Ground surface anomaly detection method and device based on space-frequency characteristics and map characteristics

Abstract

The invention relates to the technical field of remote sensing image anomalies, in particular to a ground surface anomaly detection method and device based on space-frequency characteristics and atlas characteristics, and the method comprises the steps: obtaining a single-scene hyperspectral image, determining at least one image scene of the single-scene hyperspectral image, decomposing the single-scene hyperspectral image into space-frequency domain comprehensive characteristic information, constructing a space-frequency domain abnormal feature index by using the space-frequency domain comprehensive feature information; based on the earth surface anomaly knowledge graph, using a channel attention mechanism to carry out deep mining on the plurality of features so as to select high attention features satisfying a preset condition for different scenes of the at least one image scene, and constructing a graph feature comprehensive index used for earth surface anomaly discovery, and obtaining an anomaly detection result by using the space-frequency domain anomaly characteristic index and the map characteristic comprehensive index. Therefore, the problems of earth surface feature discreteness, uncertainty, time-varying property, low operation efficiency and the like existing in earth surface anomaly detection in related technologies are solved.

Description

Technical Field

[0001] The present application relates to the field of remote sensing image anomaly technology, and in particular to a method and device for detecting surface anomalies based on space-frequency features and spectral features. Background Art

[0002] Affected by human activities and climate change, abnormal events on the ground are frequent and the damage is increasing. Satellite remote sensing has become a key means of monitoring surface anomalies due to its own advantages, and there is an urgent need for its early detection and immediate diagnosis. In related technologies, the anomaly detection method mainly extracts background information from images and uses the difference between background and anomalies to achieve distinction. According to different background information extraction methods, it can be mainly divided into anomaly detection algorithms based on statistical models, anomaly detection algorithms based on geometric models, and anomaly detection algorithms based on deep learning.

[0003] However, in related technologies, the statistical model assumes that the background follows a specific distribution. The RX algorithm has good performance but a high false alarm rate. The geometric modeling method uses feature vectors to represent the background, which is not good for anomaly representation. The sparse representation method does not fully combine the spatial spectrum characteristics, and its accuracy and effectiveness need to be optimized. The deep learning method has high accuracy but large computational complexity and slow convergence, which makes it difficult to meet the needs of on-orbit real-time detection. In summary, the current surface anomaly detection has the limitations of discrete, uncertain, and time-varying surface features, as well as low computational efficiency, which urgently need to be improved. Summary of the invention

[0004] The present application provides a method and device for detecting surface anomalies based on space-frequency features and spectral features, so as to solve the problems in related technologies of surface anomaly detection, such as discreteness, uncertainty, time-varying nature of surface features, and low computational efficiency.

[0005] The first aspect of the present application provides a surface anomaly detection method based on space-frequency features and spectral features, comprising the following steps: obtaining a surface anomaly knowledge graph; obtaining a single-scene hyperspectral image, and determining at least one image scene of the single-scene hyperspectral image, and decomposing the single-scene hyperspectral image into space- and frequency-domain comprehensive feature information, so as to construct a space- and frequency-domain anomaly feature index using the space- and frequency-domain comprehensive feature information; based on the surface anomaly knowledge graph, using a channel attention mechanism to perform deep mining on multiple features, so as to select high-attention features that meet preset conditions for different scenes of the at least one image scene, and construct a spectral feature comprehensive index for surface anomaly discovery, so as to obtain an anomaly detection result using the space- and frequency-domain anomaly feature index and the spectral feature comprehensive index.

[0006] Through the above technical solutions, the embodiments of the present application can effectively improve the extraction efficiency and accuracy of abnormal features by acquiring the surface anomaly knowledge graph and single-scene hyperspectral images, combined with the spatial and frequency domain comprehensive feature information. Secondly, the channel attention mechanism is used to deeply mine the features, and high-attention features can be selected for different image scenes, thereby reducing feature redundancy and improving the accuracy and reliability of anomaly detection. Finally, by comprehensively using the spatial and frequency domain anomaly feature index and the spectral feature comprehensive index, the comprehensiveness and accuracy of anomaly detection are enhanced, providing strong technical support for the timely discovery and response of surface abnormal events.

[0007] Optionally, in one embodiment of the present application, the acquiring of a single-view hyperspectral image, determining at least one image scene of the single-view hyperspectral image, and decomposing the single-view hyperspectral image into spatial and frequency domain comprehensive feature information, so as to construct a spatial and frequency domain abnormal feature index using the spatial and frequency domain comprehensive feature information, includes: reading the single-view hyperspectral image; converting the single-view hyperspectral image into an image in double-precision format; iterating each pixel of the image to calculate the outer product and average value of each band value of each pixel, and calculating a covariance matrix based on the outer product and average value of each band value of each pixel; calculating at least one eigenvalue and eigenvector of the covariance matrix to sort at least one eigenvalue, determine multiple main components, and calculate the contribution rate of each eigenvalue, and find the principal component whose cumulative contribution rate is greater than a preset threshold; determining the eigenvector corresponding to the principal component, so as to read the band values ​​of each pixel position and perform principal component transformation to obtain a matrix after principal component transformation.

[0008] Through the above technical solution, the embodiment of the present application can effectively extract the covariance matrix of the image and its eigenvalues ​​and eigenvectors by performing double-precision format conversion and pixel-by-pixel iterative calculation on the single-view hyperspectral image, thereby realizing principal component analysis. This process not only improves the processing efficiency of image data, but also accurately identifies and screens out the main components, ensuring that the extracted information has a high contribution rate, thereby optimizing subsequent anomaly detection and analysis. This method reduces computational complexity while improving data processing accuracy.

[0009] Optionally, in one embodiment of the present application, the single-scene hyperspectral image is acquired, at least one image scene of the single-scene hyperspectral image is determined, and the single-scene hyperspectral image is decomposed into spatial and frequency domain comprehensive feature information to construct a spatial and frequency domain abnormal feature index using the spatial and frequency domain comprehensive feature information, and also includes: normalizing the matrix after the principal component transformation, calling the discrete fractional Fourier transform function for each pixel point of the matrix after the principal component transformation, calculating the entropy of each band, and taking the maximum value to assign to FrFE to obtain the corresponding order; looping pixel points, calling the function to perform discrete Fourier transform, calculating the eigenvector matrix of DFRFT, using the eigenvector matrix and the input vector to calculate DFRFT to obtain result data, and standardizing the result data.

