Super-pixel-level hyperspectral image mikania micrantha identification method

Abstract

The embodiment of the invention provides a super-pixel-level hyperspectral image mikania micrantha recognition method, and the method comprises the steps: obtaining corresponding super-pixel-level first sub-images through carrying out the preset-scale segmentation of a target image, further determining a self-adaptive Gaussian kernel, obtaining a wave band similarity matrix through employing the wave band in each first sub-image as a unit, and carrying out the recognition of the mikania micrantha. According to the method, a plurality of sub-images are obtained, super-pixel-level spectral clustering is carried out, a plurality of band clusters are obtained, data processing and screening are carried out on bands in each band cluster, comprehensive ranking indexes are determined, and corresponding sub-images are obtained to identify a target object, so that segmentation, data processing and data analysis are carried out on the target image in a spatial dimension and a band dimension; according to the method, a clustering mode and an index ranking mode are combined, a corresponding target wave band is selected, then object recognition is achieved, the problem that combination of the two types of methods is obstructed is solved, effectiveness of object recognition and physical interpretability of features are guaranteed, and hyperspectral image object recognition under the condition of low data volume is achieved.

Description

Technical Field

[0001] The present application relates to the field of remote sensing, and in particular to a method for identifying Mikania micrantha in super-pixel-level hyperspectral images. Background Art

[0002] Hyperspectral images contain rich spatial information and detailed spectral information, which can be used to detect, identify and analyze objects in the field of remote sensing. However, the manual labeling of hyperspectral images is too labor-intensive, resulting in insufficient data sets to support the recognition of target objects. In addition, the amount of computation required in the data processing of hyperspectral images is usually high.

[0003] Therefore, object recognition based on hyperspectral images requires data dimensionality reduction. Band selection is an optional way to achieve data dimensionality reduction because it can maintain the physical meaning of the data.

[0004] However, in the process of band selection, due to the insufficient amount of data in the aforementioned dataset, band selection can be achieved through unsupervised methods, including clustering-based or ranking-based methods. These two types of methods have different limitations. For example, the clustering-based method has a better recognition effect, but the parameter adjustment is difficult and inefficient, while the feature ranking-based method cannot fully express the data connection between each band. Therefore, these two methods can be combined, but there are obstacles in the combination process and they cannot be simply combined. Therefore, a super-pixel-level hyperspectral image Mikania micrantha recognition method is needed to realize the combination of clustering-based and ranking-based recognition methods, improve the effectiveness of recognition, and reduce the required amount of data and the complexity of data processing. Summary of the invention

[0005] The present application provides a super-pixel-level hyperspectral image Mikania micrantha recognition method, which is used to realize the combination of two types of recognition methods, namely clustering-based and ranking-based, to solve the problem of obstacles in combining the two types of methods and the problem that recognition effect and physical interpretability cannot be achieved simultaneously in current recognition methods.

[0006] In a first aspect, the present application provides a super-pixel-level hyperspectral image Mikania micrantha recognition method, the method comprising:

[0007] Performing multi-scale super-pixel segmentation on the target image to obtain a plurality of super-pixel-level first sub-images;

[0008] Wherein, the first sub-image is used to indicate the spatial structure information of each object in the target image;

[0009] According to each of the first sub-images, an adaptive Gaussian kernel is determined to obtain a band similarity matrix;

[0010] Performing superpixel-level spectral clustering according to the band similarity matrix to obtain a plurality of band clusters corresponding to each of the first sub-images;

[0011] Determining a comprehensive ranking index, and determining a target band in each of the band clusters according to the comprehensive ranking index;

[0012] In each of the first sub-images, a second sub-image is determined according to the target band, and a target object in the target image is identified according to each of the second sub-images.

[0013] As an optional implementation manner, performing multi-scale superpixel segmentation on the target image to obtain a plurality of superpixel-level first sub-images includes:

[0014] Determine a segmentation scale factor, and determine an average number of pixels included in each first sub-image according to the segmentation scale factor and the spatial resolution of the target image;

[0015] Determining a first number of first sub-images according to the average number of pixels and the total number of pixels of the target image;

[0016] According to a simple non-iterative clustering algorithm, superpixel segmentation is performed on the target image to obtain first sub-images at the first number of superpixel levels;

[0017] Each of the first sub-images contains a second number of bands.

[0018] As an optional implementation manner, determining an adaptive Gaussian kernel according to each of the first sub-images to obtain a band similarity matrix includes:

[0019] Determining a second number of band images corresponding to each of the first sub-images according to a second number of bands of each of the first sub-images;

[0020] Determine a band vector according to each of the band images, and determine a fully connected graph according to each of the band vectors;

[0021] The nodes of the fully connected graph are the band vectors, and the edges of the fully connected graph include distance parameters between the band vectors.

[0022] Determine a minimum spanning tree according to the fully connected graph, and determine a band similarity parameter according to the edge weights of the minimum spanning tree;

[0023] A band similarity matrix is ​​generated according to the band similarity parameters.

[0024] As an optional implementation manner, performing superpixel-level spectral clustering according to the band similarity matrix to obtain a plurality of band clusters corresponding to each of the first sub-images includes:

[0025] Determining a normalized Laplace matrix according to the band similarity matrix;

[0026] Performing eigendecomposition according to the normalized Laplace matrix to obtain a third number of eigenvalues ​​and corresponding eigenvectors, and obtaining a band information matrix;

[0027] The first dimension of the band information matrix is ​​the second number, and the second dimension of the band information matrix is ​​the third number;

[0028] The band information matrix is ​​standardized in units of rows to obtain a standardized band information matrix, thereby obtaining a fourth number of band clusters.

[0029] As an optional implementation, the comprehensive ranking index includes information entropy, distance factor and superpixel clustering variance ratio;

[0030] The comprehensive ranking indicators are as follows:

[0031] In the calculation band of the target band cluster, the value and probability distribution of each pixel are determined, and the information entropy of the calculation band is obtained according to the value and probability distribution of each pixel through an information entropy calculation formula;

[0032] and determining a first cluster distance from the calculation band to the center of the target band cluster and a second cluster distance from the calculation band to the centers of other band clusters, and subtracting the sum of the second cluster distances from the first cluster distance to obtain a distance factor for the calculation band;

[0033] And, clustering each of the first sub-images to obtain a plurality of superpixel clusters, and using each of the superpixel clusters as a classification label, and determining the superpixel cluster variance ratio of the calculation band according to the ratio of the between-class variance to the within-class variance;

[0034] Determining the target band in each of the band clusters according to the comprehensive ranking index includes:

[0035] According to the sum of the normalized values ​​of the information entropy, the distance factor and the superpixel clustering variance ratio, the discriminative score of the comprehensive ranking index is determined, and ranking is performed according to the discriminative score, and a preset number of target bands in each of the band clusters are selected.

[0036] In a second aspect, the present application further provides a device to implement the method described in the first aspect, the device comprising:

[0037] A processing module, used for performing multi-scale super-pixel segmentation on the target image to obtain a plurality of super-pixel-level first sub-images;

[0038] Wherein, the first sub-image is used to indicate the spatial structure information of each object in the target image;

[0039] The determination module is used to determine an adaptive Gaussian kernel according to each of the first sub-images to obtain a band similarity matrix;

[0040] The processing module is used to perform super-pixel level spectral clustering according to the band similarity matrix to obtain a plurality of band clusters corresponding to each of the first sub-images;

[0041] The determination module is used to determine the comprehensive ranking index;

[0042] The processing module is used to determine the target band in each of the band clusters according to the comprehensive ranking index;

[0043] The determining module is used to determine a second sub-image in each of the first sub-images according to the target band;

[0044] The processing module is used to identify the target object in the target image according to each of the second sub-images.

