Multi-scale super-pixel-level ground feature classification and identification method and device, medium and product

Abstract

The invention discloses a multi-scale super-pixel-level ground feature classification and recognition method and device, a medium and a product, and relates to the field of image processing. The method comprises the following steps: respectively carrying out principal component analysis on optical and radar remote sensing images to extract principal component feature images; normalizing the principal component feature image, and determining a super-pixel clustering target criterion; performing super-pixel segmentation on the principal component feature image based on a super-pixel clustering target criterion by adopting a plurality of segmentation sizes; the features of texture, polarization and the like are extracted based on the preprocessed optical and radar remote sensing images, the pixel mean value of feature images corresponding to all pixels included in each segmented superpixel is taken as the feature value of each superpixel on different feature images, the extracted multi-dimensional feature value is taken as the input of a machine learning model, and ground feature classification is identified. The adaptability and robustness of superpixel segmentation can be enhanced, diversified classification tasks and complex scenes can be better served, and therefore the precision and efficiency of ground feature classification and recognition are remarkably improved.

Description

Technical Field

[0001] The present application relates to the field of image processing, and in particular to a multi-scale super-pixel level object classification and recognition method, device, medium and product. Background Art

[0002] Superpixels are irregular pixel sets with specific properties that are between a single pixel and a complete image object, and are composed of a series of pixels with similar features and adjacent positions in the original image. As an important technology in image processing, superpixel segmentation uses specific standards to upgrade the image processing level from a single pixel to the regional level, thereby significantly reducing the complexity of image post-processing and providing great convenience for subsequent image analysis. However, in the face of the rapid development of multi-platform, multi-type remote sensing sensors and imaging technologies, most of the existing superpixel segmentation algorithms are limited to the processing of a single RGB image, which makes it difficult to effectively deal with the complex data sources in the context of remote sensing big data, becoming a key bottleneck restricting its further application.

[0003] The known superpixel segmentation algorithms represented by the simple linear iterative clustering method are mainly based on the color and spatial proximity of a single image for clustering. It is difficult to achieve the complementary advantages of multi-source remote sensing images and has poor adaptability to complex scenes. Although the generated superpixels maintain good consistency in shape and size, the fit between their boundaries and the true contour of the target may not be high enough, and over-segmentation is prone to occur. It is difficult to ensure the accuracy and robustness of segmentation, resulting in large errors in subsequent tasks such as object classification and target recognition. Summary of the invention

[0004] The purpose of this application is to provide a multi-scale superpixel-level object classification and identification method, equipment, medium and product, which can enhance the adaptability and robustness of superpixel segmentation, better serve diverse classification tasks and complex scenes, and thus significantly improve the accuracy and efficiency of object classification and identification.

[0005] To achieve the above objectives, this application provides the following solutions: In a first aspect, the present application provides a multi-scale super-pixel level ground object classification and recognition method, comprising: Acquiring remote sensing images, and dividing the remote sensing images into optical remote sensing images and radar remote sensing images according to the acquisition sources of the remote sensing images; Preprocessing the optical remote sensing image and the radar remote sensing image respectively; Conduct principal component analysis on the preprocessed optical remote sensing images and radar remote sensing images respectively; The optical remote sensing image after principal component analysis and the radar remote sensing image after principal component analysis that meet the set requirements are used as principal component feature images; Normalizing the principal component feature image to obtain a normalized result; Combining the normalization result and the row and column position information of pixels in the principal component feature image, a distance measurement of the distribution of cluster centers of each pixel is performed to obtain the superpixel clustering target criterion; Determine a plurality of segmentation sizes, and use the segmentation sizes to perform superpixel segmentation on the principal component feature image based on the superpixel clustering target criterion to obtain a segmentation result; Extracting feature images based on the preprocessed optical remote sensing image and the radar remote sensing image; the feature images at least include texture feature images and polarization feature images; The pixel mean of all pixels contained in each superpixel in the segmentation result corresponding to different feature images is taken as the feature value of each superpixel on each feature image; The feature value of each superpixel on different feature images is used as the input of the machine learning model to achieve ground object classification and recognition.