[0010] Through the above technical solution, the embodiment of the present application can normalize the matrix after the principal component transformation and use the discrete fractional Fourier transform to calculate the entropy value of each band, thereby extracting the features with the maximum amount of information and providing an effective feature basis for subsequent anomaly detection. The algorithm can make full use of space-frequency information and improve the ability to distinguish abnormal signals by looping each pixel point and calculating the eigenvector matrix of DFRFT. At the same time, the result data is standardized to further enhance the stability and accuracy of the algorithm. This process not only improves the accuracy of anomaly detection, but also effectively reduces noise interference, optimizes computing efficiency, and adapts to the needs of real-time monitoring.

[0011] Optionally, in one embodiment of the present application, it also includes: generating feedback data through the spatial-frequency domain abnormal feature index and the spectral feature comprehensive index; and using the feedback data to perform image cropping to mark abnormal areas.

[0012] Through the above technical solution, the embodiment of the present application can generate feedback data by combining the spatial-frequency domain abnormal feature index and the spectral feature comprehensive index to achieve accurate identification and labeling of abnormal areas in remote sensing images. Using feedback data for image cropping not only improves the visualization effect of abnormal areas, but also enhances the accuracy and real-time performance of abnormal detection, thereby providing effective technical support for responding to natural disasters, environmental pollution and other emergencies.

[0013] The second aspect of the present application provides a surface anomaly detection device based on space-frequency features and spectral features, including: an acquisition module for acquiring a surface anomaly knowledge graph; a construction module for acquiring a single-scene hyperspectral image, determining at least one image scene of the single-scene hyperspectral image, and decomposing the single-scene hyperspectral image into space- and frequency-domain comprehensive feature information, so as to construct a space- and frequency-domain anomaly feature index using the space- and frequency-domain comprehensive feature information; a detection module for deeply mining multiple features based on the surface anomaly knowledge graph using a channel attention mechanism, so as to select high-attention features that meet preset conditions for different scenes of the at least one image scene, and construct a spectral feature comprehensive index for surface anomaly discovery, so as to obtain an anomaly detection result using the space- and frequency-domain anomaly feature index and the spectral feature comprehensive index.

[0014] Through the above technical solutions, the embodiments of the present application can effectively improve the extraction efficiency and accuracy of abnormal features by acquiring the surface anomaly knowledge graph and single-scene hyperspectral images, combined with the spatial and frequency domain comprehensive feature information. Secondly, the channel attention mechanism is used to deeply mine the features, and high-attention features can be selected for different image scenes, thereby reducing feature redundancy and improving the accuracy and reliability of anomaly detection. Finally, by comprehensively using the spatial and frequency domain anomaly feature index and the spectral feature comprehensive index, the comprehensiveness and accuracy of anomaly detection are enhanced, providing strong technical support for the timely discovery and response of surface abnormal events.

[0015] Optionally, in one embodiment of the present application, the construction module includes: a reading unit for reading the single-scene hyperspectral image; a conversion unit for converting the single-scene hyperspectral image into an image in double-precision format; a first calculation unit for iterating each pixel of the image to calculate the outer product and average of each band value of each pixel, and calculate the covariance matrix based on the outer product and average of each band value of each pixel; a second calculation unit for calculating at least one eigenvalue and eigenvector of the covariance matrix to sort at least one eigenvalue, determine multiple main components, and calculate the contribution rate of each eigenvalue to find the principal component whose cumulative contribution rate is greater than a preset threshold; a principal component transformation unit for determining the eigenvector corresponding to the principal component to read the band values ​​of each pixel position and perform principal component transformation to obtain a matrix after principal component transformation.

[0016] Through the above technical solution, the embodiment of the present application can effectively extract the covariance matrix of the image and its eigenvalues ​​and eigenvectors by performing double-precision format conversion and pixel-by-pixel iterative calculation on the single-view hyperspectral image, thereby realizing principal component analysis. This process not only improves the processing efficiency of image data, but also accurately identifies and screens out the main components, ensuring that the extracted information has a high contribution rate, thereby optimizing subsequent anomaly detection and analysis. This method reduces computational complexity while improving data processing accuracy.

[0017] Optionally, in one embodiment of the present application, the building module includes: a normalization unit, used to normalize the matrix after the principal component transformation, so as to call the discrete fractional Fourier transform function for each pixel point of the matrix after the principal component transformation, calculate the entropy of each band, and take the maximum value and assign it to FrFE to obtain the corresponding order; a standardization unit, used to cyclically process the pixel points, call the function to perform discrete Fourier transform, calculate the eigenvector matrix of DFRFT, use the eigenvector matrix and the input vector to calculate DFRFT to obtain result data, and standardize the result data.

[0018] Through the above technical solution, the embodiment of the present application can normalize the matrix after the principal component transformation and use the discrete fractional Fourier transform to calculate the entropy value of each band, thereby extracting the features with the maximum amount of information and providing an effective feature basis for subsequent anomaly detection. The algorithm can make full use of space-frequency information and improve the ability to distinguish abnormal signals by looping each pixel point and calculating the eigenvector matrix of DFRFT. At the same time, the result data is standardized to further enhance the stability and accuracy of the algorithm. This process not only improves the accuracy of anomaly detection, but also effectively reduces noise interference, optimizes computing efficiency, and adapts to the needs of real-time monitoring.

[0019] Optionally, in one embodiment of the present application, it also includes: a feedback module for generating feedback data through the spatial-frequency domain abnormal feature index and the spectral feature comprehensive index; a cropping module for using the feedback data to perform image cropping to mark abnormal areas.

[0020] Through the above technical solution, the embodiment of the present application can generate feedback data by combining the spatial-frequency domain abnormal feature index and the spectral feature comprehensive index to achieve accurate identification and labeling of abnormal areas in remote sensing images. Using feedback data for image cropping not only improves the visualization effect of abnormal areas, but also enhances the accuracy and real-time performance of abnormal detection, thereby providing effective technical support for responding to natural disasters, environmental pollution and other emergencies.

[0021] The third aspect of the present application provides an electronic device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the surface anomaly detection method based on space-frequency features and spectral features as described in the above embodiments.

[0022] The fourth aspect of the present application provides a computer-readable storage medium, which stores a computer program. When the program is executed by a processor, it implements the above-mentioned surface anomaly detection method based on space-frequency characteristics and spectral characteristics.

[0023] The fifth aspect of the present application provides a computer program, including a computer program, which, when executed, is used to implement the above-mentioned surface anomaly detection method based on space-frequency features and spectral features.