[0045] As an optional implementation manner, the processing module performs multi-scale superpixel segmentation on the target image to obtain a plurality of superpixel-level first sub-images in a specific manner including:

[0046] Determine a segmentation scale factor, and determine an average number of pixels included in each first sub-image according to the segmentation scale factor and the spatial resolution of the target image;

[0047] Determining a first number of first sub-images according to the average number of pixels and the total number of pixels of the target image;

[0048] According to a simple non-iterative clustering algorithm, the target image is segmented into superpixels to obtain first sub-images at the first number of superpixel levels;

[0049] Each of the first sub-images contains a second number of bands.

[0050] As an optional implementation manner, the processing module determines the adaptive Gaussian kernel according to each of the first sub-images to obtain a specific manner of the band similarity matrix, including:

[0051] Determining a second number of band images corresponding to each of the first sub-images according to a second number of bands of each of the first sub-images;

[0052] Determine a band vector according to each of the band images, and determine a fully connected graph according to each of the band vectors;

[0053] The nodes of the fully connected graph are the band vectors, and the edges of the fully connected graph include distance parameters between the band vectors.

[0054] Determine a minimum spanning tree according to the fully connected graph, and determine a band similarity parameter according to the edge weights of the minimum spanning tree;

[0055] A band similarity matrix is ​​generated according to the band similarity parameters.

[0056] As an optional implementation manner, the processing module performs superpixel-level spectral clustering according to the band similarity matrix to obtain a plurality of band clusters corresponding to each of the first sub-images, including:

[0057] Determining a normalized Laplace matrix according to the band similarity matrix;

[0058] Performing eigendecomposition according to the normalized Laplace matrix to obtain a third number of eigenvalues ​​and corresponding eigenvectors, and obtaining a band information matrix;

[0059] The first dimension of the band information matrix is ​​the second number, and the second dimension of the band information matrix is ​​the third number;

[0060] The band information matrix is ​​standardized in units of rows to obtain a standardized band information matrix, thereby obtaining a fourth number of band clusters.

[0061] As an optional implementation, the comprehensive ranking index includes information entropy, distance factor and superpixel clustering variance ratio;

[0062] The specific method of determining the comprehensive ranking index by the determination module includes:

[0063] In the calculation band of the target band cluster, the value and probability distribution of each pixel are determined, and the information entropy of the calculation band is obtained according to the value and probability distribution of each pixel through an information entropy calculation formula;

[0064] and determining a first cluster distance from the calculation band to the center of the target band cluster and a second cluster distance from the calculation band to the centers of other band clusters, and subtracting the sum of the second cluster distances from the first cluster distance to obtain a distance factor for the calculation band;

[0065] And, clustering each of the first sub-images to obtain a plurality of superpixel clusters, and using each of the superpixel clusters as a classification label, and determining the superpixel cluster variance ratio of the calculation band according to the ratio of the between-class variance to the within-class variance;

[0066] The processing module determines a specific method of determining the target band in each of the band clusters according to the comprehensive ranking index, including:

[0067] According to the sum of the normalized values ​​of the information entropy, the distance factor and the superpixel clustering variance ratio, the discriminative score of the comprehensive ranking index is determined, and ranking is performed according to the discriminative score, and a preset number of target bands in each of the band clusters are selected.

[0068] In a third aspect, the present application provides an electronic device, comprising: a processor, and a memory communicatively connected to the processor;

[0069] The memory stores computer-executable instructions;

[0070] The processor executes the computer-executable instructions stored in the memory to implement the method described in the first aspect.

[0071] In a fourth aspect, the present application provides a computer-readable storage medium, wherein the computer-readable storage medium stores computer-executable instructions, and when the computer-executable instructions are executed by a processor, they are used to implement the method described in the first aspect.

[0072] The present application provides a super-pixel-level hyperspectral image Mikania micrantha recognition method. By means of super-pixel-level segmentation, the target image is segmented into super-pixels at a preset scale to obtain the first super-pixel-level sub-images corresponding to multiple scales. The spatial structure information of each object in the target image is expressed by the first sub-image, and an adaptive Gaussian kernel is further determined. The band similarity matrix is ​​obtained with the band in each first sub-image as a unit, and super-pixel-level spectral clustering is performed to obtain multiple band clusters. Data processing and screening are performed on the bands in each band cluster to determine a comprehensive ranking index, and one or more target bands corresponding to each band cluster are further determined. Furthermore, in the first sub-image, a second sub-image corresponding to each target band is determined, and the target object is identified according to the determined second sub-image. Thus, the target image is segmented, data processed and analyzed in the spatial dimension and the band dimension, and the clustering method and the index ranking method are combined to select the corresponding target band, thereby realizing object recognition, solving the problem of obstacles in combining the two methods, ensuring the effectiveness of object recognition and the physical interpretability of features, and realizing hyperspectral image object recognition with low data volume. BRIEF DESCRIPTION OF THE DRAWINGS

[0073] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0074] Figure 1 It is a flow chart of a method for identifying Mikania micrantha in a super-pixel-level hyperspectral image disclosed in an embodiment of the present application;

[0075] Figure 2 It is a flow chart of a method for identifying Mikania micrantha in a super-pixel-level hyperspectral image disclosed in an embodiment of the present application;

[0076] Figure 3 It is a flow chart of a method for identifying Mikania micrantha in a super-pixel-level hyperspectral image disclosed in an embodiment of the present application;

[0077] Figure 4 It is a flow chart of a method for identifying Mikania micrantha in a super-pixel-level hyperspectral image disclosed in an embodiment of the present application;

[0078] Figure 5 It is a structural schematic diagram of a super-pixel-level hyperspectral image object recognition device disclosed in an embodiment of the present application;

[0079] Figure 6 It is a structural schematic diagram of an electronic device disclosed in an embodiment of the present application.

[0080] The above-mentioned drawings have shown clear embodiments of the present application, which will be described in more detail hereinafter. These drawings and textual descriptions are not intended to limit the scope of the concept of the present application in any way, but to illustrate the concept of the present application to those skilled in the art by referring to specific embodiments. DETAILED DESCRIPTION

[0081] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, the same numbers in different drawings represent the same or similar elements unless otherwise indicated. The implementations described in the following exemplary embodiments do not represent all implementations consistent with the present application. Instead, they are merely examples of devices and methods consistent with some aspects of the present application as detailed in the appended claims.

[0082] Hyperspectral images contain rich spatial information and detailed spectral information, which can be used to detect, identify and analyze objects in the field of remote sensing. However, the manual labeling of hyperspectral images is too labor-intensive, resulting in insufficient data sets to support the recognition of target objects. In addition, the amount of computation required in the data processing of hyperspectral images is usually high.

[0083] Therefore, object recognition based on hyperspectral images requires data dimensionality reduction. Band selection is an optional way to achieve data dimensionality reduction because it can maintain the physical meaning of the data.

