A Hyperspectral Image Anomaly Detection Method and System Based on the Mamba Network Model

Abstract

The present application discloses a hyperspectral image anomaly detection method and system based on the Mamba network model, which relates to the technical field of hyperspectral remote sensing image processing and is applied to the Mamba network model including a preset spectral cluster unfolding component, a preset non-local reconstruction feature extraction component, and a preset spectral cluster folding component. The method includes: processing the original hyperspectral image by using the preset spectral cluster unfolding component to obtain a set of spectral clusters to be reconstructed; performing non-local feature extraction and background reconstruction feature learning on the set of spectral clusters to be reconstructed by using the preset non-local reconstruction feature extraction component to obtain a first set of reconstructed spectral clusters; processing the first set of reconstructed spectral clusters by using the preset spectral cluster folding component to obtain a second set of reconstructed spectral clusters; and determining an anomaly detection result map based on the second set of reconstructed spectral clusters. It can prevent the degradation of detection performance caused by limited non-local feature extraction and abnormal target contamination.

Description

Technical Field

[0001] The present invention relates to the technical field of hyperspectral remote sensing image processing, and particularly relates to a hyperspectral image anomaly detection method and system based on a Mamba network model. Background Art

[0002] The target detection technology of hyperspectral images can model the spectral differences between targets and backgrounds to screen out candidate target pixels and perform threshold segmentation on the candidate target pixels, thereby obtaining regions of suspected targets. Among them, when the target characteristics are not fully understood, the spectral information of the target is in an unknown state. At this time, it is necessary to use hyperspectral anomaly detection technology to achieve "blind" detection of small targets. Then, after manual screening and annotation of the target regions, a target sample library is established based on the annotated target regions to provide key information for subsequent target characteristic analysis and recognition decision-making.

[0003] Currently, there are various excellent hyperspectral anomaly detection algorithms on the market, such as anomaly detection methods based on data statistical distribution, collaborative and sparse representation models, low-rank sparse decomposition, and deep neural network models. However, model-based methods need to optimize model parameters according to the background and target characteristics of hyperspectral images, and rely too much on manual features, resulting in limited performance of anomaly detection. In addition, since the hyperspectral image anomaly detection method based on the Mamba network model in the deep neural network model realizes anomaly detection by performing deep representation learning and differential feature extraction on spectral sample data, most of these methods have problems of limited non-local feature extraction and high computational complexity, which limit the network detection performance of anomaly detection in the reconstruction learning mode.

[0004] As can be seen from the above, how to prevent the detection performance degradation accident caused by limited non-local feature extraction and abnormal target contamination in the background reconstruction learning mode during the hyperspectral image anomaly detection process based on the Mamba network model is an urgent problem to be solved at present. Summary of the Invention

[0005] In view of this, the purpose of the present invention is to provide a hyperspectral image anomaly detection method and system based on a Mamba network model, which can prevent the problem of detection performance degradation caused by limited non-local feature extraction and abnormal target contamination in the background reconstruction learning mode during the hyperspectral image anomaly detection process based on the Mamba network model. The specific solutions are as follows:

[0006] In a first aspect, the present application provides a hyperspectral image anomaly detection method based on a Mamba network model, which is applied to a Mamba network model including a preset spectral cluster unfolding component, a preset non-local reconstruction feature extraction component, and a preset spectral cluster folding component. The method includes:

[0007] Using the preset sliding window spectral cluster unfolding component and the first local spatial and spectral feature compensation component in the preset spectral cluster unfolding component to process the original hyperspectral image in sequence, so as to obtain a set of spectral clusters to be reconstructed corresponding to the original hyperspectral image;

[0008] Using the preset non-local reconstruction feature extraction component to perform non-local feature extraction and background reconstruction feature learning on each of the spectral clusters to be reconstructed in the set of spectral clusters to be reconstructed, so as to obtain a first set of reconstructed spectral clusters;

[0009] Using the second local spatial and spectral feature compensation component and the preset sliding window spectral cluster folding component in the preset spectral cluster folding component to process each of the first reconstructed spectral clusters in the first set of reconstructed spectral clusters in sequence, so as to obtain a second set of reconstructed spectral clusters;

[0010] Based on the second set of reconstructed spectral clusters, determining an anomaly detection result map corresponding to the original hyperspectral image.

