Long-range drone image sub-scene segmentation method, system, medium and equipment

Abstract

The present invention discloses a method, system, medium, and device for segmenting subscenes in long-strip drone images, relating to the field of digital photogrammetry. The method mainly comprises: constructing a scene graph using connection point constraints between images, simplifying it, and eliminating gross errors; analyzing the camera parameters of a single image to optimize the gradient direction by combining neighboring images that have a connection relationship with the current image; setting connection edge weights based on similarity metrics between image nodes, and achieving robust subscene segmentation using a normalized graph cut method. Implementing the method, system, medium, and device for segmenting subscenes in long-strip drone images provided by the present invention can improve the robustness of subscene segmentation in long-strip drone images and help improve the accuracy of subscene reconstruction.

Description

Technical Field

[0001] The present invention relates to the field of digital photogrammetry, and more specifically to a method, system, medium and device for segmenting sub-scenes of long-flight unmanned aerial vehicle (UAV) images. Background Art

[0002] Unmanned aerial vehicle (UAV) photogrammetry is an important means of achieving three-dimensional scene reconstruction and is widely used in digital cities, disaster warning, power inspection, and other fields. Traditionally, large-scale, regional UAV imagery uses a multi-strip structure for data collection. These image strips have strong constraints, so sub-scene segmentation and model merging are effective approaches to parallelize processing and improve 3D reconstruction efficiency. However, for strip-like scenes, UAV imagery often uses rectangular or S-shaped flight patterns with two strips to achieve efficient data collection, resulting in weaker image strip constraints. When performing sub-scene segmentation and model merging on such scenes, the stability of the internal structure of the segmented sub-scenes must be considered. Current sub-scene segmentation methods primarily target large-scale, regional scenes and typically use scene graphs to describe the images and their connectivity. Sub-scene segmentation can be transformed into a scene graph segmentation problem by incorporating graph theory. Edge weights are designed based on the number of matching points between image pairs, visual similarity, and epipolar geometric error. Sub-model segmentation is then performed based on graph theory, including graph cuts and community detection algorithms. However, the above methods use connection point information or image similarity measurement to design the weight information of the connection edges in the graph. These methods cannot directly reflect the size of the connection point error and cannot guarantee the minimum error during the adjustment optimization between sub-scenes. Summary of the Invention

[0003] The purpose of the present invention is to provide a method, system, medium and equipment for sub-scene segmentation of long-strip UAV images, which can improve the robustness of sub-scene segmentation in long-strip UAV images and help improve the accuracy of sub-scene reconstruction.

[0004] The present invention provides a long-strip unmanned aerial vehicle image sub-scene segmentation method, comprising the following steps: S1: obtaining an initial scene graph, preprocessing the initial scene graph, and obtaining a preprocessed initial scene graph; S2: obtaining a single image camera parameter optimization gradient direction according to the preprocessed initial scene graph; S3: obtaining an image connection edge weight according to the single image camera parameter optimization gradient direction and the preprocessed initial scene graph; and performing sub-scene segmentation using a normalized graph cut algorithm based on the image connection edge weight.

[0005] Furthermore, step S2 specifically includes: S21: according to the preprocessed initial scene graph, obtaining the optimized gradient direction of the reference image camera parameter in the current image pair, such as the formula:

[0006] ,

[0007] ,

[0008] in, Indicates the steepest direction of the reference image camera parameter optimization in the current image pair, that is, the optimization gradient direction of the reference image camera parameter in the current image pair; Indicates the Image nodes, is the Jacobian matrix of the objective function, is the error vector; To optimize the objective function; For images and images The relative translation vector between Indicates in the image The normalized three-dimensional coordinates of the feature points in , Indicates in the image Normalized three-dimensional coordinates of the feature points in ; S22: According to the optimized gradient direction of the reference image camera parameters in the current image pair, the principal component analysis algorithm is used to obtain the optimized gradient direction of the single image camera parameters.

[0009] Furthermore, the gradient direction and the pre-processed initial scene graph are optimized based on the camera parameters of a single image to obtain the edge weight of the image, as shown in the formula:

[0010] ,

[0011] in, is the gradient consistency cost function, that is, the image The connection edge weights of For images The camera parameter optimization gradient direction, express The transpose of For images The camera parameter optimization gradient direction, Represents the norm.