[0084] However, in the process of band selection, due to the insufficient amount of data in the aforementioned dataset, band selection can be achieved through unsupervised methods, including clustering-based or ranking-based methods. These two types of methods have different limitations. For example, the clustering-based method has a better recognition effect, but the parameter adjustment is difficult and inefficient, while the feature ranking-based method cannot fully express the data connection between each band. Therefore, these two methods can be combined, but there are obstacles in the combination process and they cannot be simply combined. Therefore, a super-pixel-level hyperspectral image Mikania micrantha recognition method is needed to realize the combination of clustering-based and ranking-based recognition methods, improve the effectiveness of recognition, and reduce the required amount of data and the complexity of data processing.

[0085] In some implementations of the present application, each method is explained by taking Mikania micrantha identification in remote sensing hyperspectral images as an example.

[0086] Mikania micrantha is an invasive plant native to Central and South America, belonging to the genus Mikania of the Asteraceae family. It is a fast-growing perennial herb with strong invasive ability, which has serious impacts on agriculture, forestry and natural ecosystems. It can crowd out local species in new environments, resulting in reduced biodiversity and damaged ecosystem service functions. Dynamic monitoring and scientific management of Mikania micrantha through hyperspectral remote sensing technology can slow down its spread and protect the health and biodiversity of ecosystems. Hyperspectral images combine rich spatial information with detailed spectral information, which can achieve accurate detection, identification and attribute analysis of Mikania micrantha. Hyperspectral image classification is the key to hyperspectral image analysis technology and the basis of other applications. However, since manual labeling of samples is time-consuming and expensive, and often contains labeling errors, the number of labeled samples is often insufficient. Therefore, using hyperspectral data for downstream tasks such as Mikania micrantha identification may encounter the Hughes phenomenon, that is, the performance decreases instead of increases as the spectral dimension increases. In addition, the high spectral dimension leads to a large amount of calculation, which makes data processing and application difficult. There are two main ways to reduce data dimensionality: feature extraction and band selection. Among them, band selection can better remove interference and highlight signal features related to the research task because it can retain the original meaning of the data. Feature extraction will transform the data, thereby destroying the physical meaning of the data, while band selection can retain the physical meaning of the data by screening the original band group and selecting a part of the bands to form a band subset.

[0087] Due to the scarcity of samples of Mikania micrantha labeled in hyperspectral images, unsupervised band selection algorithms are more suitable. The two main methods of unsupervised band selection are clustering-based and ranking-based methods. On the one hand, the ranking-based band selection method scores and ranks each band according to a certain criterion, and selects several bands with the highest scores to form the spectral subset for Mikania micrantha identification. The maximum variance principal component analysis and the maximum signal-to-noise ratio principal component analysis, these two methods use the weighted sum of the transformed basis vectors as the score of each band for ranking. The method of ranking bands using information entropy and information divergence uses matched filters and coherence estimators to rank all spectral bands. However, the ranking-based method can often only capture the local correlation between hyperspectral bands and cannot fully model the correlation between bands.

[0088] On the other hand, clustering-based methods aim to select spectral bands with low correlation for Mikania identification, explore the similarity of bands through clustering algorithms, and select representative bands in each cluster to reduce band redundancy. Band clustering methods based on spectral information divergence select the band closest to the cluster center by calculating spectral divergence. The method of clustering statistical features using K-means clustering also selects the band closest to the cluster center through clustering. Compared with ranking algorithms, clustering-based methods usually have better Mikania identification performance due to lower redundancy of their band subsets. However, the parameter adjustment difficulty and time complexity of clustering algorithms are relatively high, and the selected hyperspectral image band subsets may also be unstable.

[0089] Considering the limitations of a single method, a combined strategy based on ranking and clustering mines correlations through clustering, and then selects the desired spectral identification bands of Mikania micrantha through ranking. The method combining linear spectral unmixing and sequential clustering technology first extracts the spectrum of the sample through the unmixing method, then calculates the sample variance in each band and executes the sequential clustering strategy, and finally selects the band with large variance in each cluster. The optimal clustering band selection framework first uses a dynamic programming method to search for the best clustering result that meets a certain clustering criterion on the entire band set, and then uses a cluster ranking strategy method to select bands within each cluster. Considering the low-dimensional subspace structure of high-dimensional data, some algorithms based on sparse subspace clustering are combined into band selection. The sparse subspace clustering method uses a sparse self-representation matrix to construct a similarity matrix, and applies spectral clustering on the similarity matrix to obtain the clustering result. It is suitable for data with multiple low-dimensional subspaces and nonlinear structures. The combined strategy based on ranking and clustering can better take into account the local correlation and global heterogeneity of the hyperspectral image bands of Mikania micrantha. However, there are still many technical details to be optimized in terms of full exploration of spatial structure and in-depth mining of continuous spectra. The feature clustering and band selection effects for Mikania micrantha identification need to be improved.

[0090] In some feasible implementations, some clustering algorithms do not take into account the shape and structure of hyperspectral data, and therefore model the band correlation inaccurately. These algorithms usually assume that the data has certain specific distribution forms, which is not always appropriate for the complex shapes in actual data and may ignore important Mikania structural information in hyperspectral images.

[0091] In some feasible implementations, it is assumed that the hyperspectral band distribution conforms to the subspace property. Although this simplifies the problem processing, it ignores the complexity and diversity of the data, reducing the universality of the method for the Mikania micrantha dataset. In addition, only the first principal component is used for superpixel segmentation, which can only obtain suboptimal segmentation effects, affecting the clustering and selection effects.

[0092] In some feasible implementations, the proposed discrimination index is based on the total covariance of superpixels and the intra-superpixel covariance, which selects bands that meet the high homogeneity of superpixels and high discreteness of different superpixels. However, due to the spectral variation of Mikania micrantha in different regions, this method will amplify the influence of spectral variation. In addition, in some feasible implementations, the ranking index is too single and ignores the importance of other aspects. The properties of the selected bands cannot take into account high information content, low correlation and high separability.

[0093] In this application, in order to address the problems of not considering data shape and universality for different data sets, this application proposes an adaptive Gaussian kernel calculation method, which can adaptively adjust the Gaussian kernel parameters according to different superpixel sizes, different numbers of clusters and different spatial patterns, and therefore has better clustering adaptability for the Mikania structure of hyperspectral images.

[0094] In order to solve the problem of assuming that the distribution of hyperspectral bands conforms to the subspace properties and ignoring the variability of data, unlike the methods that have strong assumptions on the data structure, the superpixel-level adaptive spectral clustering method proposed in this application has fewer constraints on the data structure, has better spatial consistency, and enhances the integration of the spectral and spatial features of Mikania micrantha.

[0095] In order to solve the problem that the ranking index is too single and the distinguishability index is insufficient in modeling spectral variability, this application proposes superpixel clustering variance ratio as a separability index and a comprehensive ranking index band selection method to reduce the interference of spectral variability of Mikania micrantha. All bands are used for superpixel segmentation, and a heuristic formula for multi-scale superpixel segmentation parameters is proposed based on the spatial resolution of the hyperspectral image, making the selected spectral bands more reasonable.