[0011] Optionally, the step of using the preset sliding window spectral cluster unfolding component and the first local spatial and spectral feature compensation component in the preset spectral cluster unfolding component to process the original hyperspectral image in sequence, so as to obtain a set of spectral clusters to be reconstructed corresponding to the original hyperspectral image, includes:

[0012] Using the preset sliding window spectral cluster unfolding component in the preset spectral cluster unfolding component to perform spectral cluster unfolding on the original hyperspectral image, so as to obtain a number of spectral clusters to be processed corresponding to the original hyperspectral image; the sizes of the spectral clusters to be processed are the same;

[0013] In the first local spatial and spectral feature compensation component, using a number of depthwise separable convolution branches to perform feature extraction on each of the spectral clusters to be processed respectively, so as to obtain corresponding first local spatial and spectral neighborhood features; the convolution kernel sizes corresponding to the depthwise separable convolution branches are all different;

[0014] Using a preset splicing and convolution rule to splice and convolve the first local spatial and spectral neighborhood features respectively, so as to obtain first splicing features corresponding to the first local spatial and spectral neighborhood features respectively;

[0015] In the first local space and the spectral feature compensation component, several depthwise separable convolution branches are used again to separately extract corresponding second local space and spectral neighborhood features from each of the first stitching features, and the preset stitching and convolution rules are used to stitch and convolve each of the second local space and spectral neighborhood features to obtain second stitching features corresponding to each of the second local space and spectral neighborhood features;

[0016] The preset residual connection is used to fuse each of the to-be-processed spectral clusters with the corresponding second stitching feature to obtain several to-be-reconstructed spectral clusters and a to-be-reconstructed spectral cluster set including each of the to-be-reconstructed spectral clusters.

[0017] Optionally, the preset non-local reconstruction feature extraction component is used to separately perform non-local feature extraction and background reconstruction feature learning on each of the to-be-reconstructed spectral clusters in the to-be-reconstructed spectral cluster set to obtain a first reconstructed spectral cluster set, including:

[0018] In the preset non-local reconstruction feature extraction component, a conversion operation is separately performed on each of the to-be-reconstructed spectral clusters in the to-be-reconstructed spectral cluster set to obtain corresponding initial two-dimensional sequence features;

[0019] The first convolutional layer, root mean square layer normalization layer, and linear transformation layer in the preset non-local reconstruction feature extraction component are used to sequentially process each of the initial two-dimensional sequence features to obtain corresponding to-be-processed two-dimensional sequence features;

[0020] Each of the to-be-processed two-dimensional sequence features is input into the first branch, so as to use the second convolutional layer, the first preset activation function, and the preset visual state space model in the first branch to separately process each of the to-be-processed two-dimensional sequence features to obtain corresponding first processing results;

[0021] Each of the to-be-processed two-dimensional sequence features is input into the second branch, so as to use the second preset activation function in the second branch to separately process each of the to-be-processed two-dimensional sequence features to obtain corresponding second processing results;

[0022] The bottleneck residual connection in the preset non-local reconstruction feature extraction component is used to fuse the first processing result, the second processing result, and the initial two-dimensional sequence feature to obtain a corresponding fusion result, and an inverse conversion operation is separately performed on each of the fusion results to obtain a first reconstructed spectral cluster set.

[0023] Optionally, the step of inputting each of the to-be-processed two-dimensional sequence features into the first branch, so as to use the second convolutional layer, the first preset activation function, and the preset visual state space model in the first branch to separately process each of the to-be-processed two-dimensional sequence features to obtain corresponding first processing results, includes:

[0024] Input each of the two-dimensional sequence features to be processed into the first branch, so as to process each of the two-dimensional sequence features to be processed respectively by using the second convolutional layer and the first preset activation function in the first branch to obtain corresponding features to be fused;

[0025] In the preset visual state space model, scan each of the features to be fused by using a preset four-way scanning strategy, and fuse the scanned results to obtain a corresponding first processing result; the first processing result is non-local features corresponding to each of the spectral clusters to be reconstructed.