[0012] The present invention also provides a long-strip UAV image sub-scene segmentation system, comprising the following modules: a preprocessing module, configured to: obtain an initial scene graph, preprocess the initial scene graph, and obtain a preprocessed initial scene graph; an optimized gradient direction calculation module, configured to: obtain a single image camera parameter optimized gradient direction based on the preprocessed initial scene graph; an image connection edge weight module, configured to: obtain an image connection edge weight based on the single image camera parameter optimized gradient direction and the preprocessed initial scene graph; and perform sub-scene segmentation using a normalized graph cut algorithm based on the image connection edge weight.

[0013] Furthermore, the optimized gradient direction calculation module is specifically configured as follows: according to the preprocessed initial scene graph, the optimized gradient direction of the reference image camera parameters in the current image pair is obtained, such as the formula:

[0014] ,

[0015] ,

[0016] in, Indicates the steepest direction of the reference image camera parameter optimization in the current image pair, that is, the optimization gradient direction of the reference image camera parameter in the current image pair; Indicates the Image nodes, is the Jacobian matrix of the objective function, is the error vector; To optimize the objective function; For images and images The relative translation vector between Indicates in the image The normalized three-dimensional coordinates of the feature points in , Indicates in the image The normalized three-dimensional coordinates of the feature points in the image pair are obtained; according to the optimized gradient direction of the reference image camera parameters in the current image pair, the principal component analysis algorithm is used to obtain the optimized gradient direction of the single image camera parameters.

[0017] Furthermore, the gradient direction and the pre-processed initial scene graph are optimized based on the camera parameters of a single image to obtain the edge weight of the image, as shown in the formula:

[0018] ,

[0019] in, is the gradient consistency cost function, that is, the image The connection edge weights of For images The camera parameter optimization gradient direction, express The transpose of For images The camera parameter optimization gradient direction, Represents the norm.

[0020] The present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the above-mentioned long-strip drone image sub-scene segmentation method.

[0021] The present invention also provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, the steps of the above-mentioned long-strip drone image sub-scene segmentation method are implemented.

[0022] The present invention also provides a computer program product, comprising a computer program, which, when executed by a processor, implements the steps of the above-mentioned long-strip drone image sub-scene segmentation method.

[0023] The implementation of the long-strip drone image sub-scene segmentation method, system, medium, and device provided by the present invention has the following beneficial effects:

[0024] The present invention uses the connection point constraints between images to construct a scene graph, and simplifies and removes gross errors; combines neighboring images that have a connection relationship with the current image to analyze the camera parameters of a single image and optimize the gradient direction; sets the weights of the connection edges in the scene graph through the similarity measurement between image nodes, and uses the normalized graph cut method to achieve robust segmentation of sub-scenes; effectively ensures that the sub-scenes have a robust connection structure, which can improve the robustness of sub-scene segmentation in long-range drone images, ensure that the internal size of the segmented sub-scenes is balanced and the structure is stable, provide a stable internal structure for the parallel reconstruction of sub-scenes, and provide important structural support for the subsequent reconstruction and merging of sub-scene models, which helps to improve the accuracy of sub-scene reconstruction, and thus achieve high-precision and rapid reconstruction of long-range drone images. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] The present invention will be further described below with reference to the accompanying drawings and embodiments, in which:

[0026] Figure 1 This is a flow chart of the long-strip UAV image sub-scene segmentation method provided by the present invention;

[0027] Figure 2 is the initial scene graph provided by the present invention;

[0028] Figure 3 It is a schematic diagram of a local closed-loop constraint provided by the present invention;

[0029] Figure 4 This is a schematic diagram of gradient direction analysis for optimizing parameters of a single-image camera provided by the present invention;

[0030] Figure 5 1 is a schematic diagram of image block segmentation results based on normalized graph cuts provided by the present invention;

[0031] Figure 6 It is a structural block diagram of the computer device provided by the present invention. DETAILED DESCRIPTION

[0032] In order to have a clearer understanding of the technical features, purposes and effects of the present invention, specific embodiments of the present invention are now described in detail with reference to the accompanying drawings.