[0024] The embodiment of the present application can combine the surface anomaly knowledge graph with the single-scene hyperspectral image, and use the spatial and frequency domain comprehensive feature information to significantly improve the extraction efficiency and accuracy of abnormal features. At the same time, the application of the channel attention mechanism enables the selection of high-attention features in different image scenes, thereby reducing feature redundancy and further improving the accuracy and reliability of anomaly detection. In addition, the implementation of principal component analysis effectively identifies and screens out the main components through double-precision format conversion and pixel-by-pixel iterative calculation, optimizing the subsequent anomaly detection and analysis process. The introduction of discrete fractional Fourier transform provides a rich feature basis for anomaly detection, improves the ability to resolve abnormal signals, reduces noise interference, and adapts to the needs of real-time monitoring. Finally, by comprehensively using the spatial and frequency domain abnormal feature index and the spectral feature comprehensive index to generate feedback data, accurate identification and annotation of abnormal areas in remote sensing images are achieved, providing strong technical support for responding to emergencies such as natural disasters and environmental pollution.

[0025] Additional aspects and advantages of the present application will be given in part in the description below, and in part will become apparent from the description below, or will be learned through the practice of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] The above and / or additional aspects and advantages of the present application will become apparent and easily understood from the following description of the embodiments in conjunction with the accompanying drawings, in which:

[0027] Figure 1 The schematic diagram of the overall network framework for surface anomaly detection based on space-frequency features and spectral features;

[0028] Figure 2 A flow chart of a method for detecting surface anomalies based on space-frequency features and spectral features provided according to an embodiment of the present application;

[0029] Figure 3 It is a flowchart of a single-view high (multi) spectral surface anomaly discovery method based on space-frequency anomaly index combined with feature knowledge according to one embodiment of the present application;

[0030] Figure 4 It is a flowchart of a method for discovering surface anomalies in single-view hyper (multi) spectral images based on a comprehensive index of knowledge graph features according to one embodiment of the present application;

[0031] Figure 5 A schematic diagram of the abnormal detection results of the California forest fire surface according to a specific embodiment of the present application;

[0032] Figure 6 This is a schematic diagram of the surface anomaly detection results of the oil spill in the Gulf of Mexico according to a specific embodiment of the present application;

[0033] Figure 7 This is a schematic diagram of the abnormal type discovery result of the fire surface according to a specific embodiment of the present application;

[0034] Figure 8 A schematic diagram of the results of discovering abnormal types of marine oil spill surface according to a specific embodiment of the present application;

[0035] Fig. 9 A schematic diagram of extracting flood surface anomaly type features according to a specific embodiment of the present application;

[0036] Fig.10 A schematic diagram of extracting features of abnormal landslide surface types according to a specific embodiment of the present application;

[0037] Fig.11 A schematic diagram of abnormal discovery results of landslide and flood areas according to a specific embodiment of the present application;

[0038] Fig.12 A schematic diagram of the structure of a surface anomaly detection device based on space-frequency characteristics and spectrum characteristics provided in an embodiment of the present application;

[0039] Fig.13 The figure is a structural example diagram of an electronic device provided according to an embodiment of the present application. DETAILED DESCRIPTION

[0040] Embodiments of the present application are described in detail below, and examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present application, and should not be construed as limiting the present application.

[0041] The following describes the surface anomaly detection method and device based on space-frequency features and spectral features of the embodiment of the present application with reference to the accompanying drawings. In view of the above-mentioned related technologies mentioned in the background technology, the surface anomaly detection has the limitations of discreteness, uncertainty, time-varying, and low computational efficiency of surface features. The present application provides a surface anomaly detection method based on space-frequency features and spectral features. In this method, the extraction efficiency and accuracy of abnormal features can be effectively improved by obtaining the surface anomaly knowledge graph and the single-view hyperspectral image, combining the spatial and frequency domain comprehensive feature information. Secondly, the channel attention mechanism is used to deeply mine the features, and high-attention features can be selected for different image scenes, thereby reducing feature redundancy and improving the accuracy and reliability of abnormal discovery. Finally, by comprehensively using the space-frequency domain abnormal feature index and the spectral feature comprehensive index, the comprehensiveness and accuracy of anomaly detection are enhanced, providing strong technical support for the timely discovery and response of surface abnormal events. As a result, the problems of discreteness, uncertainty, time-varying, and low computational efficiency of surface anomaly detection in the related technologies are solved.

[0042] In order to more clearly show the overall structure and workflow of the present invention, Figure 1 A network framework for surface anomaly detection and diagnosis based on the spatial-frequency characteristic anomaly index and the spectral characteristic comprehensive index is presented, which provides a basis for the detailed description of the subsequent steps.

[0043] Specifically, Figure 2 A flow chart of a method for detecting anomalies on the ground based on space-frequency features and spectral features provided in an embodiment of the present application.

[0044] like Figure 2 As shown, the surface anomaly detection method based on space-frequency characteristics and spectrum characteristics includes the following steps:

[0045] In step S201, a surface anomaly knowledge graph is obtained.

[0046] It is understandable that surface anomaly refers to the sudden change of typical surface types or states such as vegetation, water bodies, bare land, and artificial surfaces under the influence of natural or human factors, which threatens the natural environment and human social security to varying degrees, and usually causes the fluctuation values ​​of remote sensing response characteristics such as spectrum, space, scattering, and radiation on remote sensing images to exceed the threshold of normal surface changes. The surface anomaly knowledge graph input in the embodiment of the present application includes but is not limited to anomaly discovery methods, surface anomaly types, typical anomaly features, etc.

[0047] In some embodiments, the surface anomaly knowledge graph is the application of the knowledge graph in the field of surface anomaly discovery and diagnosis. It is a structured knowledge base that formally describes the concepts, entities, attributes and their relationships applied in the field of surface anomalies, so that concepts and entities are interconnected to form a network knowledge structure. The input graph contains knowledge such as surface anomaly discovery methods, surface anomaly types, and typical anomaly characteristics. In remote sensing images, surface anomalies can be divided into invasive and trait types, and the occurrence scenes are divided into four categories: vegetation, water bodies, artificial surfaces and bare land. Anomalies can be discovered and diagnosed through response characteristics such as spectral characteristics and texture characteristics. Relevant characteristic indicators include VASTI (Vegetation Adjusted Soil Temperature Index), STD (Standard Deviation), 3DC (Three-Dimensional Color Index), NIR (Near Infrared), SWIR (Short-Wave Infrared), BVDI (Bare Soil Vegetation Difference Index), HRI (Hotness Ratio Index), SVI (Spectral Vegetation Index), etc.

[0048] The embodiments of the present application can integrate relevant methods, types and features of surface anomaly discovery through a structured knowledge base, and can systematically improve the accuracy and efficiency of surface anomaly detection.

[0049] In step S202, a single-view hyperspectral image is obtained, and at least one image scene of the single-view hyperspectral image is determined, and the single-view hyperspectral image is decomposed into spatial and frequency domain comprehensive feature information, so as to construct a spatial and frequency domain abnormal feature index using the spatial and frequency domain comprehensive feature information.