[0006] Optionally, the optical remote sensing image after principal component analysis and the radar remote sensing image after principal component analysis that meet the set requirements are used as the principal component feature image, including: The optical remote sensing image after principal component analysis and the radar remote sensing image after principal component analysis with cumulative contributions exceeding 90% are used as the principal component feature images.

[0007] Optionally, the segmentation size is used to perform superpixel segmentation on the principal component feature image based on the superpixel clustering target criterion to obtain a segmentation result, including: Using the segmentation size, and based on the superpixel clustering target criterion, the principal component feature image is segmented into maximum-scale superpixels to obtain superpixel blocks; Multi-scale superpixel segmentation is performed on the superpixel block to obtain the segmentation result.

[0008] Optionally, the segmentation size is adopted and the principal component feature image is segmented into maximum-scale superpixels based on the superpixel clustering target criterion to obtain superpixel blocks, including: Arrange the segmentation sizes in descending order according to superpixel sizes to obtain a scale sequence; For the first segmentation size in the scale sequence, completing seed point initialization and seed point gradient minimum position adjustment; For each seed point, determining the joint distance from the pixel point in the principal component feature image to the seed point within a search space range based on the superpixel clustering target criterion; the search space range is determined based on the distance between two adjacent seed points; A cluster center is assigned to each pixel point according to the minimum joint distance, and iterative clustering is performed to determine the changes in the eigenvalues ​​and pixel positions of all cluster centers in each principal component feature image after two adjacent iterations. When the changes in the eigenvalues ​​and pixel positions of the cluster centers in each principal component feature image are less than the first set threshold, the iteration is stopped to obtain a superpixel block.

[0009] Optionally, using the segmentation size to perform multi-scale superpixel segmentation on the superpixel block to obtain the segmentation result includes: Determine the distance index from all pixels in the superpixel block to the cluster center; When the distance index is greater than or equal to a second set threshold, continue to segment the superpixel block until the distance index is less than the second set threshold, and obtain a clustering result; The clustering result is subjected to post-segmentation processing to obtain the segmentation result.

[0010] Optionally, performing post-segmentation processing on the clustering result to obtain the segmentation result includes: Determine whether there are any abnormalities in the clustering results; the abnormalities include multi-connectivity in the clustering results, superpixel size being smaller than a set value, and a single superpixel being divided into multiple discontinuous superpixels; When there are abnormalities in the clustering result, post-segmentation processing is performed by enhancing connectivity to obtain the segmentation result.

[0011] Optionally, the machine learning model adopts a support vector machine.

[0012] In a second aspect, the present application provides a computer device comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of any of the multi-scale super-pixel-level object classification and recognition methods described above.

[0013] In a third aspect, the present application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of any of the multi-scale super-pixel-level ground object classification and recognition methods described above.

[0014] In a fourth aspect, the present application provides a computer program product, including a computer program, which, when executed by a processor, implements the steps of any of the multi-scale super-pixel level ground object classification and recognition methods described above.

