, the electronic device may include: a processor 510, a communications interface 520, a memory 530, and a communication bus 540, wherein the processor 510, the communications interface 520, and the memory 530 communicate with each other via the communication bus 540. The processor 510 can call logical instructions in the memory 530 to execute a multi-objective separation method based on a split-loop strategy, the method including:
[0082] Obtain the binary raster mask image output by the semantic segmentation model; wherein, the semantic segmentation model is used to perform semantic segmentation on any image;
[0083] The connected regions in the binarized raster mask image are vectorized to generate an initial vector feature queue;
[0084] The initial vector feature queue is updated to obtain the target vector feature queue.
[0085] Furthermore, the logical instructions in the aforementioned memory 530 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.
[0086] On the other hand, the present invention also provides a computer program product, the computer program product comprising a computer program that can be stored on a non-transitory computer-readable storage medium, wherein when the computer program is executed by a processor, the computer is able to execute the multi-objective separation method based on the split-loop strategy provided by the above methods, the method comprising:
[0087] Obtain the binary raster mask image output by the semantic segmentation model; wherein, the semantic segmentation model is used to perform semantic segmentation on any image;
[0088] The connected regions in the binarized raster mask image are vectorized to generate an initial vector feature queue;
[0089] The initial vector feature queue is updated to obtain the target vector feature queue.
[0090] In another aspect, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the multi-objective separation method based on the split-loop strategy provided by the above methods, the method comprising:
[0091] Obtain the binary raster mask image output by the semantic segmentation model; wherein, the semantic segmentation model is used to perform semantic segmentation on any image;
[0092] The connected regions in the binarized raster mask image are vectorized to generate an initial vector feature queue;
[0093] The initial vector feature queue is updated to obtain the target vector feature queue.
[0094] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.
[0095] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.
[0096] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element. Furthermore, it should be noted that the scope of the methods and apparatuses in the embodiments of this application is not limited to performing functions in the order shown or discussed, but may also include performing functions substantially simultaneously or in the reverse order, depending on the functions involved. For example, the described methods may be performed in a different order than described, and various steps may be added, omitted, or combined. Additionally, features described with reference to certain examples may be combined in other examples.
[0097] In this application's embodiments, "determine B based on A" means that factor A must be considered when determining B. It is not limited to "B can be determined based solely on A," but should also include: "determine B based on A and C," "determine B based on A, C, and E," "determine C based on A, and further determine B based on C," etc. Additionally, it can include using A as a condition for determining B, for example, "when A meets the first condition, determine B using the first method"; another example, "when A meets the second condition, determine B," etc.; another example, "when A meets the third condition, determine B based on the first parameter," etc. Of course, it can also be a condition where A is a factor in determining B, for example, "when A meets the first condition, determine C using the first method, and further determine B based on C," etc.
[0098] It should also be noted that the terms "target," "first," and "second" in this invention are used to distinguish similar objects, not to describe a specific order or sequence. It should be understood that such terms can be used interchangeably where appropriate so that embodiments of this application can be implemented in orders other than those illustrated or described herein, and the objects distinguished by "first" and "second" are generally of the same class, without limiting the number of objects; for example, the first object can be one or more.
[0099] In this invention, the term "multiple" refers to two or more, and other quantifiers are similar.
[0100] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims
1. A multi-objective separation method based on a split-loop strategy, characterized in that, include: Obtain the binary raster mask image output by the semantic segmentation model; wherein, the semantic segmentation model is used to perform semantic segmentation on any image; The connected regions in the binarized raster mask image are vectorized to generate an initial vector feature queue; The initial vector feature queue is updated to obtain the target vector feature queue; The step of vectorizing the connected regions in the binarized raster mask image to generate an initial vector feature queue includes: Each of the multiple connected regions in the binarized raster mask image is converted into multiple vector polygon objects; each connected region corresponds one-to-one with a vector polygon object. Delete pixels with a value of 0 from each of the vector polygon objects to obtain the corresponding initial vector features; Based on the initial vector features, generate the initial vector feature queue; The step of updating the initial vector feature queue to obtain the target vector feature queue includes: Repeatedly execute the preset iterative steps to split each initial vector element in the initial vector element queue until the area of all initial vector elements is less than or equal to a preset threshold, and use the first intermediate vector element queue obtained in the last iteration as the target vector element queue. The iterative steps include: Negative buffer erosion is applied to the intermediate vector elements after the previous iteration to obtain the intermediate vector elements after the current iteration. Based on the area and topology type of the intermediate vector elements after this round of iteration, the first intermediate vector element queue and the second intermediate vector element queue after this round of iteration are obtained. The intermediate vector elements used in the first round of iterations are the initial vector elements.
2. The multi-objective separation method based on a split-loop strategy according to claim 1, characterized in that, The process of obtaining the first and second intermediate vector feature queues based on the area and topology type of the intermediate vector features after the current iteration includes: If the area of the intermediate vector element after the current iteration is less than or equal to a preset threshold, the intermediate vector element after the current iteration is added to the first intermediate vector element queue after the previous iteration to obtain the first intermediate vector element queue after the current iteration; and the intermediate vector element after the previous iteration is removed from the second intermediate vector element queue after the previous iteration to obtain the second intermediate vector element queue after the current iteration. If the topological type of the intermediate vector features after this iteration is multiple independent polygons, the intermediate vector features in the second intermediate vector feature queue after the previous iteration are replaced with the intermediate vector features after this iteration to obtain the second intermediate vector feature queue after this iteration.
3. The multi-objective separation method based on a split-loop strategy according to claim 2, characterized in that, Before adding the intermediate vector features from the current iteration to the queue of the first intermediate vector features from the previous iteration, the method further includes: The intermediate vector features after this iteration are subjected to positive buffer dilation; the positive buffer dilation is used to restore the scale of the intermediate vector features after the negative buffer erosion.
4. A multi-target separation device based on a splitting loop strategy, characterized in that, include: An extraction module is used to obtain a binary raster mask image output by a semantic segmentation model; wherein, the semantic segmentation model is used to perform semantic segmentation on any image; The conversion module is used to perform vectorization conversion on connected regions in the binary raster mask image to generate an initial vector feature queue; the vectorization conversion on connected regions in the binary raster mask image to generate the initial vector feature queue includes: Each of the multiple connected regions in the binarized raster mask image is converted into multiple vector polygon objects; each connected region corresponds one-to-one with a vector polygon object. Delete pixels with a value of 0 from each of the vector polygon objects to obtain the corresponding initial vector features; Based on the initial vector features, generate the initial vector feature queue; An update module is used to update the initial vector feature queue to obtain a target vector feature queue; the step of updating the initial vector feature queue to obtain the target vector feature queue includes: Repeatedly execute the preset iteration to split each initial vector element in the initial vector element queue until the area of all initial vector elements is less than or equal to the preset threshold, and use the first intermediate vector element queue obtained in the last iteration as the target vector element queue. The iteration includes: Negative buffer erosion is applied to the intermediate vector elements after the previous iteration to obtain the intermediate vector elements after the current iteration. Based on the area and topology type of the intermediate vector elements after this round of iteration, the first intermediate vector element queue and the second intermediate vector element queue after this round of iteration are obtained. The intermediate vector elements used in the first round of iterations are the initial vector elements.
5. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the multi-objective separation method based on the split-loop strategy as described in any one of claims 1 to 3.
6. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the multi-objective separation method based on the split-loop strategy as described in any one of claims 1 to 3.
7. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the multi-objective separation method based on the split-loop strategy as described in any one of claims 1 to 3.