[0096] The technical concept of the present application is to segment the target image into superpixels at a preset scale by means of superpixel segmentation, obtain the first sub-images at the superpixel level corresponding to multiple scales, express the spatial structure information of each object in the target image through the first sub-image, further determine the adaptive Gaussian kernel, take the bands in each first sub-image as the unit, obtain the band similarity matrix, perform superpixel spectral clustering, obtain multiple band clusters, perform data processing and screening on the bands in each band cluster, determine the comprehensive ranking index, and further determine one or more target bands corresponding to each band cluster. Furthermore, in the first sub-image, a second sub-image corresponding to each target band is determined, and the target object is identified according to the determined second sub-image. Thus, the target image is segmented, data processed and analyzed in the spatial dimension and the band dimension, and the clustering method and the index ranking method are combined to select the corresponding target band, thereby realizing object recognition, solving the problem of obstacles in combining the two methods, ensuring the effectiveness of object recognition and the physical interpretability of features, and realizing hyperspectral image object recognition with low data volume.

[0097] Embodiment 1

[0098] See also Figure 1 , Figure 1 1 is a flow chart of a method for identifying Mikania micrantha in a super-pixel hyperspectral image disclosed in an embodiment of the present application. Figure 1 As shown, the method includes:

[0099] S101, performing multi-scale super-pixel segmentation on the target image to obtain a plurality of super-pixel-level first sub-images;

[0100] The first sub-image is used to indicate the spatial structure information of each object in the target image;

[0101] S102, determining an adaptive Gaussian kernel according to each first sub-image, and obtaining a band similarity matrix;

[0102] S103, performing superpixel-level spectral clustering according to the band similarity matrix to obtain a plurality of band clusters corresponding to each first sub-image;

[0103] S104, determining a comprehensive ranking index, and determining a target band in each band cluster according to the comprehensive ranking index;

[0104] S105 . In each first sub-image, determine a second sub-image according to the target band, and identify the target object in the target image according to each second sub-image.

[0105] See also Figure 2 , Figure 2 is a flow chart of a method for identifying Mikania micrantha in a super-pixel-level hyperspectral image disclosed in an embodiment of the present application, Figure 2 The whole process of target band identification corresponding to the embodiment of the present application is shown. In a specific application scenario, the present application adopts a hyperspectral band selection method of Mikania micrantha based on superpixel-level adaptive spectral clustering (Superpixel-level Adaptive Spectral Clustering, SpASC), including multi-scale superpixel segmentation, superpixel-level adaptive Gaussian kernel spectral clustering and spectral band selection of comprehensive ranking indicators. The input of the SpASC method is a hyperspectral image containing Mikania micrantha objects. First, multi-scale superpixel segmentation is used to obtain segmentation results at several scales, which can extract the spatial structure information of different Mikania micrantha; then, the similarity matrix between bands is calculated using superpixel-level adaptive Gaussian kernel spectral clustering and spectral clustering is performed to obtain the band clustering results, considering different spatial patterns and the number of clusters for adaptive information extraction; finally, the comprehensive ranking index is used in each band cluster to select the ranking band of Mikania micrantha, taking into account aspects such as the amount of information, difference, and distinction.

[0106] By means of superpixel segmentation, the target image is segmented into superpixels at a preset scale to obtain the first superpixel sub-images corresponding to multiple scales. The spatial structure information of each object in the target image is expressed by the first sub-image, and the adaptive Gaussian kernel is further determined. The band similarity matrix is ​​obtained with the band in each first sub-image as a unit, and superpixel spectral clustering is performed to obtain multiple band clusters. Data processing and screening are performed on the bands in each band cluster, and a comprehensive ranking index is determined, and one or more target bands corresponding to each band cluster are further determined. Furthermore, in the first sub-image, a second sub-image corresponding to each target band is determined, and the target object is identified according to the determined second sub-image. Thus, the target image is segmented, data processed and analyzed in the spatial dimension and the band dimension. The clustering method and the index ranking method are combined to select the corresponding target band, and then object recognition is realized, which solves the problem of obstacles in combining the two methods, ensures the effectiveness of object recognition and the physical interpretability of features, and realizes hyperspectral image object recognition with low data volume.

[0107] As an optional implementation, multi-scale super-pixel segmentation is performed on the target image to obtain a plurality of super-pixel-level first sub-images, including:

[0108] Determine a segmentation scale factor, and determine an average number of pixels contained in each first sub-image according to the segmentation scale factor and the spatial resolution of the target image;

[0109] Determining a first number of first sub-images according to an average number of pixels and a total number of pixels of the target image;

[0110] According to a simple non-iterative clustering algorithm, superpixel segmentation is performed on the target image to obtain a first number of superpixel-level first sub-images;

[0111] Each first sub-image includes a second number of bands.

[0112] For multi-scale superpixel segmentation, the present application can use an improved Simple Non-Iterative Clustering (SNIC) algorithm to perform superpixel segmentation on a hyperspectral image containing Mikania micrantha objects, and segment the original hyperspectral image I into multiple superpixels. Where N SP represents the number of pixels, B represents the number of bands, represents the jth pixel vector of the ith superpixel, represents the kth band vector of the ith superpixel, n i It is SP i The number of pixels in the SNIC superpixel segmentation algorithm is the number of superpixels N. SP , can be obtained by the heuristic formula corresponding to this application, and the heuristic formula may include:

[0113]

[0114] in, represents the average number of pixels contained in the superpixel, T represents the spatial resolution of the hyperspectral image, ζ represents the scale factor, and N is the total number of pixels in the target image. By selecting different scale factors, the SNIC algorithm parameters at multiple scales can be obtained, and the corresponding superpixel segmentation results can be generated to represent the spatial structure information of different Mikania micrantha.

[0115] It should be noted that the value of the heuristic formula can be adaptively adjusted according to the specific application scenario.

[0116] In addition, in some implementations, the above-mentioned superpixel segmentation algorithm can also achieve similar effects through fast shift image segmentation, watershed segmentation, etc.

[0117] By determining the segmentation scale factor and according to the spatial resolution of the target image, the average number of pixels of the first sub-image and the number of first sub-images are determined through a heuristic data determination method, and the target image is divided into super-pixel-level sub-images with a basically uniform data volume, thereby reducing the difficulty of subsequent data processing, improving the effectiveness of segmentation, ensuring the effectiveness of the information contained in each first sub-image, and then ensuring the effectiveness of object recognition, and realizing hyperspectral image object recognition with low data volume.

[0118] See also Figure 3, Figure 3 1 is a flow chart of a method for identifying Mikania micrantha in a super-pixel hyperspectral image disclosed in an embodiment of the present application. Figure 3 As shown, as an optional implementation, according to each first sub-image, an adaptive Gaussian kernel is determined to obtain a band similarity matrix, including:

[0119] Determine a second number of band images corresponding to each first sub-image according to the second number of bands of each first sub-image;

[0120] According to each band image, a band vector is determined, and a fully connected graph is determined according to each band vector;

[0121] Among them, the nodes of the fully connected graph are band vectors, and the edges of the fully connected graph include the distance parameters between each band vector;

[0122] According to the fully connected graph, the minimum spanning tree is determined, and according to the edge weight of the minimum spanning tree, the band similarity parameter is determined;

[0123] Generate a band similarity matrix based on the band similarity parameters.

[0124] Furthermore, according to each band similarity matrix, the band similarity matrix corresponding to the first sub-image at each scale may be aggregated to obtain a comprehensive, aggregated band similarity matrix.

[0125] It should be noted that the distance parameter may also be calculated using other distance metrics, such as the Mahalanobis distance.