[0015] According to the specific embodiments provided in this application, this application has the following technical effects: The present application provides a multi-scale super-pixel-level object classification and recognition method, equipment, medium and product. By dividing remote sensing images into optical remote sensing images and radar remote sensing images according to the acquisition source of remote sensing images, it is possible to integrate rich attribute features from different remote sensing data sources. Based on the super-pixel clustering target criterion, the main component feature image is super-pixel segmented to obtain the segmentation result, and the pixel mean of all pixels corresponding to the feature image contained in the segmentation result is taken as the feature value of each super-pixel on different feature images, and this feature value is used as the input of the machine learning model to realize the classification and recognition of objects, which can realize more refined and flexible processing of remote sensing images, generate super-pixel areas that are more in line with the real boundaries of objects, effectively reduce the common edge blurring problems in the classification process, enhance the adaptability and robustness of super-pixel segmentation, and enable it to better serve diverse classification tasks and complex scenes, thereby significantly improving the accuracy and efficiency of object classification and recognition. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0017] Figure 1 This is an application environment diagram of a multi-scale super-pixel-level ground object classification and recognition method in an embodiment of the present application; Figure 2 A schematic diagram of a process flow of a multi-scale super-pixel-level object classification and recognition method provided in an embodiment of the present application; Figure 3 A schematic diagram of a superpixel clustering target criterion provided in an embodiment of the present application; Figure 4 A schematic diagram of a single-scale superpixel clustering process provided by another embodiment of the present application; Figure 5 A schematic diagram of a multi-scale superpixel segmentation iterative process provided by another embodiment of the present application; Figure 6 A schematic diagram of the structure of a computer device provided in one embodiment of the present application. DETAILED DESCRIPTION

[0018] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0019] Traditional superpixel segmentation algorithms are mostly for the processing of single RGB images. Among them, the superpixel segmentation algorithm closest to this application is the simple linear iterative clustering method. The simple linear iterative clustering method uses K-means clustering to generate superpixels. Compared with other segmentation algorithms, its difference is: first, by limiting the search space, the distance calculation amount in the clustering process is greatly reduced. Second, the calculation of the distance between pixels is the weighted result of the pixel CIELab color and spatial position proximity, and the control of the superpixel segmentation size and compactness is achieved at the same time. The specific principle of the simple linear iterative clustering method is as follows: (1) Initialize seed points: according to the preset number of superpixels , divide the image evenly, and the center of each block is the initialization position of the seed point. Assume that the total number of pixels is , then the number of pixels contained in each superpixel can be approximately calculated as , the side length of each superpixel can be expressed as , .

[0020] (2) Seed point adjustment: Calculate the gradient value of each pixel within the 3×3 neighborhood of the initialized sub-points, and move the seed point to the position with the smallest gradient in the neighborhood to ensure that the seed point is located in a relatively stable and representative area, thereby improving the accuracy and stability of clustering.

[0021] (3) Cluster center allocation: Different from the K-means search of the entire image, simple linear iterative clustering limits the search space of various sub-points to , thereby accelerating the convergence of the algorithm. The distance metric involves color distance and space distance. The specific calculation formula is as follows: (1) (2) (3) in, Represents the color space distance. , Represent the seed point and search space pixel number respectively. , , Represents CIELab color space The attribute values ​​of each pixel are the current pixel brightness, red-green hue, and yellow-cyan hue. Represents the spatial position distance. , )and( , ) represent pixel positions respectively. Represents the joint distance result of color space and position space. and are normalization parameters, which are used to normalize color distance and spatial distance respectively.

[0022] Each pixel will be searched by multiple seed points, thus corresponding to the distances between multiple pixels and the seed points, and each pixel is assigned a label corresponding to the seed point with the minimum distance.

[0023] (4) Update the cluster center: The center position of all pixels belonging to a cluster center is used as the new cluster center.

[0024] (5) Iterative clustering: Repeat steps (3) to (4) until the change in the position of each cluster center after two consecutive iterations is less than the set threshold or no longer changes. The iteration is stopped.

[0025] (6) Post-segmentation processing: For the above iterative clustering process, there may be other abnormal segmentation situations in its clustering results, such as multiple connectivity, superpixel size is too small, and some superpixels are mis-segmented. These problems can be improved by enhancing connectivity, so that discontinuous superpixels or superpixels with too small sizes can be reallocated to labels of neighboring superpixels, effectively integrating and optimizing the clustering results.

[0026] In order to make the above-mentioned objects, features and advantages of the present application more obvious and easy to understand, the present application is further described in detail below in conjunction with the accompanying drawings and specific implementation methods.