[0050] It can be understood that single-scene hyperspectral imagery refers to image data of a single scene acquired through hyperspectral remote sensing technology. Hyperspectral images have dozens to hundreds of continuous spectral bands, which can provide rich surface information and capture the spectral characteristics of different objects for the identification and analysis of surface features.

[0051] In the actual implementation process, the process of the single-view high (multi) spectral surface anomaly detection method based on the space-frequency anomaly index combined with feature knowledge is as follows: Figure 3As shown in the figure, the obtained single-scene hyperspectral image data is input, and the input image is understood by the convolutional neural network. The convolutional neural network uses the principles of linear algebra (especially matrix multiplication) to identify patterns in the image, thereby performing image classification and object recognition tasks. In the application of surface anomaly detection, the convolutional network performs scene understanding and classification on the input hyperspectral image data to determine the scene to which the input image belongs. According to the surface anomaly event classification system, the input image is understood and classified into water scenes, vegetation scenes, artificial surfaces or bare land scenes, laying the scene foundation for further judging the type of surface anomaly in the image.

[0052] Furthermore, the single-view hyperspectral image is decomposed into spatial and frequency domain comprehensive feature information, and the efficiency is improved through the feature-optimized global anomaly detection algorithm, and the spatial and frequency domain anomaly feature index is constructed, and the anomaly is discovered by combining the abnormal response feature knowledge. The improved algorithm can shorten the operation time by more than 80% while meeting the high anomaly detection accuracy.

[0053] Among them, the PCA (Principal Component Analysis) principal component transformation algorithm is used to reduce the dimension of image data and improve the operation efficiency. Principal component transformation refers to the rotation of the coordinate axis to maximize the variance of the original multi-band data, obtain unrelated bands, and achieve the purpose of eliminating noise and reducing data dimension. Generally, each band has the highest correlation with PC1 (First Principal Component), and the correlation with the following principal components gradually decreases. In practical applications, only the first few principal components can be used for processing. For example, after the PCA transformation of TM (Thematic Map), its PC1, PC2 (Second Principal Component), and PC3 (Third Principal Component) contain more than 95% of the information, while the first 20 components of the principal component analysis of hyperspectral images contain all the information. Therefore, selecting some components (PC1) as remote sensing indexes reduces the correlation between the original data, which is of great significance for the rapid extraction of disaster bodies. At the same time, it can also reduce the time for discovering surface anomalies, which is conducive to the immediate discovery of surface anomalies.

[0054] The specific implementation process includes:

[0055] 1) First read the image data and convert it into double precision format, iterate each pixel of the image, and then calculate the outer product and average of each band value of each pixel to obtain the covariance matrix C, which is defined as C = rv ′ *v.

[0056] 2) Use the eig function to calculate the eigenvalues ​​and eigenvectors of the covariance matrix C.

[0057] 3) Extract the eigenvector newlamda, sort the eigenvalues ​​from small to large to get newy, calculate the proportion of each eigenvalue to the total eigenvalues, and get the contribution rate.

[0058] 4) Calculate the cumulative contribution rate. When the cumulative contribution rate is greater than 0.99, record the current serial number and exit the loop. According to the serial number of the principal component, find the corresponding eigenvector from the eigenvector matrix T and record it in NT.

[0059] 5) For each pixel of the image, read its band values, and then perform dot product with the eigenvector of the principal component to obtain the result after principal component transformation.

[0060] Furthermore, FrFE (Fractional Fourier Entropy) is calculated to enrich the amount of information. That is, the optimal fractional transform coefficient ρ is solved by spectral feature extraction based on FRFT (Fractional Fourier Transform) and Shannon entropy maximization method, and the suspected abnormal information in the image is extracted. Fractional Fourier transform can use the representation of the signal in the intermediate domain, including reflection spectrum information and Fourier domain information, to more clearly distinguish between background objects and abnormal objects.

[0061] Specifically, the fractional Fourier transform is a representation method in the fractional Fourier domain formed by rotating the coordinate axis of the signal counterclockwise around the origin in the time-frequency plane by any angle. It is a generalized Fourier transform. The hyperspectral anomaly detection method based on FrFE uses FrFT as a preprocessing method, and obtains the intermediate domain features between the original reflection spectrum and the Fourier transform with complementary intensity through the space-frequency representation method, which is conducive to removing noise to improve the resolution of abnormal points and backgrounds, and enrich the image information. The specific implementation process includes:

[0062] 1) Data reshaping: Normalize the matrix DataTest after principal component transformation, calculate the number of rows, columns and bands of DataTest, and change its shape from three-dimensional (rows, columns, bands) to two-dimensional (number of pixels, bands).

[0063] 2) Calculate FrFE and optimal order: call FrFEorder function to process DataTest.

[0064] Among them, the calculation process of FrFEorder function is as follows:

[0065] First, get the size of DataTest and initialize a zero matrix E. Loop p from 0 to 1 with a step size of 0.1. In each loop, initialize an all-zero three-dimensional matrix im1 with the same size as DataTest. For each pixel in DataTest, call the Disfrft function to perform discrete fractional Fourier transform processing and store the result in im1. Perform center normalization on the data in im1. Calculate the entropy of each band of im1 and store the result in the corresponding column of E. Increase index by 1. After jumping out of the loop, find the maximum value in each row of E, and then find the maximum value among these maximum values ​​and assign it to FrFE. At the same time, find the row number where this maximum value is located, divide it by 10 and assign it to the order to obtain the optimal order of fractional Fourier entropy.

[0066] The calculation process of the Disfrft function is as follows:

[0067] Calculate the length N of the input vector and determine whether N is an even number. Calculate an offset vector shft for subsequent data rearrangement. Ensure that the input vector is a column vector. If the parameter p is not provided, it is set to N / 2. Then, p is restricted to be between 2 and N-1. Call the dFRFT function to calculate the eigenvector matrix E of the DFRFT. Use E and the input vector to calculate the result of the DFRFT.

[0068] The dFRFT function is used to calculate the eigenvector matrix of DFRFT. The calculation process of the dFRFT function is as follows:

[0069] Define global variables E_saved and p_saved, which are used to store previously calculated eigenvector matrices and the corresponding approximate orders so that they can be reused to reduce the amount of calculation. Check whether the length of the global variable E_saved is equal to the current N, and whether p_saved is equal to the current p. This is to determine whether the previously stored eigenvector matrix can be reused. If E_saved and p_saved do not match the current N and p, or have not been calculated before, then call the make_E function to calculate the new eigenvector matrix E. After the calculation is completed, update E_saved and p_saved to store this newly calculated matrix and approximate order. If E_saved and p_saved match the current N and p, then directly use the stored eigenvector matrix E_saved as the result. The purpose of the function is to return an NxN matrix E that contains the eigenvectors of the discrete Fourier transform matrix.