[0033] Figure 1 A schematic diagram of a long-strip drone image sub-scene segmentation method of this embodiment is shown. In this embodiment, the long-strip drone image sub-scene segmentation method includes the following steps:

[0034] S1: Obtain an initial scene graph, preprocess the initial scene graph, and obtain a preprocessed initial scene graph;

[0035] As an exemplary embodiment, in step S1, the initial scene graph is subjected to gross error elimination using a connection point information constraint method and an adaptive local closed-loop constraint method to obtain a preprocessed initial scene graph;

[0036] S2: Based on the preprocessed initial scene graph, the gradient direction of the camera parameter optimization of a single image is obtained;

[0037] In an exemplary embodiment, step S2 specifically includes:

[0038] S21: Based on the preprocessed initial scene graph, the optimized gradient direction of the reference image camera parameters in the current image pair is obtained, such as the formula:

[0039] ,

[0040] ,

[0041] in, Indicates the steepest direction of the reference image camera parameter optimization in the current image pair, that is, the optimization gradient direction of the reference image camera parameter in the current image pair; Indicates the Image nodes, is the Jacobian matrix of the objective function, is the error vector; To optimize the objective function; For images and images The relative translation vector between , indicating that in the image Feature points in The normalized three-dimensional coordinates of From global coordinate system to image The rotation matrix of the camera coordinate system, is the pixel coordinate of the feature point in the two-dimensional image; Indicates in the image Feature points in Normalized three-dimensional coordinates of feature points and are image points with the same name;

[0042] S22: according to the optimized gradient direction of the reference image camera parameters in the current image pair, the principal component analysis algorithm is used to obtain the optimized gradient direction of the single image camera parameters;

[0043] S3: Optimize the gradient direction and the preprocessed initial scene graph based on the camera parameters of a single image to obtain the image's connection edge weights; then use the normalized graph cut algorithm to perform sub-scene segmentation based on the image's connection edge weights;

[0044] In an exemplary embodiment, the above-mentioned optimization of the gradient direction and the pre-processed initial scene graph based on the camera parameters of a single image is performed to obtain the edge weights of the image, such as the formula:

[0045] ,

[0046] in, is the gradient consistency cost function, that is, the image The connection edge weights of For images The camera parameter optimization gradient direction, express The transpose of For images The camera parameter optimization gradient direction, Represents the norm.

[0047] In one embodiment, the above-mentioned long-strip drone image sub-scene segmentation method includes: first, using the connection point constraints between images to construct a scene graph, and simplifying and removing gross errors; second, combining neighboring images that have a connection relationship with the current image, analyzing the camera parameters of a single image to optimize the gradient direction; finally, setting the connection edge weights through the similarity measurement between image nodes, and using the normalized graph cut method to achieve robust segmentation of the sub-scene.

[0048] In another embodiment, the above-mentioned long-strip drone image sub-scene segmentation method includes the following steps:

[0049] 1. Scene graph preprocessing: For a given scene graph with epipolar geometry, there are a large number of redundant connecting edges. Furthermore, due to the image capture lighting, viewing angle, and the presence of repeated structures in the scene, gross errors may occur in the epipolar geometry between connected images in the scene graph. Therefore, it is necessary to perform preprocessing operations to simplify the initial scene graph and remove gross errors. This embodiment first simplifies the subscene graph by using constraints such as the number of connection points, connection point distribution, and intersection angles. Then, using adaptive local closed-loop constraints, it searches for image nodes with large errors in the local closed-loop connection graph to remove gross errors in the scene graph.

[0050] 2. Single-image camera parameter optimization gradient direction analysis: For the sub-scene graph preprocessed in the above steps, use the image nodes and the connected domain images to calculate the camera parameter optimization gradient direction of a single image; specifically, first use the image as the reference image, then combine it with the connected domain images to construct an image pair, and use the kernel line constraint as the optimization cost function to find the steepest direction of the reference image parameters in the current image pair as the optimization gradient direction of the reference image camera parameters in the current image pair; after calculating all adjacent image pairs with the reference image, there are differences in the optimization gradient direction of the reference image in each image pair; finally, use the series of optimized gradient directions calculated for each image pair, combined with the principal component analysis method, to obtain the final optimized gradient direction of the current reference image;

[0051] 3. Sub-scene segmentation based on normalized graph cuts: In order to ensure sufficient balance between each segmented sub-scene and facilitate subsequent parallel reconstruction of the sub-scene model; this embodiment adopts the normalized graph cut method. On the basis of the above-mentioned scene graph preprocessing, the weights of the connecting edges in the scene graph are set according to the optimized gradient direction angle of the image, and finally combined with the normalized graph cut algorithm to ultimately achieve robust segmentation of the sub-scenes.