[0027] Terminology explanation: (1) Superpixel: An irregular pixel block with certain visual significance composed of a series of adjacent pixels with similar texture, color, brightness and other characteristics.

[0028] (2) Superpixel segmentation: The process of subdividing a digital image into multiple superpixels is a refined form of image segmentation that can provide more meaningful image regions and facilitate subsequent processing.

[0029] The multi-scale super-pixel level ground object classification and recognition method provided in the embodiment of the present application can be applied to Figure 1In the application environment shown. Among them, the terminal 102 communicates with the server 104 through the network. The data storage system can store the data that the server 104 needs to process. The data storage system can be set up separately, integrated on the server 104, or placed on the cloud or other servers. The terminal 102 can send the remote sensing image to the server 104. After the server 104 receives the remote sensing image, for the remote sensing image, the server 104 divides the remote sensing image into an optical remote sensing image and a radar remote sensing image according to the acquisition source of the remote sensing image. The optical remote sensing image and the radar remote sensing image are preprocessed respectively. The preprocessed optical remote sensing image and the radar remote sensing image are respectively subjected to principal component analysis. The optical remote sensing image after the principal component analysis and the radar remote sensing image after the principal component analysis that meet the set requirements are used as the principal component feature image. The principal component feature image is normalized to obtain a normalized result. Combined with the normalized result and the pixel row and column position information in the principal component feature image, the distance measurement of the distribution of each pixel cluster center is performed to obtain the superpixel clustering target criterion. Determine multiple segmentation sizes, and use the segmentation sizes to perform superpixel segmentation on the main component feature image based on the superpixel clustering target criterion to obtain a segmentation result. Extract feature images including at least a texture feature image and a polarization feature image based on the preprocessed remote sensing image. Take the pixel mean of all pixels contained in each superpixel in the segmentation result corresponding to different feature images as the feature value of each superpixel on each feature image. Use the feature value of each superpixel on different feature images as the input of the machine learning model to achieve ground object classification and recognition. The server 104 can feed back the obtained ground object classification and recognition results to the terminal 102. In addition, in some embodiments, the multi-scale superpixel-level ground object classification and recognition method can also be implemented by the server 104 or the terminal 102 alone, such as the terminal 102 can directly perform ground object classification and recognition on the remote sensing image, or the server 104 can obtain the remote sensing image from the data storage system and perform ground object classification and recognition on the remote sensing image.

[0030] The terminal 102 may be, but is not limited to, various desktop computers, laptop computers, smart phones, tablet computers, IoT devices, and portable wearable devices. The IoT devices may be smart speakers, smart TVs, smart air conditioners, smart vehicle-mounted devices, etc. The portable wearable devices may be smart watches, smart bracelets, head-mounted devices, etc. The server 104 may be implemented as an independent server or a server cluster consisting of multiple servers, or may be a cloud server.

[0031] In an exemplary embodiment, Figure 2 As shown, a multi-scale super-pixel level ground object classification and recognition method is provided. The method is executed by a computer device, and can be executed by a computer device such as a terminal or a server alone, or can be executed by a terminal and a server together. In the embodiment of the present application, the method is applied to Figure 1 The server 104 in the example is used as an example to illustrate, including the following steps 200 to 209. Among them: Step 200: Acquire remote sensing images, and classify the remote sensing images into optical remote sensing images and radar remote sensing images according to the acquisition sources of the remote sensing images.

[0032] Step 201: Preprocess the optical remote sensing image and the radar remote sensing image respectively.

[0033] Step 202: performing principal component analysis on the preprocessed optical remote sensing image and radar remote sensing image respectively.

[0034] Step 203: The optical remote sensing image after principal component analysis and the radar remote sensing image after principal component analysis that meet the set requirements are used as principal component feature images. The principal component feature images may also be referred to as principal component images.

[0035] Step 204: normalize the principal component feature image to obtain a normalized result.

