Iterative spectrum-space hyperspectral anomaly detection method and system based on multi-feedback mechanism

Abstract

The invention provides an iterative spectrum-space hyperspectral anomaly detection method and system based on a multi-feedback mechanism. The method comprises the steps of obtaining a to-be-detected hyperspectral image; adopting a traditional anomaly detection algorithm to calculate an anomaly score of each pixel point of the hyperspectral image, and generating a spectral anomaly graph; binarization processing is carried out on the spectrum abnormal image, an abnormal target area is extracted, and a spectrum foreground image is generated; performing spatial filtering processing on the spectral anomaly graph, extracting local spatial information, and generating a spatial filtering graph; combining the spectral anomaly graph, the spectral foreground graph and the spatial filtering graph according to weights to obtain a new hyperspectral image; updating the original hyperspectral image with the new hyperspectral image; repeating the steps; after each iteration, calculating the similarity between the spectral foreground images generated in the current iteration and the previous iteration; if the similarity reaches a preset convergence threshold value, iteration is stopped, and a final anomaly detection result is output. According to the detection method provided by the invention, the detection precision is improved, and the calculation efficiency is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of hyperspectral anomaly detection, and particularly to an iterative spectral-spatial hyperspectral anomaly detection method and system based on a multi-feedback mechanism. Background Art

[0002] Hyperspectral imaging technology has been widely applied in the fields of remote sensing, environmental monitoring, and target detection because it can simultaneously obtain the spatial and spectral information of targets. However, hyperspectral data brings significant challenges to the detection of abnormal targets due to its high dimensionality and noise complexity.

[0003] Hyperspectral anomaly detection (HAD) aims to distinguish abnormal targets and background information from hyperspectral images. Hyperspectral data usually presents a three-dimensional structure, where the spectral dimension contains rich material composition information, while the spatial dimension provides the shape and distribution characteristics of targets. The main technical difficulties in its research are that the background of hyperspectral data is complex and the spectral variations are diverse, making it easy for the detection of abnormal targets to be interfered by the background. At the same time, there may be noise interference in hyperspectral images, which poses higher requirements for the robustness of detection algorithms, and how to effectively combine high-dimensional spectral information and spatial information has become the key to improving detection performance.

[0004] In recent years, statistical methods based on spectral features, low-rank sparse decomposition, tensor decomposition, and deep learning methods have made great progress. In particular, the joint utilization of spectral and spatial information has significantly improved the separation ability of abnormal targets.

[0005] Traditional statistical methods, such as the Reed-Xiaoli anomaly detector (RXAD), are one of the most classic hyperspectral anomaly detection methods. RXAD uses the spectral mean and covariance matrix to distinguish the background and targets by calculating the Mahalanobis distance. However, this method has poor adaptability to high-dimensional data and is vulnerable to noise.

[0006] Methods based on low-rank sparse decomposition (LRSD) assume that hyperspectral images consist of a low-rank background and sparse abnormal targets. Robust principal component analysis (RPCA) and low-rank sparse matrix decomposition (LRaSMD) are typical methods among them, which can effectively separate the background and abnormal targets. However, these methods often lose the three-dimensional structural characteristics of hyperspectral images.

[0007] Tensor decomposition methods can preserve the three-dimensional characteristics of hyperspectral data. Models based on tensor principal component analysis (TPCA) and low-rank tensor representation (such as PCA-TLRSR) can model the background more accurately. However, these methods rely on complex regularization strategies and have high computational costs.

[0008] Joint methods of spectral and spatial information have become the mainstream of research in recent years. For example: the dual-window RX algorithm: combines local and global window analysis of spectral features, improving the detection ability of abnormal targets. Total variation regularization (TV): optimizes target separation by jointly considering spectral smoothness and spatial sparsity.