[0052] In another embodiment, the above-mentioned long-strip drone image sub-scene segmentation method includes the following steps:

[0053] Step 1: Scene graph preprocessing:

[0054] The input of this embodiment is the initial scene graph with epipolar geometric relationship ,like Figure 2 As shown; the vertex V of the scene graph represents the image sequence, and the edge E of the scene graph represents the relative pose of the image pair with a connection relationship This embodiment uses connection point information and adaptive local closed-loop constraints to simplify the scene graph and eliminate gross errors.

[0055] The connection point information constraints are as follows: UAV images have a high degree of overlap. The number and distribution of connection points can be used to eliminate weakly connected image pairs and achieve redundancy of the scene graph connection edges. First, all neighboring images that have a connection relationship with the current image are selected, sorted in descending order according to the number of connection points, and the connection edges smaller than the threshold are deleted in the scene graph. Second, among the remaining adjacent images, a weighted function that considers the distribution of connection points and the intersection angle is designed and the weights are sorted, retaining the previous image. adjacent images;

[0056] Among them, the adaptive local closed loop constraint is specifically: for image triplets with interconnected relationships , the relative rotation matrix should satisfy ,in is the identity matrix, Represents the metric between two matrices using distance metrics, is the threshold; in order to avoid the inefficiency caused by traversing all triples in the flight strip and the instability of the local triangle closed loop constraint, this embodiment analyzes the stable connection edges in the scene graph and performs adaptive local closed loop verification based on this, such as Figure 3 As shown; first build according to the image node Triangulated network, and verify whether all triangle vertices meet the angle constraint relationship, retain the ones that meet Threshold-constrained connected edges ; Secondly, iterate through , and along the edge Vertex Search for image nodes that are connected to it within a fixed distance , if the vertex and If there is a connection relationship between them along the retained edges, verify whether the closed loop composed of these edges satisfies the constraint relationship (where the vertex and Image Node connection), if it satisfies, the connection edge is retained ;Finally, after several iterations, the rough edges of the scene graph are eliminated;

[0057] Step 2: Single image camera parameter optimization gradient direction analysis:

[0058] In the preprocessed scene graph In the example, traverse the image nodes one by one , and find the node All neighboring nodes that have a connection relationship , computing nodes Optimize the objective function between neighboring nodes The steepest gradient direction ,

[0059] (1)

[0060] in is the Jacobian matrix of the objective function, is the error vector; kernel line constraint is used as the cost function, and the calculation formula is:

[0061] (2)

[0062] in, Representing feature points The normalized three-dimensional coordinates of Represents the transition from the global coordinate system to the image The rotation matrix of the camera coordinate system, For images and images The relative translation vector between

[0063] Calculate the node by formula (1) All optimized gradients between neighboring nodes ,like Figure 4 Finally, the principal component analysis algorithm is used to extract the main direction of the series of gradient vectors as the image The optimized gradient direction ;

[0064] Step 3: Sub-scene segmentation based on normalized graph cuts:

[0065] In the preprocessed scene graph In the example, the connection edge weights are set by the similarity measurement between image nodes, and the sub-scene segmentation problem can be solved using the minimum cut theory; the image is calculated through the above steps. The optimized gradient direction , using the gradient consistency cost function Computational imaging The connection edge weight is:

[0066] (3)

[0067] In order to ensure the balance of the size of the segmented sub-scenes, the normalized graph cut algorithm is used to Segmentation is performed, and the segmentation results are as follows Figure 5 As shown, it can provide a basic analysis unit for the subsequent parallel reconstruction of sub-scenes.

[0068] Through the above steps, the robust segmentation of the long-distance drone image sub-scenes is achieved, which can ensure the stability of the internal structure of the sub-scene, improve the accuracy of the sub-scene model reconstruction, and provide important structural support for the subsequent sub-model merging.