[0036] Step 205: combining the normalization result and the row and column position information of the pixels in the principal component feature image, performing distance measurement for the distribution of the cluster centers of each pixel, and obtaining the superpixel clustering target criterion.

[0037] Step 206: Determine multiple segmentation sizes, and use the segmentation sizes to perform superpixel segmentation on the principal component feature image based on the superpixel clustering target criterion to obtain a segmentation result. The segmentation sizes of superpixels of multiple scales can be determined in combination with research data and specific application scenarios.

[0038] Step 207: Extract feature images based on the preprocessed remote sensing images (ie, the preprocessed optical remote sensing images and radar remote sensing images). The feature images at least include texture feature images and polarization feature images.

[0039] Step 208: Take the pixel mean of all pixels contained in each superpixel in the segmentation result corresponding to different feature images as the feature value of each superpixel in each feature image.

[0040] Step 209: Using the feature values ​​of each superpixel on different feature images as input to the machine learning model to achieve object classification and recognition.

[0041] Implementing the above steps 200 to 209 can integrate rich attribute features from different remote sensing data sources, achieve more sophisticated and flexible processing of remote sensing images, generate superpixel areas that are more closely aligned with the actual boundaries of ground objects, and effectively reduce the common edge blurring problems in the classification process. It can not only enhance the adaptability and robustness of superpixel segmentation, but also enable it to better serve diverse classification tasks and complex scenarios, thereby significantly improving the accuracy and efficiency of ground object classification and identification.

[0042] In another exemplary embodiment of the present application, in order to reduce information redundancy while reducing the number of input images and improving segmentation efficiency, the principal component feature image with a cumulative contribution of more than 90% can be used to participate in the superpixel segmentation of the target area. Based on this, the above step 203 can be replaced by: taking the optical remote sensing image after principal component analysis with a cumulative contribution of more than 90% and the radar remote sensing image after principal component analysis as the principal component feature image.

[0043] In another exemplary embodiment of the present application, the process of formulating the superpixel clustering target criterion in step 204 can be described as follows: The side length of each superpixel is taken as the distance between two adjacent seed points. Based on this, the search space range of various sub-points can be set as .

[0044] like Figure 3 As shown, each principal component feature image involved in the segmentation is normalized, and the distance measurement of each pixel cluster center distribution is performed by combining the normalized result and the pixel row and column position information. The calculation formula is as follows: (4) (5) (6) in, represents the distance in the principal component feature space, Indicates the number of principal component feature images involved in segmentation, Pixels With pixels In the The eigenvalues ​​on the normalized principal component eigenimage, represents the standardized spatial distance, represents the spatial distance coefficient, and Represent the pixel row and column positions respectively, It is the joint distance measurement result between each principal component feature and the pixel row and column position space.

[0045] In another exemplary embodiment of the present application, the implementation process of step 205 can be described as follows: 1) Using the segmentation size, perform maximum scale superpixel segmentation on the principal component feature image based on the superpixel clustering target criterion to obtain superpixel blocks. Among them: 1-1) Arrange the segmentation sizes in descending order according to the superpixel size to obtain a scale sequence for scale-by-scale segmentation.

[0046] 1-2) For the first segmentation size in the scale sequence, complete the seed point initialization and the seed point gradient minimum position adjustment.

[0047] 1-3), for each seed point, determine the search space range (e.g. ) is the joint distance from the pixel point to the seed point in the principal component feature image. For example, the joint distance from the pixel point to the seed point can be calculated by combining the pixel value of each pixel in the normalized principal component image and its spatial position according to formula (4) to formula (6).

[0048] 1-4), assign a cluster center to each pixel point according to the minimum joint distance, perform iterative clustering, and use the average position of all pixels in each superpixel block in each iteration as the new cluster center, determine the changes in the eigenvalues ​​and pixel positions of all cluster centers in each principal component feature image after two consecutive iterations, and when the changes in the eigenvalues ​​and pixel positions of the cluster centers in each principal component feature image are less than the first set threshold When , the iteration stops and the super pixel block is obtained.