[0026] Optionally, the step of using the second local space and spectral feature compensation component and the preset sliding window spectral cluster folding component in the preset spectral cluster folding component to process each first reconstructed spectral cluster in the first reconstructed spectral cluster set to obtain a second reconstructed spectral cluster set includes:

[0027] Use the third convolutional layer and the second local space and spectral feature compensation component in the preset spectral cluster folding component to process each first reconstructed spectral cluster in the first reconstructed spectral cluster set respectively to obtain corresponding compensated spectral clusters;

[0028] Use the preset sliding window spectral cluster folding component to perform folding operations on each of the compensated spectral clusters to obtain a second reconstructed spectral cluster set.

[0029] Optionally, after using the second local space and spectral feature compensation component and the preset sliding window spectral cluster folding component in the preset spectral cluster folding component to process each first reconstructed spectral cluster in the first reconstructed spectral cluster set to obtain a second reconstructed spectral cluster set, it further includes:

[0030] Use a preset reconstruction loss calculation function to calculate the reconstruction loss between the second reconstructed spectral cluster set and the spectral cluster set to be reconstructed, so as to obtain a corresponding reconstruction loss result, and use the reconstruction loss result to adjust the model parameters of the Mamba network model.

[0031] Optionally, the step of determining an anomaly detection result map corresponding to the original hyperspectral image based on the second reconstructed spectral cluster set includes:

[0032] Use a preset reconstruction residual technique and determine a reconstruction residual map corresponding to each of the reconstructed spectral clusters in the second reconstructed spectral cluster set based on the second reconstructed spectral cluster set;

[0033] Fuse each of the reconstructed residual maps using a preset fusion technique to obtain an anomaly detection result map corresponding to the original hyperspectral image.

[0034] In a second aspect, the present application provides a hyperspectral image anomaly detection system based on a Mamba network model, which is applied to a Mamba network model including a preset spectral cluster unfolding component, a preset non-local reconstruction feature extraction component, and a preset spectral cluster folding component. The system includes:

[0035] A first spectral cluster set determination module, configured to sequentially process an original hyperspectral image using a preset sliding window spectral cluster unfolding component and a first local spatial and spectral feature compensation component in the preset spectral cluster unfolding component to obtain a set of spectral clusters to be reconstructed corresponding to the original hyperspectral image;

[0036] A second spectral cluster set determination module, configured to perform non-local feature extraction and background reconstruction feature learning on each of the spectral clusters to be reconstructed in the set of spectral clusters to be reconstructed using the preset non-local reconstruction feature extraction component to obtain a first set of reconstructed spectral clusters;

[0037] A third spectral cluster set determination module, configured to sequentially process each first reconstructed spectral cluster in the first set of reconstructed spectral clusters using a second local spatial and spectral feature compensation component and a preset sliding window spectral cluster folding component in the preset spectral cluster folding component to obtain a second set of reconstructed spectral clusters;

[0038] Anomaly detection result map determination module, configured to determine an anomaly detection result map corresponding to the original hyperspectral image based on the second set of reconstructed spectral clusters.

[0039] Optionally, the first spectral cluster set determination module includes:

[0040] A spectral cluster unfolding unit, configured to perform spectral cluster unfolding on the original hyperspectral image using a preset sliding window spectral cluster unfolding component in the preset spectral cluster unfolding component to obtain a plurality of spectral clusters to be processed corresponding to the original hyperspectral image; the sizes of the spectral clusters to be processed are the same;

[0041] A feature extraction unit, configured to perform feature extraction on each of the spectral clusters to be processed using a plurality of depthwise separable convolution branches in the first local spatial and spectral feature compensation component to obtain corresponding first local spatial and spectral neighborhood features; the convolution kernel sizes corresponding to the depthwise separable convolution branches are all different;