[0009] Deep learning techniques have also developed rapidly in the field of hyperspectral anomaly detection. For example: methods based on convolutional neural networks (CNNs) extract spectral-spatial features through multi-scale convolution. Generative adversarial networks (GANs) are used to generate high-quality foreground maps and spatial filter maps, thus improving the detection performance. However, deep learning methods rely on a large amount of labeled data and their generalization ability is limited. Summary of the Invention

[0010] In view of this, embodiments of the present invention provide an iterative spectral-spatial hyperspectral anomaly detection method and system based on a multi-feedback mechanism to eliminate or improve one or more defects existing in the prior art.

[0011] On the one hand, the present invention provides an iterative spectral-spatial hyperspectral anomaly detection method based on a multi-feedback mechanism, the method comprising the following steps:

[0012] Obtain a hyperspectral image to be detected;

[0013] In one iteration, use a traditional anomaly detection algorithm to calculate the anomaly score of each pixel point of the hyperspectral image, generating a spectral anomaly map; perform binary processing on the spectral anomaly map to extract the abnormal target area, generating a spectral foreground map; perform spatial filtering on the spectral anomaly map to extract local spatial information, generating a spatial filter map; combine the spectral anomaly map, the spectral foreground map, and the spatial filter map according to a preset weight to obtain a new hyperspectral image; update the original hyperspectral image with the new hyperspectral image; perform the above iterative steps;

[0014] After each iteration, calculate the similarity between the current iteration and the spectral foreground map generated in the previous iteration; if the similarity reaches a preset convergence threshold, stop the iteration and output the final anomaly detection result.

[0015] In some embodiments of the present invention, using a traditional anomaly detection algorithm to calculate the anomaly score of each pixel point of the hyperspectral image includes:

[0016] The Reed-Xiaoli anomaly detection algorithm is used to calculate the anomaly score of each pixel, and the calculation formula is:

[0017] ADMap(i,j) = (X(i,j,:) - μ) T Σ -1 (X(i,j,:) - μ);

[0018] Among them, ADMap is the spectral anomaly map, which is a two-dimensional matrix representing the anomaly score of each pixel; represents the spectral mean vector; represents the covariance matrix; (i,j) represents the pixel position in the hyperspectral image.

[0019] In some embodiments of the present invention, the spectral anomaly map is binarized to extract the anomaly target area, including:

[0020] The Otsu threshold segmentation algorithm is used to process the spectral anomaly map, automatically calculate the segmentation threshold, and mark the part of the spectral anomaly map with pixel values greater than the segmentation threshold as the anomaly target area;

[0021] The formula of the Otsu threshold segmentation algorithm is:

[0022]

[0023] Among them, is the spectral foreground map, which is a binary matrix marking the anomaly pixel positions; ADMao represents the spectral anomaly map; τ represents the segmentation threshold automatically calculated by the Otsu threshold segmentation algorithm.

[0024] In some embodiments of the present invention, the spectral anomaly map is spatially filtered to extract local spatial information, including:

[0025] The Gaussian filter is used to process the spectral anomaly map to reflect the anomaly degree around each pixel, and the calculation formula is:

[0026] SFMap(i,j) = ∑ p,q G(p,q)·ADMap(i + p,j + q);

[0027] In the formula, G(p,q) represents the Gaussian kernel function, which is used to represent the coordinates of the pixel offset in the two-dimensional plane; p represents the offset of the current pixel relative to the central pixel in the horizontal direction; q represents the offset of the current pixel relative to the central pixel in the vertical direction; the formula of the Gaussian kernel function is:

[0028]

[0029] Among them, SFMap is the spatial filter map, which is a continuous value matrix used to reflect the local anomaly distribution of each pixel; ADMap represents the spectral anomaly map; σ is the standard deviation of the Gaussian kernel, which is used to control the smoothing intensity of the filter.

[0030] In some embodiments of the present invention, the spectral anomaly map, the spectral foreground map, and the spatial filter map are combined according to a preset weight to obtain a new hyperspectral image, including:

[0031] X ′ (i,j,:) = X(i,j,:) ⊙ FGMap(i,j) + λ · SFMap(i,j);

[0032] Among them, X ′ (i,j,:) represents the new hyperspectral image; X(i,j,:) represents the original hyperspectral image; ⊙ represents element-wise multiplication; FGMap represents the spectral foreground map; λ is the preset weight parameter used to balance the influence of the spectral foreground map and the spatial filter map; SFMap represents the spatial filter map.