[0049] Furthermore, the first threshold value can be set Collectively referred to as the cluster center change threshold, based on this, the cluster center change calculation formula can be expressed as: (7) (8) in, To end the The difference matrix between the principal component eigenvalues ​​and pixel spatial position coordinates of the new and old cluster centers in the iteration, is the total number of cluster centers, is the number of principal component images involved in segmentation, , Respectively sequence The first The new and old cluster centers are in The eigenvalues ​​on the principal component images are is the spatial distance coefficient, , Represents the two adjacent iterations The pixel locations of the new and old cluster centers, The single-scale superpixel segmentation is terminated at The cluster center change value of the iteration.

[0050] The essence of the above steps 1-2) to 1-4) is a Figure 4 The single-scale superpixel clustering process shown.

[0051] 2) Perform multi-scale superpixel segmentation on the superpixel block to obtain the segmentation result. 2-1) Determine the distance index from all pixels in the superpixel block to the cluster center.

[0052] 2-2) When the distance index is greater than or equal to the second set threshold When the distance index is less than the second set threshold, continue to segment the superpixel block until The clustering result is obtained.

[0053] 2-3) Perform post-segmentation processing on the clustering results to obtain the segmentation results.

[0054] The essence of the above steps 2-1) to 2-3) is to calculate the distance index from all pixels in each superpixel block obtained by segmentation to the cluster center based on the maximum scale superpixel segmentation. Figure 5 As shown, if the distance index corresponding to the super pixel block is less than the second set threshold , then the super pixel block will no longer be processed in the subsequent scale segmentation steps, and other super pixel blocks exceeding the set threshold will continue to be segmented. The loop iterates and processes the areas that need to be segmented at each scale.

[0055] Further, the second threshold is set It is collectively called the super pixel block no longer segmentation threshold. Based on this, the distance index calculation formula is: (9) (10) in, For the The difference matrix between the principal component eigenvalues ​​of each pixel and the cluster center and the pixel spatial position coordinates in the superpixel block, is the total number of pixels contained in the superpixel block, and They are the first pixels and The cluster center of the superpixel block is The eigenvalues ​​on the principal component images are and Respectively represent the first pixels and The pixel position of the cluster center of each superpixel block, Reflects the The distance from all pixels in a superpixel block to the cluster center.

[0056] In another exemplary embodiment of the present application, for abnormal situations such as multi-connectivity, superpixel size being too small, and a single superpixel being divided into multiple discontinuous superpixels in the clustering results, clustering post-processing can be performed by enhancing connectivity to reallocate discontinuous superpixels or superpixels that are too small to labels of neighboring superpixels. Region, process the associated connected domains with the same label as the main connected domain, count the label values ​​of the 4-neighborhood points of the contour points of each associated connected domain, and update the pixels of the connected domain to be processed with the label values ​​that the contour points touch the most. Among them, the abnormalities in the clustering results are mainly obtained through visual inspection or label inspection. Assuming that the pixels that should have belonged to the same superpixel are mistakenly divided into different superpixels, and these superpixels are not continuous in space, it can be determined that a single superpixel has been divided into multiple discontinuous superpixels. Alternatively, if adjacent pixels have different labels, and the superpixels corresponding to these labels should be visually continuous, it can also be determined that discontinuous segmentation has occurred.

[0057] Based on this, the implementation process of steps 2-3) can be described as: Step 2-31), determine whether there are any abnormalities in the clustering results. Abnormalities include multi-connectivity in the clustering results, superpixel size smaller than a set value, and a single superpixel being divided into multiple discontinuous superpixels.

[0058] Step 2-32), when there are abnormalities in the clustering results, post-segmentation processing is performed by enhancing connectivity to obtain the segmentation results.