[0070] The make_E function is used to calculate the eigenvectors and eigenvalues ​​of a specific order of DFRFT. The calculation process of the make_E function is as follows:

[0071] Initialize a series of variables, including the difference matrix d2, the polynomial d_p, and the zero vector st. The polynomial is constructed by convolution operation and accumulated into s to construct the first column of the matrix H. The circulant matrix consisting of the values ​​of s is added to the diagonal matrix consisting of the real part of the Fourier transform of s to construct the matrix H. The transformation matrix V is constructed based on the transformation of the identity matrix for subsequent eigenvector calculation. VHV is calculated by the transformation matrix V and the matrix H. Then, the eigenvectors of the first and second halves of VHV are calculated respectively. In order to ensure that the order of the eigenvectors matches the requirements of the DFT, the eigenvectors are rearranged. The purpose of the make_E function is to construct the eigenvector matrix E of a DFRFT of a specific order p. This matrix is ​​obtained by constructing and diagonalizing a special matrix H, which is a combination of a circulant matrix and a diagonal matrix. Then, the eigenvectors are calculated and rearranged by the transformation matrix V.

[0072] Furthermore, global RX (Reed-Xiaoli, an algorithm) anomaly detection is performed. This method performs global RX detection on the input data cube, returns the detection result, and obtains the anomaly value. The specific implementation process includes: reshaping the input data into a two-dimensional matrix X of size (number of rows and columns) * number of bands, calculating the mean value X_mean of each band of X and the covariance matrix cov_X of X, calculating the transpose of X minus the repeated matrix of X_mean to obtain Y, and initializing the global RX detection result GRX_Detect to a zero matrix of size 1*(number of rows and columns). Calculate the pseudo-inverse cov_inv of cov_X. For each column of Y, calculate its product with cov_inv, and then store the result in GRX_Detect. Finally, GRX_Detect is reshaped to the size of the number of rows * the number of columns, and then returned to obtain the detection result of the global RX detection.

[0073] Specifically, the global anomaly detection algorithm is used to detect outliers in the data, identify abnormal pixels in the image, and find abnormal features in the image. The specific steps include:

[0074] 1) Define the function GRX (Global RX, global RX anomaly detection algorithm), and the input parameter is the hyperspectral (multi-spectral) image.

[0075] 2) Use the size function to get the size of the input data, which are the number of rows, columns, and bands.

[0076] 3) Use the reshape function to convert the three-dimensional data cube into a two-dimensional matrix X, where each row represents a pixel and each column represents a band.

[0077] 4) Calculate the mean X_mean of X to get the average value of each band. The result is a column vector.

[0078] 5) Calculate the covariance matrix cov_X of X.

[0079] 6) Compute the matrix Y, which is the column of the data matrix X minus the mean X_mean.

[0080] 7) Initialize the result matrix GRX_Detect to be a zero matrix with size 1 row and number of columns times the number of rows.

[0081] 8) Calculate the pseudo-inverse cov_inv of the covariance matrix.

[0082] 9) Using a for loop, for each pixel, calculate its outlier in the transformed space, i.e., the product of each column of Y with the pseudo-inverse of the covariance matrix, and then multiplied by the transpose of Y.

[0083] 10) Finally, the one-dimensional detection result GRX_Detect is reshaped into a two-dimensional matrix with the same number of rows and columns as the input data to obtain the anomaly detection result.

[0084] The embodiment of the present application can effectively identify and extract key information from the image by acquiring a single-view hyperspectral image and performing scene classification. After decomposing the image into spatial and frequency domain comprehensive feature information, these feature information can be fully utilized to construct a spatial and frequency domain abnormal feature index, thereby improving the accuracy and efficiency of anomaly detection. This method not only enhances the ability to identify surface anomalies, but also significantly shortens the calculation time, meets the needs of real-time monitoring, and helps to promptly discover and respond to surface abnormal events.

[0085] In step S203, based on the surface anomaly knowledge graph, a channel attention mechanism is used to deeply mine multiple features, so as to select high-attention features that meet preset conditions for different scenes of at least one image scene, and construct a comprehensive index of graph features for surface anomaly discovery, so as to obtain anomaly detection results using the space-frequency domain anomaly feature index and the graph feature comprehensive index.

[0086] It is understandable that the surface anomalies of single-view high (multi) spectral images are discovered by combining the comprehensive index of knowledge graph features. The channel attention mechanism is used to deeply mine massive features, select high-attention features for different scenes, reduce feature redundancy, and construct a comprehensive feature index for anomaly detection. Under the condition of integrating the knowledge system of anomaly graph features, the accuracy of anomaly detection is improved and the false alarm rate is reduced.

[0087] In the actual implementation process, the process of the surface anomaly detection method of single-view hyper (multi) spectral image based on the comprehensive index of knowledge graph features is as follows: Figure 4As shown in the figure, initial features are obtained from the image after scene understanding, including texture features, spectral features, and spectral features. Specific indexes include 3DC, SVI, NIR, HRI, VASTI, STD, BVDI, and entropy. These initial features provide an important information basis for subsequent anomaly detection.

[0088] Furthermore, SENet (Squeeze-and-Excitation Networks) is used to deeply mine these initial features. SENet obtains the importance of each feature channel through automatic learning and assigns different weight values ​​to each feature. In this way, the neural network can focus on feature channels that are useful for surface anomaly detection, while suppressing those feature channels that contribute less to anomaly detection. In this way, the importance of different feature channels is dynamically adjusted, so that the attention ranking of features is different in different scenarios, thereby constructing a comprehensive feature index for specific scenarios.

[0089] Specifically, there are many features in the image, many of which are strongly correlated with surface anomaly detection. By screening the features related to anomaly detection, a knowledge system of anomaly map features is constructed, which includes texture features, spectral features and other important indexes related to anomaly detection. For example, in water scenes, the BVDI index is an important feature, while in artificial surface scenes, the VASTI index is more important. Using the channel attention mechanism, features are weighted for different scenes to reduce feature redundancy and thus improve the accuracy of anomaly detection. The core principle of the attention mechanism is to weight the input features to ensure that the features that the network pays attention to are given greater weights, while irrelevant features are given smaller weights. This method not only reduces feature redundancy, but also effectively improves the accuracy of anomaly detection and reduces the false alarm rate.