[0033] In some embodiments of the present invention, after each iteration, the similarity between the current iteration and the spectral foreground map generated in the previous iteration is calculated, including:

[0034] The Tanimoto index is used to measure the similarity, and the calculation formula of the Tanimoto index is:

[0035]

[0036] Among them, FGMap k represents the spectral foreground map generated in the k-th iteration; ‖·‖ represents the number of elements in the set; ∩ and ∪ represent the set intersection and union respectively.

[0037] In some embodiments of the present invention, the similarity is compared with the preset convergence threshold, and the calculation formula is:

[0038] Tanimoto ≥ ∈;

[0039] Among them, Tanimoto represents the similarity; ∈ represents the preset convergence threshold;

[0040] When the above calculation formula is satisfied, the iteration is stopped, and the final anomaly detection result is output.

[0041] On the other hand, the present invention also provides an iterative spectral-spatial hyperspectral anomaly detection system based on a multi-feedback mechanism. The system is executed to implement the steps of any one of the methods mentioned above. The system includes:

[0042] A data input module, configured to obtain a hyperspectral image to be detected and input it into the data detection module;

[0043] A data detection module, including an initial detection module, a feedback generation module, and a data update module, configured to iteratively execute the following steps: In the initial detection module, a traditional anomaly detection algorithm is used to calculate the anomaly score of each pixel point of the hyperspectral image, generating a spectral anomaly map; In the feedback generation module, the spectral anomaly map is binarized to extract the anomaly target area, generating a spectral foreground map; The spectral anomaly map is spatially filtered to extract local spatial information, generating a spatial filter map; In the data update module, the spectral anomaly map, the spectral foreground map, and the spatial filter map are combined according to a preset weight to obtain a new hyperspectral image; The new hyperspectral image is used to update the original hyperspectral image;

[0044] A dynamic stop module, configured to calculate the similarity between the current iteration and the spectral foreground map generated in the previous iteration after each iteration of the data detection module; If the similarity reaches a preset convergence threshold, the iteration is stopped;

[0045] A result output module, configured to output the final anomaly detection result after the dynamic stop module stops the iteration.

[0046] On the other hand, the present invention also provides a computer-readable storage medium, on which a computer program is stored, and when the program is executed by a processor, the steps of any one of the methods mentioned above are implemented.

[0047] On the other hand, the present invention also provides a computer program product, including a computer program / instructions, and when the computer program / instructions are executed by a processor, the steps of any one of the methods mentioned above are implemented.

[0048] The present invention provides an iterative spectral-spatial hyperspectral anomaly detection method and system based on a multi-feedback mechanism, including: acquiring a hyperspectral image to be detected; calculating the anomaly score of each pixel point of the hyperspectral image using a traditional anomaly detection algorithm to generate a spectral anomaly map; performing binary processing on the spectral anomaly map to extract the anomaly target area and generate a spectral foreground map; performing spatial filtering processing on the spectral anomaly map to extract local spatial information and generate a spatial filtering map; combining the spectral anomaly map, the spectral foreground map, and the spatial filtering map according to weights to obtain a new hyperspectral image; updating the original hyperspectral image with the new hyperspectral image; repeating the above steps; after each iteration, calculating the similarity between the current iteration and the spectral foreground map generated in the previous iteration; if the similarity reaches a preset convergence threshold, stop the iteration and output the final anomaly detection result. The detection method provided by the present invention improves the detection accuracy, significantly enhances the calculation efficiency, reduces the dependence on labeled data, and has strong practicability and application prospects.

[0049] Additional advantages, objects, and features of the present invention will be partly described below, and will partly become apparent to those of ordinary skill in the art after studying the following, or can be learned from the practice of the present invention. The objects and other advantages of the present invention can be realized and obtained by the structure specifically pointed out in the specification and the drawings.