[0059] In another exemplary embodiment of the present application, since the proposed multi-scale superpixel segmentation method limits the search space of cluster centers, a small number of fused principal component features can be used to perform local clustering of a small number of categories. However, for the classification and recognition of objects in the entire area, texture features and polarization features can be further extracted based on the original image to jointly form a multi-dimensional feature space for classification. The pixel mean of the feature image corresponding to all pixels in each superpixel block is taken as the feature value of each superpixel on the feature image, which is used as the input data of a machine learning model (such as a support vector machine, etc.) to realize the classification and recognition of objects.

[0060] Based on the above description, this application first performs principal component dimensionality reduction on multi-source remote sensing images according to the optical and radar categories, and takes the principal component feature image with a cumulative contribution of more than 90% as the segmentation input data (that is, the principal component analysis fusion result of multi-source remote sensing data is used as the superpixel segmentation input data), and combines the selected principal component feature space and pixel position space to formulate superpixel clustering target criteria. Secondly, combined with the distribution characteristics of the objects in the target area, multiple superpixel segmentation sizes are determined, and the superpixel sizes are segmented from large to small. The fused multi-source remote sensing feature space and the spatial position distance of each pixel are combined to formulate the judgment rules for whether the segmentation of each scale should be stopped and whether each superpixel block should continue to be segmented, and post-segmentation processing is performed. Finally, for different image features, the feature mean of all pixels corresponding to each superpixel is taken as the feature value of each dimension to form a multi-dimensional feature space for machine learning object classification and recognition.

[0061] In summary, this application proposes a new multi-source remote sensing data fusion framework for multi-scale superpixel segmentation, which can fully consider the complementarity and differences between different data sources, and accurately describe the characteristics of ground objects while reducing the amount of clustering calculations. Secondly, the use of multi-scale segmentation strategies can better adapt to complex scenes, improve the fit between segmentation boundaries and the true contours of different ground object categories, and effectively reduce over-segmentation, thereby ensuring the reliability of segmentation results and improving the accuracy of subsequent tasks such as ground object classification and target recognition.

[0062] In an exemplary embodiment, a computer device is provided. The computer device may be a server or a terminal. The internal structure diagram thereof may be as follows: Figure 6 As shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, referred to as I / O) and a communication interface. The processor, the memory and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store land object classification and identification data. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a multi-scale super-pixel-level land object classification and identification method is implemented.

[0063] Those skilled in the art will understand that Figure 6The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components. In an exemplary embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the steps in the above-mentioned method embodiments when executing the computer program.

[0064] In an exemplary embodiment, a computer-readable storage medium is provided, storing a computer program, and when the computer program is executed by a processor, the steps in the above method embodiments are implemented.

[0065] In an exemplary embodiment, a computer program product is provided, including a computer program, and when the computer program is executed by a processor, the steps in the above method embodiments are implemented.

[0066] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.

[0067] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM may be in various forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).

[0068] The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. The non-relational database may include a distributed database based on blockchain, etc., but is not limited thereto. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but is not limited thereto.

[0069] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0070] This article uses specific examples to illustrate the principles and implementation methods of this application. The description of the above embodiments is only used to help understand the method and core ideas of this application. At the same time, for those skilled in the art, according to the ideas of this application, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting this application.

Claims

1. A multi-scale super-pixel level ground object classification and recognition method, characterized in that: The multi-scale super-pixel level ground object classification and recognition method comprises: Acquiring remote sensing images, and dividing the remote sensing images into optical remote sensing images and radar remote sensing images according to the acquisition sources of the remote sensing images; Preprocessing the optical remote sensing image and the radar remote sensing image respectively; Conduct principal component analysis on the preprocessed optical remote sensing images and radar remote sensing images respectively; The optical remote sensing image after principal component analysis and the radar remote sensing image after principal component analysis that meet the set requirements are used as principal component feature images; Normalizing the principal component feature image to obtain a normalized result; Combining the normalization result and the row and column position information of pixels in the principal component feature image, a distance measurement of the distribution of cluster centers of each pixel is performed to obtain the superpixel clustering target criterion; Determine a plurality of segmentation sizes, and use the segmentation sizes to perform superpixel segmentation on the principal component feature image based on the superpixel clustering target criterion to obtain a segmentation result; Extracting feature images based on the preprocessed optical remote sensing image and the radar remote sensing image; the feature images at least include texture feature images and polarization feature images; The pixel mean of all pixels contained in each superpixel in the segmentation result corresponding to different feature images is taken as the feature value of each superpixel on each feature image; The feature value of each superpixel on different feature images is used as the input of the machine learning model to achieve ground object classification and recognition.