[0090] In the embodiment of the present application, a channel attention mechanism is used to optimize features in different scenarios. The specific model used is SENet, which is divided into two parts: compression and excitation. The global spatial information is compressed, and then feature learning is performed in the channel dimension to form the importance of each channel to the channel. Finally, different weights are assigned to each channel through the excitation part. For any given transformation: Ftr:X→U, where X∈R H′×W′×C′ , U∈R H×W×C , Ftr is used as a convolution operator. The specific steps include:

[0091] 1) Squeeze (Fsq): Through global average pooling, the two-dimensional features (H×W) of each channel are compressed into a real number, and the feature map is transformed from

[0092] The feature U is compressed by a squeeze operation to aggregate the feature maps across the spatial dimension H×W to generate a channel descriptor as shown below:

[0093] H×W×C→1×1×C,

[0094] The global spatial information is compressed into the above channel descriptors so that these channel descriptors can be used by the layers they are input to. Global average pooling is used here, and the formula is as follows:

[0095]

[0096] 2) Excitation (Fex): Generate a weight value for each feature channel, build the correlation between channels through two fully connected layers, and the number of output weight values ​​is the same as the number of channels in the input feature map.

[0097] Each channel learns the activation of a specific sample through a self-selected gate mechanism based on channel dependence, so that it can learn to use global information, selectively emphasize information features, and suppress less useful features. Sigmoid is used here, and the ReLU function is embedded in the middle to limit the complexity of the model and help training; the gating mechanism is parameterized through the bottleneck formed by two fully connected layers (FC), that is, W1 is used to reduce the dimension and W2 is used to increase the dimension. The formula is as follows:

[0098] s=F ex (z,W)=σ(g(z,W))=σ(W2δ(W1z)).

[0099] 3) Scale (Fscale): Add the normalized weight obtained above to the features of each channel. The method used in this invention is multiplication, where each channel is multiplied by the weight coefficient. Feature map Each feature is multiplied by its weight and then added to get the feature comprehensive index, and surface anomalies in the image are discovered based on the feature comprehensive index.

[0100] After finding anomalies in the hyperspectral (multi-) spectral image through the spatial-frequency feature anomaly index and the spectral feature comprehensive index, data feedback about the abnormal scene is obtained. The image is cropped using the feedback abnormal data, and the abnormal area is cropped and marked.

[0101] The present application is described in detail below with reference to several specific embodiments.

[0102] The experiment of surface anomaly detection based on fractional Fourier transform used four comparative methods: global and local anomaly detection of fractional Fourier transform under principal component analysis; global and local anomaly detection of fractional Fourier transform of original image. The experiment selected California forest fire and Gulf of Mexico oil spill incident, and the surface anomaly detection results are as follows: Figure 5 and Figure 6 shown.

[0103] Under two typical surface anomaly images, global RX detection has better anomaly detection effect, while local RX detection may have false detection in marine oil spill detection. The global anomaly detection based on fractional Fourier transform under principal component analysis also has a relatively impressive performance in terms of results. Considering the timeliness and rapidity requirements of on-orbit instant detection of surface anomalies, principal component analysis is performed on multispectral data to retain the main information of the data while improving the computational efficiency. The anomaly detection algorithm time calculation is performed on four groups of surface anomaly data, and it is found that the surface anomaly detection time under PCA principal component analysis is effectively reduced. Therefore, from the perspective of timeliness and the accuracy of detection results, PCA transformation post-processing is more efficient and can also effectively detect surface anomaly information.

[0104] A typical surface anomaly discovery experiment was conducted on the surface anomaly discovery model. For two different types of surface anomaly events, fire and marine oil spill, surface anomalies were discovered and detected, and thresholds were set to finally output the surface anomaly discovery results, such as Figure 7 , Figure 8 shown.

[0105] In order to find the types of surface anomalies in fires, we conducted surface anomaly detection in remote sensing images of 6 different spatial regions. The results can be used to find and feedback surface anomalies. After setting the threshold, the surface anomaly area can be effectively output, completing the function of surface anomaly detection. At the same time, due to the superiority of its algorithm, it can effectively avoid interference from clouds, smoke, etc. in the surface anomaly detection stage, and output the fire point location more accurately.

[0106] For the detection of surface anomalies of marine oil spills, we conducted surface anomalies detection of marine oil spills in four different spatial regions, covering open source datasets of marine oil spills and hyperspectral images related to the Gulf of Mexico oil spill. The results are as follows: Figure 8 The above images are used to conduct an experiment on the discovery of surface anomalies of marine oil spills. The experimental results show that this method can also be used to discover and detect surface anomalies of marine oil spills. At the same time, by setting the threshold, it can effectively output the anomaly results and anomaly areas.

[0107] This application uses the comprehensive index of graph features to detect surface anomalies for two different types of surface abnormal disasters, landslide and flood disasters. The high attention feature results are as follows Fig. 9 and Fig.10 As shown. Through the constructed comprehensive index I, surface anomalies of two types of surface anomaly images, landslide and flood, were discovered and detected. Among them, 6 abnormal images of flood events in different spatial regions were detected, including India, Pakistan, central and western Europe, etc., and the threshold segmentation was performed by the maximum variance method between classes, which can better discover the surface anomaly area and output the surface anomaly events; 8 abnormal landslide images in different spatial regions were detected, and the threshold segmentation was performed by the maximum variance method between classes, which can better discover the surface anomaly area and output the surface anomaly events. The results are shown in Fig.11 .

[0108] The embodiment of the present application can use the channel attention mechanism to deeply mine multiple features based on the surface anomaly knowledge graph, and can effectively select high-attention features for different image scenes, thereby constructing a comprehensive index of graph features for surface anomaly detection. This method not only improves the accuracy of anomaly detection and reduces the false alarm rate, but also makes full use of the combination of the spatial-frequency domain anomaly feature index and the graph feature comprehensive index, enhances the reliability and real-time performance of the anomaly detection results, and meets the actual needs of surface anomaly monitoring.

[0109] According to the surface anomaly detection method based on space-frequency features and spectral features proposed in the embodiment of the present application, the extraction efficiency and accuracy of abnormal features can be effectively improved by obtaining the surface anomaly knowledge graph and the single-scene hyperspectral image, and combining the comprehensive feature information in the space and frequency domains. Secondly, the channel attention mechanism is used to deeply mine the features, and high-attention features can be selected for different image scenes, thereby reducing feature redundancy and improving the accuracy and reliability of anomaly detection. Finally, by comprehensively using the space-frequency domain anomaly feature index and the spectral feature comprehensive index, the comprehensiveness and accuracy of anomaly detection are enhanced, providing strong technical support for the timely discovery and response of surface abnormal events.

[0110] Next, a surface anomaly detection device based on space-frequency features and spectrum features proposed in accordance with an embodiment of the present application will be described with reference to the accompanying drawings.