[0126] By determining the band vectors corresponding to each band in the first sub-image, a fully connected graph is generated, and then the distance relationship between the band vectors is indicated according to the fully connected graph, and then a minimum spanning tree is generated to achieve the generation of an adaptive Gaussian kernel, and then the current band similarity parameters can be determined to generate a band similarity matrix. Further, according to the band similarity matrix corresponding to the first sub-image at each segmentation scale, an aggregation matrix can be determined, so as to perform super-pixel-level adaptive spectral clustering, provide a basis for subsequent data processing, ensure the effectiveness of object recognition, and realize hyperspectral image object recognition under low data volume.

[0127] See also Figure 4 , Figure 4 1 is a flow chart of a method for identifying Mikania micrantha in a super-pixel hyperspectral image disclosed in an embodiment of the present application. Figure 4 As shown, as an optional implementation, superpixel-level spectral clustering is performed according to the band similarity matrix to obtain multiple band clusters corresponding to each first sub-image, including:

[0128] According to the band similarity matrix, a normalized Laplace matrix is ​​determined;

[0129] Performing eigendecomposition according to the normalized Laplace matrix to obtain the third number of eigenvalues ​​and corresponding eigenvectors, and obtaining a band information matrix;

[0130] Wherein, the first dimension of the band information matrix is ​​the second quantity, and the second dimension of the band information matrix is ​​the third quantity;

[0131] The band information matrix is ​​normalized in units of rows to obtain a normalized band information matrix, thereby obtaining a fourth number of band clusters.

[0132] Specifically, the above two implementations can be divided into three parts: superpixel-level adaptive similarity calculation, multi-scale aggregation, and spectral clustering. First, the superpixel-level adaptive similarity calculation uses an adaptive Gaussian kernel (AGK) to calculate the band similarity matrix for each superpixel separately, such as Figure 3 As shown, at one scale, each superpixel adaptively obtains a Gaussian kernel parameter related to the size of Mikania micrantha through the AGK module for Gaussian similarity calculation.

[0133] like Figure 4 As shown in Figure 1, the AGK module first uses the band vectors to construct a fully connected graph, where the nodes are the band vectors and the weights of the edges between the nodes are the Euclidean distances between the bands. The j-th band vector in With the kth band vector The band distance between them is calculated as shown in the formula:

[0134]

[0135] Use distance to build the graph structure, and use Kruskal algorithm to build the minimum spanning tree, sort the edge weights of the tree, and then take the median of the first K largest edge weights as The calculation method of band similarity of Mikania micrantha hyperspectral image using adaptive Gaussian kernel is shown in the formula:

[0136]

[0137] The band similarity matrices of all superpixels are averaged to obtain a band similarity matrix Pζ at a certain scale, as shown in the formula:

[0138]

[0139] Next, the similarity matrices of multiple scales can be aggregated, as shown in the following formula:

[0140]

[0141] Where v represents the number of scales.

[0142] Furthermore, the spectral clustering algorithm and the similarity matrix O provided in the previous embodiment can be used to cluster the bands of the hyperspectral image of Mikania micrantha. The module first calculates the similarity matrix D based on O and obtains the Laplace matrix L. The solution of the normalized cut is expressed as L sym and the quadratic form of the eigenvectors corresponding to the K smallest eigenvalues. Therefore, normalizing L gives the Laplace matrix L sym , perform eigenvalue decomposition to obtain M and Λ, extract the eigenvectors corresponding to the smallest K eigenvalues ​​from the matrix M, and form a matrix V∈R B×K , where B represents the number of bands. Then, the rows of V are normalized to unit length, and these normalized rows are stacked to construct the matrix Y∈R B×K To represent the information of the graph, each row can be regarded as a K-dimensional graph node. The K-means clustering algorithm is applied to the matrix Y to obtain K band clusters.

[0143] In addition, the spectral clustering method can also be replaced by some commonly used clustering methods, such as K-means clustering and density clustering after adaptive adjustment according to the application scenario of this application.

[0144] Through the band similarity matrix under the aforementioned first sub-image, in this embodiment, the band similarity matrix can also be an aggregation matrix determined according to the band similarity matrix corresponding to the first sub-image under each segmentation scale, and then by processing the matrix data, a band information matrix is ​​obtained, thereby determining the band cluster, providing a data basis for the combination of subsequent ranking methods, solving the problem of obstacles in combining the two methods, ensuring the effectiveness of object recognition and the physical interpretability of features, and realizing hyperspectral image object recognition with low data volume.

[0145] As an optional implementation, the comprehensive ranking index includes information entropy, distance factor and superpixel clustering variance ratio;

[0146] Determine comprehensive ranking indicators, including:

[0147] In the calculation band of the target band cluster, the value and probability distribution of each pixel are determined, and the information entropy of the calculation band is obtained according to the value and probability distribution of each pixel through the information entropy calculation formula;

[0148] and determining a first cluster distance from the calculation band to the center of the target band cluster and a second cluster distance from the calculation band to the centers of other band clusters, and subtracting the sum of the second cluster distances from the first cluster distance to obtain a distance factor for the calculation band;

[0149] and, clustering each first sub-image to obtain a plurality of superpixel clusters, and using each superpixel cluster as a classification label, and determining a superpixel cluster variance ratio of a calculation band according to a ratio of an inter-class variance to an intra-class variance;

[0150] According to the comprehensive ranking indicators, the target bands in each band cluster are determined, including:

[0151] According to the sum of the normalized values ​​of information entropy, distance factor and superpixel clustering variance ratio, the discriminative score of the comprehensive ranking index is determined, and the ranking is performed according to the discriminative score, and a preset number of target bands in each band cluster are selected.

[0152] Information entropy h∈R B Calculation. In the lth band, the pixel value can be regarded as a discrete random variable Y l , the value range is The corresponding probability distribution is expressed as The pixel value in the lth band is The information entropy of the lth band of the Mikania micrantha hyperspectral image is as shown in the formula:

[0153]

[0154] Where N is the number of pixels, is a random variable Y l Pick The amount of information when the random variable Y l The mathematical expectation of the amount of information. The larger the information entropy, the higher the uncertainty of the band and the greater the amount of information contained.

[0155] The distance factor γ is calculated. The K-means clustering in the last step of spectral clustering can obtain the center of each cluster and the distance matrix. The elements in the distance matrix represent the distance between the band and the cluster center. The distance matrix is ​​represented by C∈R B×K , C i,j Represents the distance between the i-th band and the j-th cluster center. According to the K-means algorithm, the column index corresponding to the minimum value of each row in the distance matrix is ​​the cluster label of the band. Calculate each band cluster label cluster(l), which represents the cluster to which the l-th band belongs. By calculating the distance between each band and the center of its cluster and the distance to other cluster centers, the distance factor γ∈R of the spectral band of the Mikania image can be obtained. B , calculated by the following formula:

[0156]

[0157] The size of the distance factor depends on the distance from the center of the cluster to which it belongs and the distance from the centers of other clusters. The larger the distance factor, the stronger the representativeness of the band and the weaker the correlation.