2. The multi-scale super-pixel level ground object classification and recognition method according to claim 1, characterized in that: The optical remote sensing image after principal component analysis and the radar remote sensing image after principal component analysis that meet the set requirements are used as the principal component feature image, including: The optical remote sensing image after principal component analysis and the radar remote sensing image after principal component analysis with cumulative contributions exceeding 90% are used as the principal component feature images.

3. The multi-scale super-pixel level ground object classification and recognition method according to claim 1, characterized in that: The segmentation size is adopted, and superpixel segmentation is performed on the principal component feature image based on the superpixel clustering target criterion to obtain a segmentation result, including: Using the segmentation size, and based on the superpixel clustering target criterion, the principal component feature image is segmented into maximum-scale superpixels to obtain superpixel blocks; Multi-scale superpixel segmentation is performed on the superpixel block to obtain the segmentation result.

4. The multi-scale super-pixel level ground object classification and recognition method according to claim 3, characterized in that: Using the segmentation size, the principal component feature image is segmented into maximum-scale superpixels based on the superpixel clustering target criterion to obtain superpixel blocks, including: Arrange the segmentation sizes in descending order according to superpixel sizes to obtain a scale sequence; For the first segmentation size in the scale sequence, completing seed point initialization and seed point gradient minimum position adjustment; For each seed point, determining the joint distance from the pixel point in the principal component feature image to the seed point within a search space range based on the superpixel clustering target criterion; the search space range is determined based on the distance between two adjacent seed points; A cluster center is assigned to each pixel point according to the minimum joint distance, and iterative clustering is performed to determine the changes in the eigenvalues ​​and pixel positions of all cluster centers in each principal component feature image after two adjacent iterations. When the changes in the eigenvalues ​​and pixel positions of the cluster centers in each principal component feature image are less than the first set threshold, the iteration is stopped to obtain a superpixel block.

5. The multi-scale super-pixel level ground object classification and recognition method according to claim 3, characterized in that: Using the segmentation size to perform multi-scale superpixel segmentation on the superpixel block to obtain the segmentation result, including: Determine the distance index from all pixels in the superpixel block to the cluster center; When the distance index is greater than or equal to a second set threshold, continue to segment the superpixel block until the distance index is less than the second set threshold, and obtain a clustering result; The clustering result is subjected to post-segmentation processing to obtain the segmentation result.

6. The multi-scale super-pixel level ground object classification and recognition method according to claim 5, characterized in that: The clustering result is subjected to post-segmentation processing to obtain the segmentation result, including: Determine whether there are any abnormalities in the clustering results; the abnormalities include multi-connectivity in the clustering results, superpixel size being smaller than a set value, and a single superpixel being divided into multiple discontinuous superpixels; When there are abnormalities in the clustering result, post-segmentation processing is performed by enhancing connectivity to obtain the segmentation result.

7. The multi-scale super-pixel level ground object classification and recognition method according to claim 1, characterized in that: The machine learning model adopts support vector machine.

8. A computer device comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the multi-scale super-pixel level object classification and recognition method according to any one of claims 1 to 7.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the multi-scale super-pixel level ground object classification and recognition method described in any one of claims 1 to 7 is implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the multi-scale super-pixel level ground object classification and recognition method described in any one of claims 1 to 7 is implemented.

Images