[0050] Those skilled in the art will understand that the objects and advantages that can be achieved by the present invention are not limited to the above specifically described, and the above and other objects that the present invention can achieve will be more clearly understood according to the following detailed description. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] The drawings described herein are used to provide a further understanding of the present invention, form a part of this application, and do not constitute a limitation to the present invention. In the drawings:

[0052] Figure 1 It is a schematic diagram of the steps of an iterative spectral-spatial hyperspectral anomaly detection method based on a multi-feedback mechanism in an embodiment of the present invention.

[0053] Figure 2 It is a flow chart of an iterative spectral-spatial hyperspectral anomaly detection method based on a multi-feedback mechanism in an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0054] To make the objects, technical solutions, and advantages of the present invention clearer and more understandable, the present invention will be further described in detail below in conjunction with the embodiments and the drawings. Herein, the illustrative embodiments of the present invention and their descriptions are used to explain the present invention, but do not limit the present invention.

[0055] Here, it should also be noted that in order to avoid obscuring the present invention with unnecessary details, only the structures and / or processing steps closely related to the solution according to the present invention are shown in the drawings, while other details less relevant to the present invention are omitted.

[0056] It should be emphasized that the term "comprising / including" as used herein refers to the presence of features, elements, steps or components, but does not exclude the presence or addition of one or more other features, elements, steps or components.

[0057] Here, it should also be noted that if not otherwise specified, the term "connection" in this text can not only refer to direct connection, but also represent indirect connection with intermediaries.

[0058] In the following, embodiments of the present invention will be described with reference to the drawings. In the drawings, the same reference numerals represent the same or similar components, or the same or similar steps.

[0059] It should be emphasized here that the step labels mentioned hereinafter do not limit the order of the steps. Instead, it should be understood that the steps can be executed in the order mentioned in the embodiments, or different from the order in the embodiments, or several steps can be executed simultaneously.

[0060] In order to solve the problems existing in the existing hyperspectral anomaly detection technology, such as limited background suppression ability, insufficient fusion of spectral and spatial information, poor robustness, low iterative efficiency, poor convergence, and the dependence of deep learning technology on data annotation, the present invention proposes an iterative spectral-spatial hyperspectral anomaly detection method based on a multi-feedback mechanism, as Figure 1 shown, the method includes the following steps S101 to S103:

[0061] Step S101: Obtain the hyperspectral image to be detected.

[0062] Step S102: In one iteration, use a traditional anomaly detection algorithm to calculate the anomaly score of each pixel point of the hyperspectral image, generate a spectral anomaly map; perform binary processing on the spectral anomaly map to extract the anomaly target area and generate a spectral foreground map; perform spatial filtering on the spectral anomaly map to extract local spatial information and generate a spatial filtering map; combine the spectral anomaly map, the spectral foreground map and the spatial filtering map according to a preset weight to obtain a new hyperspectral image; update the original hyperspectral image with the new hyperspectral image; execute the above iterative steps.

[0063] Step S103: After each iteration, calculate the similarity between the current iteration and the spectral foreground map generated in the previous iteration; if the similarity reaches a preset convergence threshold, stop the iteration and output the final anomaly detection result.

[0064] As Figure 2As shown, it is a flowchart of an iterative spectral-spatial hyperspectral anomaly detection method based on a multi-feedback mechanism.

[0065] In step S101, hyperspectral image data to be detected is acquired. Among them, a hyperspectral image usually consists of multiple bands, and each pixel contains information in multiple spectral dimensions. The hyperspectral image data structure is a three-dimensional data cube, where X and Y are spatial dimensions, and Z is the spectral dimension.

[0066] In step S102, each iteration step can be subdivided into: step S1021 to generate a spectral anomaly map, step S1022 to generate a spectral foreground map, step S1023 to generate a spatial filter map, and step S1024 for data update.

[0067] In step S1021, a traditional anomaly detection algorithm is used to calculate the anomaly score of each pixel point in the hyperspectral image, generating a spectral anomaly map (ADMap).

[0068] In some embodiments, the traditional anomaly detection algorithm uses the Reed-Xiaoli anomaly detection algorithm.