[0042] The first splicing feature acquisition unit is configured to splice and convolve each of the first local spatial and spectral neighborhood features by using a preset splicing and convolution rule, so as to obtain first splicing features respectively corresponding to each of the first local spatial and spectral neighborhood features;

[0043] The second splicing feature acquisition unit is configured to extract corresponding second local spatial and spectral neighborhood features from each of the first splicing features again by using a plurality of depthwise separable convolution branches in the first local spatial and spectral feature compensation component, and splice and convolve each of the second local spatial and spectral neighborhood features by using a preset splicing and convolution rule, so as to obtain second splicing features respectively corresponding to each of the second local spatial and spectral neighborhood features;

[0044] The spectral cluster and feature fusion unit is configured to fuse each of the to-be-processed spectral clusters with the corresponding second splicing feature by using a preset residual connection, so as to obtain a plurality of to-be-reconstructed spectral clusters and a to-be-reconstructed spectral cluster set including each of the to-be-reconstructed spectral clusters.

[0045] Optionally, the second spectral cluster set determination module includes:

[0046] The spectral cluster conversion unit is configured to perform a conversion operation on each of the to-be-reconstructed spectral clusters in the to-be-reconstructed spectral cluster set in the preset non-local reconstruction feature extraction component, so as to obtain corresponding initial two-dimensional sequence features;

[0047] The sequence feature determination unit is configured to sequentially process each of the initial two-dimensional sequence features by using a first convolutional layer, a root mean square normalization layer, and a linear transformation layer in the preset non-local reconstruction feature extraction component to obtain corresponding to-be-processed two-dimensional sequence features;

[0048] The first processing result acquisition unit is configured to input each of the to-be-processed two-dimensional sequence features into a first branch, so as to process each of the to-be-processed two-dimensional sequence features respectively by using a second convolutional layer, a first preset activation function, and a preset visual state space model in the first branch to obtain corresponding first processing results;

[0049] The second processing result acquisition unit is configured to input each of the to-be-processed two-dimensional sequence features into a second branch, so as to process each of the to-be-processed two-dimensional sequence features respectively by using a second preset activation function to obtain corresponding second processing results;

[0050] The result inverse conversion unit is configured to fuse the first processing result, the second processing result, and the initial two-dimensional sequence feature by using a bottleneck residual connection in the preset non-local reconstruction feature extraction component to obtain a corresponding fusion result, and perform an inverse conversion operation on each of the fusion results respectively to obtain a first reconstructed spectral cluster set.

[0051] As can be seen from the above, before performing hyperspectral image anomaly detection based on the Mamba network model in this application, it is necessary to use the preset sliding window spectral cluster expansion component and the first local spatial and spectral feature compensation component in the preset spectral cluster expansion component to process the original hyperspectral image in sequence, so as to obtain a set of to-be-reconstructed spectral clusters corresponding to the original hyperspectral image; use the preset non-local reconstruction feature extraction component to perform non-local feature extraction and background reconstruction feature learning on each of the to-be-reconstructed spectral clusters in the set of to-be-reconstructed spectral clusters respectively, so as to obtain a first set of reconstructed spectral clusters; use the second local spatial and spectral feature compensation component and the preset sliding window spectral cluster folding component in the preset spectral cluster folding component to process each first reconstructed spectral cluster in the first set of reconstructed spectral clusters in sequence, so as to obtain a second set of reconstructed spectral clusters; determine an anomaly detection result map corresponding to the original hyperspectral image based on the second set of reconstructed spectral clusters.

[0052] Thus, in this application, by using the preset sliding window spectral cluster expansion component and the first local spatial and spectral feature compensation component to process the original hyperspectral image in sequence, a set of to-be-reconstructed spectral clusters is obtained; then, the preset non-local reconstruction feature extraction component is used to perform non-local feature extraction and background reconstruction feature learning on each of the to-be-reconstructed spectral clusters in the set of to-be-reconstructed spectral clusters respectively, so as to obtain a first set of reconstructed spectral clusters; subsequently, the second local spatial and spectral feature compensation component and the preset sliding window spectral cluster folding component in the preset spectral cluster folding component are used to process each first reconstructed spectral cluster in the first set of reconstructed spectral clusters in sequence, so as to obtain a second set of reconstructed spectral clusters; finally, an anomaly detection result map corresponding to the original hyperspectral image is determined based on the second set of reconstructed spectral clusters. In this way, during the hyperspectral image anomaly detection process based on the Mamba network model, the problem of detection performance degradation caused by limited non-local feature extraction and abnormal target contamination in the background reconstruction learning mode is prevented, and the user experience is improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are only the embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained according to the provided drawings without creative efforts.