[0069] This embodiment provides a long-strip drone image sub-scene segmentation system, including the following modules: a preprocessing module, configured to obtain an initial scene graph, preprocess the initial scene graph, and obtain a preprocessed initial scene graph; an optimized gradient direction calculation module, configured to obtain a single-image camera parameter optimized gradient direction based on the preprocessed initial scene graph; an image connection edge weight module, configured to obtain an image connection edge weight based on the single-image camera parameter optimized gradient direction and the preprocessed initial scene graph; and perform sub-scene segmentation using a normalized graph cut algorithm based on the image connection edge weight.

[0070] In an exemplary embodiment, the optimized gradient direction calculation module is specifically configured to obtain the optimized gradient direction of the reference image camera parameters in the current image pair according to the preprocessed initial scene graph, such as the formula:

[0071] ,

[0072] ,

[0073] in, Indicates the steepest direction of the reference image camera parameter optimization in the current image pair, that is, the optimization gradient direction of the reference image camera parameter in the current image pair; Indicates the Image nodes, is the Jacobian matrix of the objective function, is the error vector; To optimize the objective function; For images and images The relative translation vector between , indicating that in the image Feature points in The normalized three-dimensional coordinates of From global coordinate system to image The rotation matrix of the camera coordinate system, is the pixel coordinate of the feature point in the two-dimensional image; Indicates in the image Feature points in Normalized three-dimensional coordinates of feature points and are image points with the same name; according to the optimized gradient direction of the reference image camera parameters in the current image pair, the principal component analysis algorithm is used to obtain the optimized gradient direction of the single image camera parameters.

[0074] In an exemplary embodiment, the above-mentioned optimization of the gradient direction and the pre-processed initial scene graph based on the camera parameters of a single image is performed to obtain the edge weights of the image, such as the formula:

[0075] ,

[0076] in, is the gradient consistency cost function, that is, the image The connection edge weights of For images The camera parameter optimization gradient direction, express The transpose of For images The camera parameter optimization gradient direction, Represents the norm.

[0077] This embodiment provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the above-described method for segmenting subscenes of long-distance drone imagery. The storage medium may be a magnetic disk, an optical disk, a read-only memory (ROM), a random access memory (RAM), a flash memory, a hard disk drive (HDD), or a solid-state drive (SSD); the storage medium may also include a combination of the aforementioned types of memory.

[0078] This embodiment provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, the steps of the above-mentioned long-strip drone image sub-scene segmentation method are implemented.

[0079] like Figure 6As shown, the computer device 120 may include: at least one processor 121, such as a central processing unit (CPU), at least one communication interface 123, a memory 124, and at least one communication bus 122. The communication bus 122 is used to enable communication between these components. The communication interface 123 may include a display and a keyboard. Optionally, the communication interface 123 may also include a standard wired interface or a wireless interface. The memory 124 may be a high-speed random access memory (RAM) or a non-volatile memory, such as at least one disk drive. The memory 124 may optionally be at least one storage device located remote from the processor 121. The memory 124 stores application programs, and the processor 121 invokes program code stored in the memory 124 to execute any of the aforementioned method steps. The communication bus 122 may be a peripheral component interconnect (PCI) bus or an extended industry standard architecture (EISA) bus, for example. The communication bus 122 can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 6The fact that only one line is used in the figure does not mean that there is only one bus or only one type of bus. Memory 124 may include volatile memory, such as random-access memory (RAM); it may also include non-volatile memory, such as flash memory, a hard disk drive (HDD), or a solid-state drive (SSD); or it may include a combination of these types of memory. Processor 121 may be a central processing unit (CPU), a network processor (NP), or a combination of a CPU and an NP. Processor 121 may further include a hardware chip. The hardware chip may be an application-specific integrated circuit (ASIC), a programmable logic device (PLD), or a combination thereof. The PLD may be a complex programmable logic device (CPLD), a field-programmable gate array (FPGA), a generic array logic (GAL), or any combination thereof. Optionally, the memory 124 is further configured to store program instructions. The processor 121 may call the program instructions to implement the long-flight strip drone image sub-scene segmentation method of this embodiment.