[0069] The spectral information of the Reed-Xiaoli anomaly detection algorithm is compared with the global background model, and its calculation formula is shown as formula (1):

[0070] ADMap(i,j)=(X(i,j,:)-μ) T Σ -1 (X(i,j,:)-μ); (1)

[0071] Among them, ADMap is the spectral anomaly map, which is a two-dimensional matrix representing the anomaly score of each pixel; represents the spectral mean vector; represents the covariance matrix; (i,j) represents the pixel position in the hyperspectral image.

[0072] By this method, a spectral anomaly map is generated, which shows the anomaly degree of each pixel in the hyperspectral image.

[0073] In step S1022, the spectral anomaly map is binarized to extract the anomaly target area, generating a spectral foreground map (FGMap). Among them, binarization is the process of converting the pixel values in the spectral anomaly map into two possible states (usually 0 and 1, or black and white).

[0074] In some embodiments, the binarization method uses the Otsu threshold segmentation algorithm to automatically calculate the segmentation threshold, and marks the part of the pixel values in the spectral anomaly map that are greater than the segmentation threshold as the anomaly target area.

[0075] The spectral anomaly map is binarized using the Otsu threshold segmentation algorithm to extract the foreground region (i.e., potential anomaly targets), and the calculation formula is shown in formula (2):

[0076]

[0077] where FGMap ∈ {0, 1} n1×n3 is the spectral foreground map, which is a binary matrix marking the positions of abnormal pixels; ADMap represents the spectral anomaly map; τ represents the segmentation threshold automatically calculated by the Otsu threshold segmentation algorithm.

[0078] In step S1023, the spectral anomaly map is spatially filtered to extract local spatial information and generate a spatial filter map (SFMap).

[0079] In some embodiments, Gaussian filtering is used to spatially filter the spectral anomaly map to extract local spatial information. The spectral anomaly map is smoothed by the spatial filter to reflect the degree of anomaly around each pixel, and the calculation formula of Gaussian filtering is shown in formula (3):

[0080] SFMap(i, j) = ∑ p,q G(p, q)·ADMap(i + p, j + q); (3)

[0081] In formula (3), G(p, q) represents the Gaussian kernel function, which is used to represent the coordinates of the pixel offset in the two-dimensional plane; p represents the horizontal offset of the current pixel relative to the central pixel; q represents the vertical offset of the current pixel relative to the central pixel; the Gaussian kernel function is shown in formula (4):

[0082]

[0083] where SFMap is the spatial filter map, which is a continuous value matrix used to reflect the local anomaly distribution of each pixel; ADMap represents the spectral anomaly map; σ is the standard deviation of the Gaussian kernel, which is used to control the smoothing intensity of the spatial filter.

[0084] The Gaussian kernel function G(p, q) is a weight function based on two-dimensional coordinates, which is used to assign different weights to the pixels around the central pixel during the convolution process, and the offset is used to define the action range of the kernel function relative to each pixel of the image.

[0085] In step S1024, the spectral anomaly map, the spectral foreground map, and the spatial filter map are combined according to preset weights to obtain a new hyperspectral image, and the original hyperspectral image is updated with the new hyperspectral image.

[0086] In some embodiments, the spectral anomaly map, the spectral foreground map, and the spatial filter map are combined according to a preset weight to obtain a new hyperspectral image. The calculation formula is shown in formula (5):

[0087] X′(i,j,:) = X(i,j,:) ⊙ FGMap(i,j) + λ · SFMap(i,j); (5)

[0088] Wherein, X′(i,j,:) represents the new hyperspectral image; X(i,j,:) represents the original hyperspectral image; ⊙ represents element-wise multiplication; FGMap represents the spectral foreground map; λ is a preset weight parameter used to balance the influence of the spectral foreground map and the spatial filter map; SFMap represents the spatial filter map.

[0089] Update the original hyperspectral image X with the new hyperspectral image X′. The updated data cube X ′ Can more prominently highlight the abnormal targets while suppressing background interference.

[0090] Repeat step S102 to iterate the spectral information and the spatial information.