[0054] Figure 1 It is a flowchart of a hyperspectral image anomaly detection method based on the Mamba network model disclosed in this application;

[0055] Figure 2A specific false color map and distribution truth map of hyperspectral images disclosed in this application;

[0056] Figure 3 A schematic diagram of the overall framework of the Mamba network model disclosed in this application;

[0057] Figure 4 A schematic diagram of the network structure of a first local spatial and spectral feature compensation component disclosed in this application;

[0058] Figure 5 A schematic diagram of the network structure of a preset visual state space model disclosed in this application;

[0059] Figure 6 A schematic diagram of the structure of a visual 2D selective state space model disclosed in this application;

[0060] Figure 7 Schematic diagrams of the anomaly detection results corresponding to the respective anomaly detection methods disclosed in this application;

[0061] Figure 8 A schematic diagram of the structure of a hyperspectral image anomaly detection system based on the Mamba network model disclosed in this application. Detailed implementation manners

[0062] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0063] Currently, there are various hyperspectral anomaly detection algorithms on the market, such as anomaly detection methods based on data statistical distribution, collaborative and sparse representation models, low-rank sparse decomposition, and deep neural network models. However, model-based methods require model parameter tuning and optimization according to the background and target characteristics of hyperspectral images, and rely too much on handcrafted features, resulting in limited performance of anomaly detection. In addition, since the hyperspectral image anomaly detection method based on the Mamba network model in the deep neural network model realizes anomaly detection by performing deep feature learning and differential feature extraction on spectral sample data, most of these methods have problems of limited non-local feature extraction and high computational complexity, which limit the network detection performance of anomaly detection in the reconstruction learning mode. Therefore, this application provides a hyperspectral image anomaly detection method based on the Mamba network model, which can prevent the problem of degraded detection performance caused by limited non-local feature extraction and abnormal target contamination in the background reconstruction learning mode during the hyperspectral image anomaly detection based on the Mamba network model.

[0064] See Figure 1 As shown, an embodiment of the present invention discloses a hyperspectral image anomaly detection method based on the Mamba network model, which is applied to the Mamba network model including a preset spectral cluster expansion component, a preset non-local reconstruction feature extraction component, and a preset spectral cluster folding component. The method includes:

[0065] Step S11: Use the preset sliding window spectral cluster expansion component and the first local spatial and spectral feature compensation component in the preset spectral cluster expansion component to process the original hyperspectral image in sequence, and obtain a set of spectral clusters to be reconstructed corresponding to the original hyperspectral image.

[0066] In this embodiment, before performing hyperspectral image anomaly detection based on the Mamba network model, a preset sensor is required to collect the hyperspectral image to be detected for anomaly. In a specific implementation manner, the preset sensor is the Reflective Optical System Imaging Spectrometer-03 (ROSIS-03) sensor, and the collected hyperspectral image is an image of a beach scene in Pavia, Italy. Among them, the spatial size of the image is 150×150 pixels, the number of bands is 188, and in addition, the above image contains seven vehicle anomaly targets. As Figure 2 shown, Figure 2 (a) in is the false color map corresponding to the above image, Figure 2 (b) in is the distribution truth map of the corresponding anomaly targets in the image.