[0158] Superpixel clustering variance ratio η∈R B Calculation. The superpixels are further clustered into superpixel clusters using spectral information, and an indicator of the ratio of inter-cluster and intra-cluster variances based on superpixel clustering is designed. First, the mean vector of the spectral features of all pixels in each superpixel is taken as the spectral feature of the superpixel; then, all superpixels are clustered using K-means according to the superpixel spectral features to obtain the clusters of superpixel clusters; then, each superpixel cluster is regarded as a category, and the intra-class variance is defined as the variance of the superpixel spectral features in the same cluster, expressed as S w , the inter-class variance is defined as the variance of the average spectral features between different clusters, expressed as S b The symbol R represents the cluster of superpixel clustering, I = {R1, R2, …, R NC}, Where NC is the number of clusters, R i is the i-th superpixel cluster, which contains r i superpixels, represents the jth superpixel of the ith cluster. Considering each cluster as a category, the intra-class variance S of the Mikania micrantha hyperspectral image w and the between-class variance S b The calculation process is as follows:

[0159]

[0160] in, represents the kth pixel of the jth superpixel, n j Represents the number of pixels in the superpixel. i represents the mean vector of the i-th cluster, and μ represents the mean vector of all pixels; finally, the Fisher score of the l-th band of the Mikania micrantha hyperspectral image can be defined as:

[0161]

[0162] Among them, η l The larger the value of , the larger the superpixel clustering variance ratio of the l-th band, the closer the features of superpixels of the same category are, and the stronger the distinguishability of Mikania micrantha in the l-th band.

[0163] Comprehensive ranking index discrimination score ψ∈R B It can be expressed in the form of a vector, including the distance factor γ∈R B , information entropy h∈R B and the variance ratio η∈RB , which is integrated by the following formula:

[0164]

[0165] Among them, ||·||2 represents the modulus of the vector, and the comprehensive ranking index ψ is obtained by normalizing and accumulating the three indicators. Normalization can reduce the magnitude gap between different indicators. In addition, there is a positive correlation between the comprehensive ranking index of the spectral identification band of Mikania micrantha and the distance factor, information entropy and superpixel variance ratio. The larger the distance factor of the band, the larger the information entropy or the superpixel variance ratio, the larger the comprehensive ranking index. Therefore, the greater the representativeness and decorrelation of the band, the greater the amount of information, and the stronger the distinguishability of Mikania micrantha, the greater the value of its comprehensive ranking index. By selecting the band with a high score of the comprehensive ranking index in each cluster, the appropriate band can be selected.

[0166] In this embodiment, information entropy calculation can use other information metrics, such as spectral information divergence, etc., distance factor calculation can use other distance metrics, such as Mahalanobis distance, etc., and variance ratio calculation can also use other statistical indicators, such as standard deviation ratio, etc.

[0167] In summary, one or more of the calculation methods provided in each subdivided step or implementation mode of the present application can be replaced by a method that is adaptively adjusted according to the application scenario of the present application based on some commonly used data processing or data measurement means.

[0168] This implementation determines a comprehensive discriminative score by calculating the information entropy, distance factor and superpixel clustering variance ratio of each band, thereby achieving the selection of target bands in each band cluster, improving the effectiveness of band selection, and further improving the effectiveness of subsequent object recognition.

[0169] In summary, this application proposes multi-scale superpixel segmentation and heuristic formulas, using superpixels of different sizes to represent the spatial structure type of Mikania micrantha, and dynamically adjusting the size of superpixels according to the spatial resolution of different scenes, solving the problem of determining superpixel segmentation parameters, and improving the noise resistance and accuracy of similarity calculation. The adaptive Gaussian kernel designed in this application takes into account different superpixel sizes, different spatial patterns, and different numbers of clusters, independently describes the band similarity of Mikania micrantha, and makes the calculated hyperspectral band similarity matrix more reasonable. The comprehensive ranking index band selection method constructed in this application takes into account different indicators for evaluating the spectral identification bands of Mikania micrantha, covering the amount of information, differences, and distinctions, and obtains better results than a single indicator.

[0170] Embodiment 2

[0171] The present application also provides a super-pixel hyperspectral image object recognition device to implement the above method, see Figure 5 , Figure 5 Schematic diagram of the structure of a super-pixel hyperspectral image object recognition device disclosed in the embodiment of the present application. Figure 5 As shown, based on any other embodiment, the device includes:

[0172] A processing module 22 is used to perform multi-scale super-pixel segmentation on the target image to obtain a plurality of super-pixel-level first sub-images;

[0173] The first sub-image is used to indicate the spatial structure information of each object in the target image;

[0174] A determination module 21, configured to determine an adaptive Gaussian kernel according to each first sub-image to obtain a band similarity matrix;

[0175] A processing module 22, configured to perform super-pixel-level spectral clustering according to the band similarity matrix to obtain a plurality of band clusters corresponding to each first sub-image;

[0176] A determination module 21, used to determine a comprehensive ranking indicator;

[0177] A processing module 22, for determining a target band in each band cluster according to the comprehensive ranking index;

[0178] A determination module 21, configured to determine a second sub-image in each first sub-image according to a target band;

[0179] The processing module 22 is used to identify the target object in the target image according to each second sub-image.

[0180] By means of superpixel segmentation, the target image is segmented into superpixels at a preset scale to obtain the first superpixel sub-images corresponding to multiple scales. The spatial structure information of each object in the target image is expressed by the first sub-image, and the adaptive Gaussian kernel is further determined. The band similarity matrix is ​​obtained with the band in each first sub-image as a unit, and superpixel spectral clustering is performed to obtain multiple band clusters. Data processing and screening are performed on the bands in each band cluster, and a comprehensive ranking index is determined, and one or more target bands corresponding to each band cluster are further determined. Furthermore, in the first sub-image, a second sub-image corresponding to each target band is determined, and the target object is identified according to the determined second sub-image. Thus, the target image is segmented, data processed and analyzed in the spatial dimension and the band dimension. The clustering method and the index ranking method are combined to select the corresponding target band, and then object recognition is realized, which solves the problem of obstacles in combining the two methods, ensures the effectiveness of object recognition and the physical interpretability of features, and realizes hyperspectral image object recognition with low data volume.

[0181] As an optional implementation manner, the processing module 22 performs multi-scale super-pixel segmentation on the target image to obtain a plurality of super-pixel-level first sub-images in a specific manner including:

[0182] Determine a segmentation scale factor, and determine an average number of pixels contained in each first sub-image according to the segmentation scale factor and the spatial resolution of the target image;

[0183] Determining a first number of first sub-images according to an average number of pixels and a total number of pixels of the target image;

[0184] According to a simple non-iterative clustering algorithm, superpixel segmentation is performed on the target image to obtain a first number of superpixel-level first sub-images;

[0185] Each first sub-image includes a second number of bands.

[0186] By determining the segmentation scale factor and according to the spatial resolution of the target image, the average number of pixels of the first sub-image and the number of first sub-images are determined through a heuristic data determination method, and the target image is divided into super-pixel-level sub-images with a basically uniform data volume, thereby reducing the difficulty of subsequent data processing, improving the effectiveness of segmentation, ensuring the effectiveness of the information contained in each first sub-image, and then ensuring the effectiveness of object recognition, and realizing hyperspectral image object recognition with low data volume.

[0187] As an optional implementation manner, the processing module 22 determines the adaptive Gaussian kernel according to each first sub-image, and obtains the specific manner of the band similarity matrix, including:

[0188] Determine a second number of band images corresponding to each first sub-image according to the second number of bands of each first sub-image;

[0189] According to each band image, a band vector is determined, and a fully connected graph is determined according to each band vector;

[0190] Among them, the nodes of the fully connected graph are band vectors, and the edges of the fully connected graph include the distance parameters between each band vector;

[0191] According to the fully connected graph, the minimum spanning tree is determined, and according to the edge weight of the minimum spanning tree, the band similarity parameter is determined;

[0192] Generate a band similarity matrix based on the band similarity parameters.