[0080] This embodiment provides a computer program product, including a computer program, which, when executed by a processor, implements the steps of the above-mentioned long-strip drone image sub-scene segmentation method.

[0081] The embodiments of the present invention are described above in conjunction with the accompanying drawings, but the present invention is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of the present invention, ordinary technicians in this field can also make many forms without departing from the scope of protection of the present invention and the claims, all of which are protected by the present invention.

Claims

1. A long-range UAV image sub-scene segmentation method, characterized in that: The following steps are involved: S1: Acquire an initial scene graph, and preprocess the initial scene graph to obtain a preprocessed initial scene graph; S2: Obtaining a single image camera parameter optimization gradient direction according to the preprocessed initial scene graph; S3: Optimizing the gradient direction and the preprocessed initial scene graph according to the camera parameters of the single image to obtain the image connection edge weights; performing sub-scene segmentation using a normalized graph cut algorithm based on the image connection edge weights; Step S2 specifically includes: S21: According to the pre-processed initial scene graph, the optimized gradient direction of the reference image camera parameter in the current image pair is obtained, such as the formula: ， ， in, Indicates the steepest direction of the reference image camera parameter optimization in the current image pair, that is, the optimization gradient direction of the reference image camera parameter in the current image pair; Indicates the Image nodes, is the Jacobian matrix of the objective function, is the error vector; To optimize the objective function; For images and images The relative translation vector between Indicates in the image The normalized three-dimensional coordinates of the feature points in , Indicates in the image The normalized three-dimensional coordinates of the feature points in; S22: Obtaining a single image camera parameter optimization gradient direction using a principal component analysis algorithm according to the optimization gradient direction of the reference image camera parameter in the current image pair.

2. The long-strip UAV image sub-scene segmentation method according to claim 1 is characterized in that: The gradient direction is optimized according to the camera parameters of the single image and the preprocessed initial scene graph to obtain the connection edge weight of the image, as shown in the formula: ， in, is the gradient consistency cost function, that is, the image The connection edge weights of For images The camera parameter optimization gradient direction, express The transpose of For images The camera parameter optimization gradient direction, Represents the norm.

3. A long-range drone image sub-scene segmentation system, characterized by: Includes the following modules: A preprocessing module is configured to: obtain an initial scene graph, preprocess the initial scene graph, and obtain a preprocessed initial scene graph; An optimized gradient direction calculation module is configured to: obtain a single image camera parameter optimized gradient direction according to the preprocessed initial scene graph; An image connection edge weight module is configured to: optimize the gradient direction and the preprocessed initial scene graph according to the single image camera parameters to obtain the image connection edge weight; and perform sub-scene segmentation using a normalized graph cut algorithm based on the image connection edge weight; The specific configuration of the optimized gradient direction calculation module is: According to the pre-processed initial scene graph, the optimized gradient direction of the reference image camera parameters in the current image pair is obtained, such as the formula: ， ， in, Indicates the steepest direction of the reference image camera parameter optimization in the current image pair, that is, the optimization gradient direction of the reference image camera parameter in the current image pair; Indicates the Image nodes, is the Jacobian matrix of the objective function, is the error vector; To optimize the objective function; For images and images The relative translation vector between Indicates in the image The normalized three-dimensional coordinates of the feature points in , Indicates in the image The normalized three-dimensional coordinates of the feature points in; According to the optimized gradient direction of the reference image camera parameters in the current image pair, the optimized gradient direction of the single image camera parameters is obtained by using a principal component analysis algorithm.

4. The long-strip UAV image sub-scene segmentation system according to claim 3 is characterized in that: The gradient direction is optimized according to the camera parameters of the single image and the preprocessed initial scene graph to obtain the connection edge weight of the image, as shown in the formula: ， in, is the gradient consistency cost function, that is, the image The connection edge weights of For images The camera parameter optimization gradient direction, express The transpose of For images The camera parameter optimization gradient direction, Represents the norm.

5. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the long-strip UAV image sub-scene segmentation method according to any one of claims 1-2 are implemented.

6. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the steps of the long-strip drone image sub-scene segmentation method according to any one of claims 1-2 are implemented.

7. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the long-strip drone image sub-scene segmentation method described in any one of claims 1-2 are implemented.

Images