[0067] It is worth mentioning that when performing anomaly detection on hyperspectral images in the embodiments of the present application, the Mamba (i.e., a selective structure state space model) network model is used to perform anomaly detection on hyperspectral images. Among them, the overall framework schematic diagram of the Mamba network model is as Figure 3 shown:

[0068] The Mamba network model includes a preset spectral cluster unfolding component, a preset non-local reconstruction feature extraction component, and a preset spectral cluster folding component. Among them, the preset spectral cluster unfolding component includes a preset sliding window spectral cluster unfolding component, a first local spatial and spectral feature compensation component, and a convolution fusion module; the preset non-local reconstruction feature extraction component includes a preset visual state space model; the preset spectral cluster folding component includes a convolution fusion module, a second local spatial and spectral feature compensation component, and a preset sliding window spectral cluster folding component.

[0069] In this embodiment, the original hyperspectral image H needs to be input into the preset spectral cluster unfolding component to use the preset spectral cluster unfolding component to process the original hyperspectral image to obtain a set of spectral clusters to be reconstructed , where is the spectral cluster corresponding to the original hyperspectral image, N is the number of spectral clusters. In addition, the preset spectral cluster unfolding component is composed of a preset sliding window spectral cluster unfolding component, a first local spatial and spectral feature compensation component, and a convolution fusion module connected in series. Specifically, using the preset sliding window spectral cluster unfolding component and the first local spatial and spectral feature compensation component in the preset spectral cluster unfolding component to process the original hyperspectral image in sequence to obtain a set of spectral clusters to be reconstructed corresponding to the original hyperspectral image may include: using the preset sliding window spectral cluster unfolding component in the preset spectral cluster unfolding component to perform spectral cluster unfolding on the original hyperspectral image to obtain a number of spectral clusters to be processed corresponding to the original hyperspectral image; the sizes of the spectral clusters to be processed are the same; in the first local spatial and spectral feature compensation component, using a number of depthwise separable convolution branches to perform feature extraction on each spectral cluster to be processed to obtain corresponding first local spatial and spectral neighborhood features; the convolution kernel sizes corresponding to the depthwise separable convolution branches are all different.

[0070] Further, after obtaining the first local spatial and spectral neighborhood features corresponding to each spectral cluster to be processed, the embodiments of the present application first need to splice and convolve each first local spatial and spectral neighborhood feature using a preset splicing and convolution rule to obtain a first splicing feature corresponding to each first local spatial and spectral neighborhood feature. Subsequently, in the first local spatial and spectral feature compensation component, several depthwise separable convolution branches are used again to extract corresponding second local spatial and spectral neighborhood features from each first splicing feature, and the preset splicing and convolution rule is used to splice and convolve each second local spatial and spectral neighborhood feature to obtain a second splicing feature corresponding to each second local spatial and spectral neighborhood feature. Finally, the preset residual connection is used to fuse each spectral cluster to be processed with the corresponding second splicing feature to obtain several spectral clusters to be reconstructed and a set of spectral clusters to be reconstructed including each spectral cluster to be reconstructed.

[0071] In a specific implementation, the preset sliding window spectral cluster unfolding component divides the original hyperspectral image H into N partially overlapping spectral cluster sets in the order from left to right and from top to bottom through a preset sliding window , where the length, width, and sliding step of the preset sliding window are set to 16 and 4 respectively, and the size of each spectral cluster is .

[0072] Further, for each spectral cluster, the embodiments of the present application need to input each spectral cluster into the first local spatial and spectral feature compensation component to extract multi-scale local spatial-spectral neighborhood features using multi-scale depthwise separable convolution, thus making up for the shortcoming of the spatial local feature forgetting in the linear scanning mechanism of the Mamba network model. The network structure of the first local spatial and spectral feature compensation component is as Figure 4 shown: For the spectral cluster input to the first local spatial and spectral feature compensation component, the embodiments of the present application will input it into 3 parallel depthwise separable convolution branches respectively. Among them, the convolution kernel sizes of the three branches are 3×3, 5×5, and 7×7 respectively, and then the following local multi-scale feature maps are obtained through the GELU activation function:

[0073] ;

[0074] Among them, represents the GELU activation function, , and represent depthwise separable convolutions with convolution kernel sizes of , and respectively. Then, through channel splicing and The convolution fuses and interacts the multi-scale local features obtained, and then uses depthwise separable convolution again to extract the features that have been fused and interacted to compensate for the spatial-spectral information of the target local neighborhood. Finally, the residual connection is used to fuse with the original input spectral cluster to obtain the final output result as follows:

[0075] ;

[0076] wherein, is the channel concatenation operation, is a 2D convolution with a convolution kernel size of 1×1, is the element-wise addition operation. Subsequently, for each spectral cluster after multi-scale spatial-spectral feature compensation, the embodiment of the present application will perform channel integration through a 2D convolution of 1×1 to uniformly adjust the number of channels to the number of spectral bands.