[0111] Fig.12 is a block diagram of a surface anomaly detection device based on space-frequency features and spectral features according to an embodiment of the present application. Fig.12 As shown, the surface anomaly detection device 10 based on space-frequency features and spectrum features includes: an acquisition module 100, a construction module 200 and a detection module 300.

[0112] Specifically, the acquisition module 100 is used to obtain a surface anomaly knowledge graph.

[0113] The construction module 200 is used to obtain a single-view hyperspectral image, determine at least one image scene of the single-view hyperspectral image, and decompose the single-view hyperspectral image into spatial and frequency domain comprehensive feature information, so as to construct a spatial and frequency domain abnormal feature index using the spatial and frequency domain comprehensive feature information.

[0114] The detection module 300 is used to perform deep mining on multiple features based on the surface anomaly knowledge graph using the channel attention mechanism, so as to select high-attention features that meet preset conditions for different scenes of at least one image scene, and construct a comprehensive index of graph features for surface anomaly discovery, so as to obtain anomaly detection results using the spatial-frequency domain anomaly feature index and the comprehensive index of graph features.

[0115] Optionally, in one embodiment of the present application, the construction module 200 includes: a reading unit, a first calculation unit, a second calculation unit and a principal component transformation unit.

[0116] The reading unit is used to read the single-view hyperspectral image; the conversion unit is used to convert the single-view hyperspectral image into an image in double-precision format.

[0117] The first calculation unit is used to iterate each pixel of the image to calculate the outer product and the average value of each band value of each pixel, and calculate the covariance matrix according to the outer product and the average value of each band value of each pixel.

[0118] The second calculation unit is used to calculate at least one eigenvalue and eigenvector of the covariance matrix to sort at least one eigenvalue, determine multiple main components, and calculate the contribution rate of each eigenvalue to find the main component whose cumulative contribution rate is greater than a preset threshold.

[0119] The principal component transformation unit is used to determine the eigenvector corresponding to the principal component, so as to read the values ​​of each band at each pixel position and perform principal component transformation to obtain a matrix after principal component transformation.

[0120] Optionally, in one embodiment of the present application, the construction module 200 includes: a normalization unit and a standardization unit.

[0121] Among them, the normalization unit is used to normalize the matrix after the principal component transformation, so as to call the discrete fractional Fourier transform function for each pixel point of the matrix after the principal component transformation, calculate the entropy of each band, and take the maximum value to assign to FrFE to obtain the corresponding order.

[0122] The standardization unit is used to process pixel points in a loop, call a function to perform discrete Fourier transform, calculate the eigenvector matrix of DFRFT, calculate DFRFT using the eigenvector matrix and the input vector to obtain result data, and standardize the result data.

[0123] Optionally, in an embodiment of the present application, it further includes: a feedback module 400 and a cropping module 500 .

[0124] Specifically, the feedback module 400 is used to generate feedback data through the space-frequency domain abnormal feature index and the spectrum feature comprehensive index.

[0125] The cropping module 500 is used to crop the image using the feedback data to mark abnormal areas.

[0126] It should be noted that the aforementioned explanation of the embodiment of the surface anomaly detection method based on space-frequency characteristics and spectrum characteristics is also applicable to the surface anomaly detection device based on space-frequency characteristics and spectrum characteristics of this embodiment, and will not be repeated here.

[0127] According to the surface anomaly detection device based on space-frequency features and spectral features proposed in the embodiment of the present application, the extraction efficiency and accuracy of abnormal features can be effectively improved by acquiring the surface anomaly knowledge graph and the single-scene hyperspectral image, and combining the comprehensive feature information in the space and frequency domains. Secondly, the channel attention mechanism is used to deeply mine the features, and high-attention features can be selected for different image scenes, thereby reducing feature redundancy and improving the accuracy and reliability of anomaly detection. Finally, by comprehensively using the space-frequency domain anomaly feature index and the spectral feature comprehensive index, the comprehensiveness and accuracy of anomaly detection are enhanced, providing strong technical support for the timely discovery and response of surface abnormal events.

[0128] Fig.13 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application. The electronic device may include:

[0129] A memory 1301 , a processor 1302 , and a computer program stored in the memory 1301 and executable on the processor 1302 .

[0130] When the processor 1302 executes the program, the surface anomaly detection method based on space-frequency features and spectrum features provided in the above embodiment is implemented.

[0131] Furthermore, the electronic device further comprises:

[0132] The communication interface 1303 is used for communication between the memory 1301 and the processor 1302 .

[0133] The memory 1301 is used to store computer programs that can be executed on the processor 1302 .

[0134] The memory 1301 may include a high-speed RAM memory, and may also include a non-volatile memory (non-volatile memory), such as at least one disk memory.

[0135] If the memory 1301, the processor 1302 and the communication interface 1303 are implemented independently, the communication interface 1303, the memory 1301 and the processor 1302 can be connected to each other through a bus and communicate with each other. The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus. The bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Fig.13 Only one thick line is used in the diagram, but this does not mean that there is only one bus or only one type of bus.

[0136] Optionally, in a specific implementation, if the memory 1301, the processor 1302 and the communication interface 1303 are integrated on a chip, the memory 1301, the processor 1302 and the communication interface 1303 can communicate with each other through an internal interface.

[0137] The processor 1302 may be a central processing unit (CPU), or an application specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of the present application.

[0138] An embodiment of the present application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-mentioned surface anomaly detection method based on space-frequency features and spectral features.

[0139] An embodiment of the present application also provides a computer program, including a computer program, which, when executed, is used to implement the above-mentioned surface anomaly detection method based on space-frequency features and spectral features.

[0140] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or N embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification and the features of the different embodiments or examples, without contradiction.

[0141] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include at least one of the features. In the description of this application, "N" means at least two, such as two, three, etc., unless otherwise clearly and specifically defined.

[0142] Any process or method description in a flowchart or otherwise described herein may be understood to represent a module, fragment or portion of code comprising one or N executable instructions for implementing the steps of a custom logical function or process, and the scope of the preferred embodiments of the present application includes alternative implementations in which functions may not be performed in the order shown or discussed, including performing functions in a substantially simultaneous manner or in reverse order depending on the functions involved, which should be understood by technicians in the technical field to which the embodiments of the present application belong.

[0143] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute instructions), or in combination with these instruction execution systems, devices or apparatuses. For the purpose of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in combination with these instruction execution systems, devices or apparatuses. More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection with one or N wirings (electronic devices), a portable computer disk box (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), a fiber optic device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be paper or other suitable medium on which the program is printed, since the program may be obtained electronically by optically scanning the paper or other medium and then editing, interpreting or processing in other suitable ways as necessary and then storing it in a computer memory.