[0193] By determining the band vectors corresponding to each band in the first sub-image, a fully connected graph is generated, and then the distance relationship between the band vectors is indicated according to the fully connected graph, and then a minimum spanning tree is generated to achieve the generation of an adaptive Gaussian kernel, and then the current band similarity parameters can be determined to generate a band similarity matrix. Further, according to the band similarity matrix corresponding to the first sub-image at each segmentation scale, an aggregation matrix can be determined, so as to perform super-pixel-level adaptive spectral clustering, provide a basis for subsequent data processing, ensure the effectiveness of object recognition, and realize hyperspectral image object recognition under low data volume.

[0194] As an optional implementation, the processing module 22 performs superpixel-level spectral clustering according to the band similarity matrix to obtain a plurality of band clusters corresponding to each first sub-image in a specific manner including:

[0195] According to the band similarity matrix, a normalized Laplace matrix is ​​determined;

[0196] Performing eigendecomposition according to the normalized Laplace matrix to obtain the third number of eigenvalues ​​and corresponding eigenvectors, and obtaining a band information matrix;

[0197] Wherein, the first dimension of the band information matrix is ​​the second quantity, and the second dimension of the band information matrix is ​​the third quantity;

[0198] The band information matrix is ​​normalized in units of rows to obtain a normalized band information matrix, thereby obtaining a fourth number of band clusters.

[0199] Through the band similarity matrix under the aforementioned first sub-image, in this embodiment, the band similarity matrix can also be an aggregation matrix determined according to the band similarity matrix corresponding to the first sub-image under each segmentation scale, and then by processing the matrix data, a band information matrix is ​​obtained, thereby determining the band cluster, providing a data basis for the combination of subsequent ranking methods, solving the problem of obstacles in combining the two methods, ensuring the effectiveness of object recognition and the physical interpretability of features, and realizing hyperspectral image object recognition with low data volume.

[0200] As an optional implementation, the comprehensive ranking index includes information entropy, distance factor and superpixel clustering variance ratio;

[0201] The specific method of determining the comprehensive ranking indicator by the determination module 21 includes:

[0202] In the calculation band of the target band cluster, the value and probability distribution of each pixel are determined, and the information entropy of the calculation band is obtained according to the value and probability distribution of each pixel through the information entropy calculation formula;

[0203] and determining a first cluster distance from the calculation band to the center of the target band cluster and a second cluster distance from the calculation band to the centers of other band clusters, and subtracting the sum of the second cluster distances from the first cluster distance to obtain a distance factor for the calculation band;

[0204] and, clustering each first sub-image to obtain a plurality of superpixel clusters, and using each superpixel cluster as a classification label, and determining a superpixel cluster variance ratio of a calculation band according to a ratio of an inter-class variance to an intra-class variance;

[0205] The processing module 22 determines the specific method of the target band in each band cluster according to the comprehensive ranking index, including:

[0206] According to the sum of the normalized values ​​of information entropy, distance factor and superpixel clustering variance ratio, the discriminative score of the comprehensive ranking index is determined, and the ranking is performed according to the discriminative score, and a preset number of target bands in each band cluster are selected.

[0207] By calculating the information entropy, distance factor and superpixel clustering variance ratio of each band, the comprehensive discriminative score is determined, thereby realizing the selection of target bands in each band cluster, improving the effectiveness of band selection, and further improving the effectiveness of subsequent object recognition.

[0208] It can be explained that it should be understood that the division of the various modules of the above device is only a division of logical functions. In actual implementation, they can be fully or partially integrated into one physical entity, or they can be physically separated. And these modules can all be implemented in the form of software called by processing elements; they can also be all implemented in the form of hardware; some modules can also be implemented in the form of software called by processing elements, and some modules can be implemented in the form of hardware. For example, the determination module 21 can be a separately established processing element, or it can be integrated in a chip of the above device. In addition, it can also be stored in the memory of the above device in the form of program code, and called and executed by a processing element of the above device. The function of the above determination module 21. The implementation of other modules is similar. In addition, these modules can be fully or partially integrated together, or they can be implemented independently. The processing element here can be an integrated circuit with signal processing capabilities. In the implementation process, each step of the above method or each module above can be completed by an integrated logic circuit of hardware in the processor element or instructions in the form of software.

[0209] Figure 6 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application, such as Figure 6 As shown, the electronic device may include: a transceiver 121 , a processor 122 , and a memory 123 .

[0210] The processor 122 executes the computer execution instructions stored in the memory, so that the processor 122 executes the scheme in the above embodiment. The processor 122 can be a general-purpose processor, including a central processing unit CPU, a network processor (network processor, NP), etc.; it can also be a digital signal processor DSP, an application-specific integrated circuit ASIC, a field programmable gate array FPGA or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components.

[0211] The memory 123 is connected to the processor 122 via a system bus and completes communication between them. The memory 123 is used to store computer program instructions.

[0212] The transceiver 121 may be used to obtain tasks to be executed and configuration information of the tasks to be executed.

[0213] The system bus can be a peripheral component interconnect (PCI) bus or an extended industry standard architecture (EISA) bus, etc. The system bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, only one thick line is used in the figure, but it does not mean that there is only one bus or one type of bus. The transceiver is used to realize the communication between the database access device and other computers (such as clients, read-write libraries, and read-only libraries). The memory may include random access memory (RAM) and may also include non-volatile memory.

[0214] The electronic device provided in the embodiment of the present application may be the terminal device of the above embodiment.

[0215] The embodiment of the present application also provides a chip for running instructions, which is used to execute the technical solution of the method for identifying Mikania micrantha in super-pixel-level hyperspectral images in the above embodiment.

[0216] The embodiment of the present application further provides a computer-readable storage medium, in which computer instructions are stored. When the computer instructions are executed on a computer, the computer executes the technical solution of the method for identifying Mikania micrantha in super-pixel-level hyperspectral images in the above embodiment.

[0217] An embodiment of the present application also provides a computer program product, which includes a computer program stored in a computer-readable storage medium. At least one processor can read the computer program from the computer-readable storage medium. When at least one processor executes the computer program, it can implement the technical solution of the super-pixel-level hyperspectral image Mikania micrantha identification method in the above embodiment.

[0218] In the several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic, for example, the division of modules is only a logical function division, and there may be other division methods in actual implementation, such as multiple modules can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, devices or modules, which can be electrical, mechanical or other forms.

[0219] The modules described as separate components may or may not be physically separated, and the components shown as modules may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the modules may be selected according to actual needs to implement the solution of this embodiment.

[0220] In addition, each functional module in each embodiment of the present application can be integrated into one processing unit, or each module can exist physically separately, or two or more modules can be integrated into one unit. The above-mentioned module-composed unit can be implemented in the form of hardware or in the form of hardware plus software functional units.

[0221] The above-mentioned integrated module implemented in the form of a software function module can be stored in a computer-readable storage medium. The above-mentioned software function module is stored in a storage medium, including a number of instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) or a processor to perform some steps of the methods of various embodiments of the present application.