[0091] In step S103, a dynamic stopping rule is designed. After each iteration, calculate the similarity between the current iteration and the spectral foreground map generated in the previous iteration; if the similarity reaches the preset convergence threshold, stop the iteration and output the final anomaly detection result.

[0092] In some embodiments, the Tanimoto index is used to measure the similarity. Among them, the calculation formula of the Tanimoto index can be shown in formula (6):

[0093]

[0094] Wherein, FGMap k Represents the spectral foreground map generated in the k-th iteration; ‖·‖ represents the number of elements in the set; ∩ and ∪ represent the set intersection and union respectively.

[0095] Based on formula (6), when comparing the similarity with the preset convergence threshold, when the calculation formula is as formula (7):

[0096] Tanimoto ≥ ∈; (7)

[0097] When formula (7) is satisfied, it indicates that the change of the feedback map has tended to be stable, stop the iteration, and output the final anomaly detection result. Among them, Tanimoto represents the similarity; ∈ represents the preset convergence threshold;

[0098] The generated final anomaly detection result can indicate the location and range of the abnormal targets through multiple iterations, helping with further analysis and application.

[0099] Correspondingly, the present invention also provides an iterative spectral-spatial hyperspectral anomaly detection system based on a multi-feedback mechanism. The system includes:

[0100] A data input module, configured to obtain a hyperspectral image to be detected and input it into the data detection module.

[0101] The data detection module further includes an initial detection module, a feedback generation module, and a data update module, and is configured to iteratively execute the following steps: In the initial detection module, a traditional anomaly detection algorithm is used to calculate the anomaly score of each pixel point of the hyperspectral image, and a spectral anomaly map is generated; in the feedback generation module, the spectral anomaly map is binarized to extract the anomaly target area, and a spectral foreground map is generated; the spectral anomaly map is spatially filtered to extract local spatial information, and a spatial filter map is generated; in the data update module, the spectral anomaly map, the spectral foreground map, and the spatial filter map are combined according to a preset weight to obtain a new hyperspectral image; the new hyperspectral image is used to update the original hyperspectral image.

[0102] A dynamic stop module, configured to calculate the similarity between the current iteration and the spectral foreground map generated in the previous iteration after each iteration of the data detection module; if the similarity reaches a preset convergence threshold, the iteration is stopped.

[0103] A result output module, configured to output the final anomaly detection result after the dynamic stop module stops the iteration.

[0104] The present invention also provides a computer-readable storage medium, on which a computer program is stored. When the program is executed by a processor, the steps of the iterative spectral-spatial hyperspectral anomaly detection method based on a multi-feedback mechanism are implemented.

[0105] Correspondingly, the present invention also provides a device, which includes a computer device. The computer device includes a processor and a memory. A computer instruction is stored in the memory, and the processor is configured to execute the computer instruction stored in the memory. When the computer instruction is executed by the processor, the device implements the steps of the method described above.

[0106] An embodiment of the present invention also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the foregoing edge computing server deployment method are implemented. The computer-readable storage medium may be a tangible storage medium, such as a random access memory (RAM), a memory, a read-only memory (ROM), an electrically programmable ROM, an electrically erasable programmable ROM, a register, a floppy disk, a hard disk, a removable storage disk, a CD-ROM, or any other form of storage medium known in the art.

[0107] In summary, the present invention provides an iterative spectral-spatial hyperspectral anomaly detection method and system based on a multi-feedback mechanism, and designs a dynamic multi-feedback mechanism, a dynamic stopping rule, an unsupervised learning mechanism, an efficient spectral and spatial information fusion, iterative optimization, and a data update strategy.

[0108] By introducing a multi-feedback mechanism, including a spectral anomaly map, a spectral foreground map, and a spatial filter map, the deep fusion of spectral and spatial information is achieved. The hyperspectral data is optimized from multiple dimensions, the distinction between abnormal targets and the background is enhanced, the background suppression ability is improved, and the accuracy and robustness of hyperspectral anomaly detection are improved. Further, in each iteration, according to the change of the feedback map, the weights of each feedback mechanism are dynamically adjusted. This enables the algorithm to flexibly adjust the feedback strategy according to the characteristics of the data, adapt to the detection requirements of different backgrounds and targets, and further improve the adaptability and accuracy of the algorithm.