[0144] It should be understood that the various parts of the present application can be implemented by hardware, software, firmware or a combination thereof. In the above embodiment, the N steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. If implemented by hardware, as in another embodiment, it can be implemented by any one or a combination of multiple of the following technologies known in the art: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0145] A person skilled in the art may understand that all or part of the steps in the method for implementing the above-mentioned embodiment may be completed by instructing related hardware through a program, and the program may be stored in a computer-readable storage medium, which, when executed, includes one or a combination of the steps of the method embodiment.

[0146] In addition, each functional unit in each embodiment of the present application may be integrated into a processing module, or each unit may exist physically separately, or two or more units may be integrated into one module. The above-mentioned integrated module may be implemented in the form of hardware or in the form of a software functional module. If the integrated module is implemented in the form of a software functional module and sold or used as an independent product, it may also be stored in a computer-readable storage medium.

[0147] The storage medium mentioned above may be a read-only memory, a magnetic disk or an optical disk, etc. Although the embodiments of the present application have been shown and described above, it can be understood that the above embodiments are exemplary and cannot be understood as limiting the present application. A person of ordinary skill in the art may change, modify, replace and modify the above embodiments within the scope of the present application.

Claims

1. A surface anomaly detection method based on space-frequency features and spectral features, characterized in that: The following steps are involved: Obtaining the knowledge graph of surface anomalies; Acquire a single-view hyperspectral image, determine at least one image scene of the single-view hyperspectral image, and decompose the single-view hyperspectral image into spatial and frequency domain comprehensive feature information, so as to construct a spatial and frequency domain abnormal feature index using the spatial and frequency domain comprehensive feature information; Based on the surface anomaly knowledge graph, a channel attention mechanism is used to deeply mine multiple features, so as to select high-attention features that meet preset conditions for different scenes of the at least one image scene, and construct a comprehensive index of graph features for surface anomaly discovery, so as to obtain anomaly detection results using the space-frequency domain anomaly feature index and the graph feature comprehensive index.

2. The method according to claim 1, characterized in that The step of acquiring a single-view hyperspectral image, determining at least one image scene of the single-view hyperspectral image, and decomposing the single-view hyperspectral image into spatial and frequency domain comprehensive feature information, so as to construct a spatial and frequency domain abnormal feature index by using the spatial and frequency domain comprehensive feature information, includes: Reading the single-view hyperspectral image; Converting the single-view hyperspectral image into an image in double-precision format; Iterate each pixel of the image to calculate the outer product and the average value of each band value of each pixel, and calculate the covariance matrix according to the outer product and the average value of each band value of each pixel; Calculate at least one eigenvalue and eigenvector of the covariance matrix to sort at least one eigenvalue, determine a plurality of main components, and calculate the contribution rate of each eigenvalue to find the main component whose cumulative contribution rate is greater than a preset threshold; The eigenvector corresponding to the principal component is determined to read the values ​​of each band at each pixel position and perform principal component transformation to obtain a matrix after principal component transformation.

3. The method according to claim 2, characterized in that The method of acquiring a single-view hyperspectral image, determining at least one image scene of the single-view hyperspectral image, and decomposing the single-view hyperspectral image into spatial and frequency domain comprehensive feature information, so as to construct a spatial and frequency domain abnormal feature index by using the spatial and frequency domain comprehensive feature information, further includes: Normalizing the matrix after the principal component transformation, calling the discrete fractional Fourier transform function for each pixel point of the matrix after the principal component transformation, calculating the entropy of each band, and taking the maximum value to assign to FrFE to obtain the corresponding order; The pixel points are processed in a loop, a function is called to perform a discrete Fourier transform, an eigenvector matrix of DFRFT is calculated, the DFRFT is calculated using the eigenvector matrix and the input vector to obtain result data, and the result data is standardized.

4. The method according to claim 1, characterized in that: Also includes: Generate feedback data through the space-frequency domain abnormal characteristic index and the spectrum characteristic comprehensive index; The feedback data is used to perform image cropping to mark abnormal areas.

5. A surface anomaly detection device based on space-frequency characteristics and spectrum characteristics, characterized in that: include: An acquisition module is used to obtain a knowledge graph of surface anomalies; A construction module is used to obtain a single-view hyperspectral image, determine at least one image scene of the single-view hyperspectral image, and decompose the single-view hyperspectral image into spatial and frequency domain comprehensive feature information, so as to construct a spatial and frequency domain abnormal feature index using the spatial and frequency domain comprehensive feature information; A detection module is used to perform deep mining on multiple features based on the surface anomaly knowledge graph using a channel attention mechanism, so as to select high-attention features that meet preset conditions for different scenes of the at least one image scene, and construct a comprehensive index of graph features for surface anomaly discovery, so as to obtain anomaly detection results using the space-frequency domain anomaly feature index and the comprehensive index of graph features.

6. The device according to claim 5, characterized in that The building blocks include: A reading unit, used for reading the single-view hyperspectral image; A conversion unit, used for converting the single-view hyperspectral image into an image in double-precision format; A first calculation unit is used to iterate each pixel of the image to calculate the outer product and the average value of each band value of each pixel, and calculate the covariance matrix according to the outer product and the average value of each band value of each pixel; A second calculation unit is used to calculate at least one eigenvalue and eigenvector of the covariance matrix, so as to sort the at least one eigenvalue, determine a plurality of main components, and calculate the contribution rate of each eigenvalue, and find out the main component whose cumulative contribution rate is greater than a preset threshold; The principal component transformation unit is used to determine the eigenvector corresponding to the principal component, so as to read the values ​​of each band at each pixel position and perform principal component transformation to obtain a matrix after principal component transformation.

7. The device according to claim 5, characterized in that The building blocks include: A normalization unit is used to perform normalization processing on the matrix after the principal component transformation, so as to call the discrete fractional Fourier transform function for each pixel point of the matrix after the principal component transformation, calculate the entropy of each band, and take the maximum value to assign to FrFE to obtain the corresponding order; The standardization unit is used for cyclically processing pixel points, calling a function to perform discrete Fourier transform, calculating an eigenvector matrix of DFRFT, using the eigenvector matrix and an input vector to calculate DFRFT to obtain result data, and performing standardization on the result data.

8. An electronic device, characterized in that: include: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the surface anomaly detection method based on space-frequency features and spectral features as described in any one of claims 1 to 4.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: The program is executed by a processor to implement the surface anomaly detection method based on space-frequency features and spectral features as described in any one of claims 1 to 4.

10. A computer program product, comprising a computer program, characterized in that The computer program is executed to implement the surface anomaly detection method based on space-frequency features and spectral features as described in any one of claims 1 to 4.

Images