[0222] It should be understood that the processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), etc. A general-purpose processor may be a microprocessor or any conventional processor. The steps of the method disclosed in the invention may be directly implemented as being executed by a hardware processor, or may be implemented by a combination of hardware and software modules in the processor.

[0223] The memory may include a high-speed RAM memory, and may also include a non-volatile storage NVM, such as at least one disk memory, and may also be a USB flash drive, a mobile hard disk, a read-only memory, a magnetic disk or an optical disk, etc.

[0224] The bus may be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. The bus may be divided into an address bus, a data bus, a control bus, etc. For ease of representation, the bus in the drawings of the present application is not limited to only one bus or one type of bus.

[0225] The above storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk. The storage medium can be any available medium that can be accessed by a general or special purpose computer.

[0226] An exemplary storage medium is coupled to a processor so that the processor can read information from the storage medium and write information to the storage medium. Of course, the storage medium can also be a component of the processor. The processor and the storage medium can be located in an application specific integrated circuit (ASIC). Of course, the processor and the storage medium can also exist as discrete components in an electronic control unit or a main control device.

[0227] Those skilled in the art can understand that all or part of the steps of implementing the above-mentioned method embodiments can be completed by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When the program is executed, the steps of the above-mentioned method embodiments are executed; and the aforementioned storage medium includes: ROM, RAM, disk or optical disk and other media that can store program codes.

[0228] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit it. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present application.

Claims

1. A super-pixel hyperspectral image recognition method for Mikania micrantha, characterized in that: The method comprises: Performing multi-scale super-pixel segmentation on the target image to obtain a plurality of super-pixel-level first sub-images; Wherein, the first sub-image is used to indicate the spatial structure information of each object in the target image; According to each of the first sub-images, an adaptive Gaussian kernel is determined to obtain a band similarity matrix; Performing superpixel-level spectral clustering according to the band similarity matrix to obtain a plurality of band clusters corresponding to each of the first sub-images; Determining a comprehensive ranking index, and determining a target band in each of the band clusters according to the comprehensive ranking index; In each of the first sub-images, a second sub-image is determined according to the target band, and a target object in the target image is identified according to each of the second sub-images.

2. The method according to claim 1, characterized in that The step of performing multi-scale super-pixel segmentation on the target image to obtain a plurality of super-pixel-level first sub-images includes: Determine a segmentation scale factor, and determine an average number of pixels included in each first sub-image according to the segmentation scale factor and the spatial resolution of the target image; Determining a first number of first sub-images according to the average number of pixels and the total number of pixels of the target image; According to a simple non-iterative clustering algorithm, the target image is segmented into superpixels to obtain first sub-images at the first number of superpixel levels; Each of the first sub-images contains a second number of bands.

3. The method according to claim 2, characterized in that The step of determining an adaptive Gaussian kernel according to each of the first sub-images to obtain a band similarity matrix includes: Determining a second number of band images corresponding to each of the first sub-images according to a second number of bands of each of the first sub-images; Determine a band vector according to each of the band images, and determine a fully connected graph according to each of the band vectors; The nodes of the fully connected graph are the band vectors, and the edges of the fully connected graph include distance parameters between the band vectors. Determine a minimum spanning tree according to the fully connected graph, and determine a band similarity parameter according to the edge weights of the minimum spanning tree; A band similarity matrix is ​​generated according to the band similarity parameters.

4. The method according to claim 3, characterized in that: The performing superpixel-level spectral clustering according to the band similarity matrix to obtain a plurality of band clusters corresponding to each of the first sub-images includes: Determining a normalized Laplace matrix according to the band similarity matrix; Performing eigendecomposition according to the normalized Laplace matrix to obtain a third number of eigenvalues ​​and corresponding eigenvectors, and obtaining a band information matrix; The first dimension of the band information matrix is ​​the second number, and the second dimension of the band information matrix is ​​the third number; The band information matrix is ​​standardized in units of rows to obtain a standardized band information matrix, thereby obtaining a fourth number of band clusters.

5. The method according to any one of claims 1 to 4, characterized in that: The comprehensive ranking indicators include information entropy, distance factor and superpixel clustering variance ratio; The comprehensive ranking indicators are as follows: In the calculation band of the target band cluster, the value and probability distribution of each pixel are determined, and the information entropy of the calculation band is obtained according to the value and probability distribution of each pixel through an information entropy calculation formula; and determining a first cluster distance from the calculation band to the center of the target band cluster and a second cluster distance from the calculation band to the centers of other band clusters, and subtracting the sum of the second cluster distances from the first cluster distance to obtain a distance factor for the calculation band; And, clustering each of the first sub-images to obtain a plurality of superpixel clusters, and using each of the superpixel clusters as a classification label, and determining the superpixel cluster variance ratio of the calculation band according to the ratio of the between-class variance to the within-class variance; Determining the target band in each of the band clusters according to the comprehensive ranking index includes: According to the sum of the normalized values ​​of the information entropy, the distance factor and the superpixel clustering variance ratio, the discriminative score of the comprehensive ranking index is determined, and ranking is performed according to the discriminative score, and a preset number of target bands in each of the band clusters are selected.

6. A super-pixel hyperspectral image object recognition device, characterized in that: The device comprises: A processing module, used for performing multi-scale super-pixel segmentation on the target image to obtain a plurality of super-pixel-level first sub-images; Wherein, the first sub-image is used to indicate the spatial structure information of each object in the target image; The determination module is used to determine an adaptive Gaussian kernel according to each of the first sub-images to obtain a band similarity matrix; The processing module is used to perform super-pixel level spectral clustering according to the band similarity matrix to obtain a plurality of band clusters corresponding to each of the first sub-images; The determination module is used to determine the comprehensive ranking index; The processing module is used to determine the target band in each of the band clusters according to the comprehensive ranking index; The determining module is used to determine a second sub-image in each of the first sub-images according to the target band; The processing module is used to identify the target object in the target image according to each of the second sub-images.

7. The device according to claim 6, characterized in that The specific method of the processing module performing multi-scale super-pixel segmentation on the target image to obtain a plurality of super-pixel-level first sub-images includes: Determine a segmentation scale factor, and determine an average number of pixels included in each first sub-image according to the segmentation scale factor and the spatial resolution of the target image; Determining a first number of first sub-images according to the average number of pixels and the total number of pixels of the target image; According to a simple non-iterative clustering algorithm, the target image is segmented into superpixels to obtain first sub-images at the first number of superpixel levels; Each of the first sub-images contains a second number of bands.

8. The device according to claim 7, characterized in that The specific method in which the processing module determines the adaptive Gaussian kernel according to each of the first sub-images to obtain the band similarity matrix includes: Determining a second number of band images corresponding to each of the first sub-images according to a second number of bands of each of the first sub-images; Determine a band vector according to each of the band images, and determine a fully connected graph according to each of the band vectors; The nodes of the fully connected graph are the band vectors, and the edges of the fully connected graph include distance parameters between the band vectors. Determine a minimum spanning tree according to the fully connected graph, and determine a band similarity parameter according to the edge weights of the minimum spanning tree; A band similarity matrix is ​​generated according to the band similarity parameters.

9. An electronic device, characterized in that: include: A processor, and a memory communicatively connected to the processor; The memory stores computer-executable instructions; The processor executes the computer-executable instructions stored in the memory to implement the method according to any one of claims 1 to 5.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-executable instructions, which are used to implement the method according to any one of claims 1 to 5 when executed by a processor.

Images