[0109] By introducing a dynamic stopping rule based on the Tanimoto index, when the similarity of the feedback map exceeds the set threshold, the algorithm automatically stops iterating, avoiding unnecessary calculations and improving the detection efficiency.

[0110] By introducing an unsupervised learning mechanism, abnormal targets are identified through the statistical characteristics and spatial structure features of the data itself, reducing the dependence on labeled data and having stronger generalization ability.

[0111] Those of ordinary skill in the art should understand that the various exemplary components, systems, and methods described in conjunction with the embodiments disclosed herein can be implemented in hardware, software, or a combination of both. Specifically, whether to execute in hardware or software depends on the specific application and design constraints of the technical solution. Professional technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of the present invention. When implemented in hardware, it can be, for example, an electronic circuit, an application-specific integrated circuit (ASIC), appropriate firmware, a plug-in, a functional card, etc. When implemented in software, the elements of the present invention are programs or code segments used to perform the required tasks. The program or code segment can be stored in a machine-readable medium or transmitted through a data signal carried in a carrier wave on a transmission medium or a communication link.

[0112] It should be clear that the present invention is not limited to the specific configurations and processes described above and shown in the figures. For the sake of brevity, the detailed description of known methods is omitted here. In the above embodiments, several specific steps are described and shown as examples. However, the method process of the present invention is not limited to the specific steps described and shown, and those skilled in the art can make various changes, modifications, and additions, or change the order between steps after understanding the spirit of the present invention.

[0113] In the present invention, features described and / or illustrated for one embodiment can be used in the same or a similar manner in one or more other embodiments, and / or combined with the features of other embodiments or replace the features of other embodiments.

[0114] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, various modifications and variations can be made to the embodiments of the present invention. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. An iterative spectral-spatial hyperspectral anomaly detection method based on a multi-feedback mechanism, characterized in that: The method comprises the following steps: Acquire a hyperspectral image to be detected; In one iteration, a traditional anomaly detection algorithm is used to calculate the anomaly score of each pixel point of the hyperspectral image to generate a spectral anomaly map; the spectral anomaly map is binarized to extract the abnormal target area and generate a spectral foreground map; the spectral anomaly map is spatially filtered to extract local spatial information and generate a spatial filtering map; the spectral anomaly map, the spectral foreground map and the spatial filtering map are combined according to preset weights to obtain a new hyperspectral image; the new hyperspectral image is used to update the original hyperspectral image; and the above iterative steps are performed; After each iteration, the similarity between the spectral foreground images generated in the current iteration and the previous iteration is calculated; if the similarity reaches a preset convergence threshold, the iteration is stopped and the final anomaly detection result is output.

2. The iterative spectral-spatial hyperspectral anomaly detection method based on a multi-feedback mechanism according to claim 1, characterized in that: The anomaly score of each pixel of the hyperspectral image is calculated using a traditional anomaly detection algorithm, including: The Reed-Xiaoli anomaly detection algorithm is used to calculate the anomaly score of each pixel. The calculation formula is: ADMap(i,j)=(X(i,j,:)-μ) T ∑-1(X(i,j,:)-μ); Among them, ADMap is the spectral anomaly map, which is a two-dimensional matrix representing the anomaly score of each pixel; represents the spectral mean vector; represents the covariance matrix; (i, j) represents the pixel position in the hyperspectral image.

3. The iterative spectral-spatial hyperspectral anomaly detection method based on multi-feedback mechanism according to claim 1 is characterized in that: Binarization is performed on the spectral anomaly map to extract the abnormal target area, including: The spectral anomaly image is processed by using the Otsu threshold segmentation algorithm, the segmentation threshold is automatically calculated, and the part of the spectral anomaly image with a pixel value greater than the segmentation threshold is marked as the abnormal target area; The formula of the Otsu threshold segmentation algorithm is: in, is the spectral foreground map, which is a binary matrix marking the abnormal pixel positions; ADMap represents the spectral anomaly map; τ represents the segmentation threshold automatically calculated by the Otsu threshold segmentation algorithm.

4. The iterative spectral-spatial hyperspectral anomaly detection method based on multi-feedback mechanism according to claim 1, characterized in that: Performing spatial filtering on the spectral anomaly map to extract local spatial information includes: The spectral anomaly map is processed by Gaussian filtering to reflect the degree of anomaly around each pixel. The calculation formula is: SFMap(i,j)=∑ p，q G(p,q)·ADMap(i+p,j+q); Wherein, G(p, q) represents a Gaussian kernel function, which is used to represent the coordinates of the pixel offset in a two-dimensional plane; p represents the horizontal offset of the current pixel relative to the center pixel; q represents the vertical offset of the current pixel relative to the center pixel; the formula of the Gaussian kernel function is: Among them, SFMap is the spatial filter map, which is a continuous value matrix used to reflect the local anomaly distribution of each pixel; ADMap represents the spectral anomaly map; σ is the standard deviation of the Gaussian kernel, which is used to control the smoothing strength of the filter.

5. The iterative spectral-spatial hyperspectral anomaly detection method based on multi-feedback mechanism according to claim 1, characterized in that: The spectral anomaly map, the spectral foreground map and the spatial filter map are combined according to preset weights to obtain a new hyperspectral image, including: X'(i,j,:)=X(i,j,:)⊙FGMap(i,j)+λ·SFMap(i,j); Among them, X′(i, j,:) represents the new hyperspectral image; X(i, j,:) represents the original hyperspectral image; ⊙ represents the point-by-point multiplication of elements; FGMap represents the spectral foreground map; λ is the preset weight parameter used to balance the influence of the spectral foreground map and the spatial filter map; SFMap represents the spatial filter map.

6. The iterative spectral-spatial hyperspectral anomaly detection method based on multi-feedback mechanism according to claim 1, characterized in that: After each iteration, the similarity between the spectral foreground image generated in the current iteration and the previous iteration is calculated, including: The similarity is measured using the Tanimoto index, which is calculated as follows: Among them, FGMap k represents the spectral foreground image generated in the kth iteration; ||·|| represents the number of elements in the set; ∩ and U represent the intersection and union of the sets, respectively.

7. The iterative spectral-spatial hyperspectral anomaly detection method based on multi-feedback mechanism according to claim 6, characterized in that: Compare the similarity with the preset convergence threshold, and the calculation formula is: Tanimoto ≥∈; Wherein, Tanimoto represents the similarity; ∈ represents the preset convergence threshold; When the above calculation formula is satisfied, the iteration is stopped and the final anomaly detection result is output.

8. An iterative spectral-spatial hyperspectral anomaly detection system based on a multi-feedback mechanism, characterized in that: The system is executed to implement the steps of the method according to any one of claims 1 to 7, and the system comprises: A data input module is used to obtain the hyperspectral image to be detected and input it into the data detection module; The data detection module includes an initial detection module, a feedback generation module and a data update module, and is used to iteratively perform the following steps: in the initial detection module, a traditional anomaly detection algorithm is used to calculate the anomaly score of each pixel point of the hyperspectral image to generate a spectral anomaly map; in the feedback generation module, the spectral anomaly map is binarized to extract the abnormal target area to generate a spectral foreground map; the spectral anomaly map is spatially filtered to extract local spatial information to generate a spatial filter map; in the data update module, the spectral anomaly map, the spectral foreground map and the spatial filter map are combined according to preset weights to obtain a new hyperspectral image; the new hyperspectral image is used to update the original hyperspectral image; A dynamic stop module is used to calculate the similarity between the spectrum foreground image generated in the current iteration and the spectrum foreground image generated in the previous iteration after each iteration of the data detection module; if the similarity reaches a preset convergence threshold, the iteration is stopped; The result output module is used to output the final anomaly detection result after the dynamic stop module stops iterating.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer program product comprising a computer program / instructions, characterized in that When the computer program / instructions are executed by a processor, the steps of the method as claimed in any one of claims 1 to 7 are implemented.

Images