[0077] Step S12: Use the preset non-local reconstruction feature extraction component to separately perform non-local feature extraction and background reconstruction feature learning on each of the to-be-reconstructed spectral clusters in the to-be-reconstructed spectral cluster set, so as to obtain the first set of reconstructed spectral clusters.

[0078] In this embodiment, each spectral cluster after multi-scale spatial-spectral feature compensation needs to be sent into a preset non-local reconstruction feature extraction component composed of a preset visual state space model for background non-local reconstruction feature learning to obtain the first set of reconstructed spectral clusters . Specifically, the preset non-local reconstruction feature extraction component is used to perform non-local feature extraction and background reconstruction feature learning on each to-be-reconstructed spectral cluster in the to-be-reconstructed spectral cluster set, so as to obtain the first reconstructed spectral cluster set, which may include: performing a conversion operation on each to-be-reconstructed spectral cluster in the to-be-reconstructed spectral cluster set in the preset non-local reconstruction feature extraction component to obtain a corresponding initial two-dimensional sequence feature; using the first convolutional layer, root mean square normalization layer, and linear transformation layer in the preset non-local reconstruction feature extraction component to process each initial two-dimensional sequence feature in sequence to obtain a corresponding to-be-processed two-dimensional sequence feature; inputting each to-be-processed two-dimensional sequence feature into the first branch, so as to use the second convolutional layer, first preset activation function, and preset visual state space model in the first branch to process each to-be-processed two-dimensional sequence feature respectively to obtain a corresponding first processing result; inputting each to-be-processed two-dimensional sequence feature into the second branch, so as to use the second preset activation function in the second branch to process each to-be-processed two-dimensional sequence feature respectively to obtain a corresponding second processing result; using the bottleneck residual connection in the preset non-local reconstruction feature extraction component to fuse the first processing result, the second processing result, and the initial two-dimensional sequence feature to obtain a corresponding fusion result, and performing an inverse conversion operation on each fusion result respectively to obtain the first reconstructed spectral cluster set.

[0079] Further, the network structure of the preset visual state space model is as Figure 5 shown: For the first reconstructed spectral cluster set input into the preset visual state space model, in the embodiment of the present application, the first reconstructed spectral cluster set is first reshaped into a two-dimensional sequence feature with a size of , where , , . Then, the two-dimensional sequence feature is passed through a 1D convolution to reduce its number of channels to , and a two-dimensional sequence feature with a size of is obtained through the root mean square normalization layer and the linear transformation layer.

[0080] Subsequently, in the embodiment of the present application, the two-dimensional sequence feature with a size of needs to be input into 2 parallel branches and through a channel Chunk operation. In a specific implementation manner, the output of branch can be expressed as ;

[0081] where and represent a 1D convolutional layer and a visual two-dimensional selective state space model respectively, and represents the SiLU activation function.

[0082] Further, another branch consists of a single SiLU activation function, that is . It is worth mentioning that in the embodiment of the present application, two branches and two-dimensional sequence features need to be fused through bottleneck residual connection , and the output result is feature sequence of, and finally the two-dimensional sequence features are output as a three-dimensional feature map with a size of through a reshaping operation, that is, the first reconstructed spectral cluster set .

[0083] Specifically, input each two-dimensional sequence feature to be processed into the first branch, so as to process each two-dimensional sequence feature to be processed respectively by using the second convolutional layer, the first preset activation function and the preset visual state space model in the first branch, and the corresponding first processing result can be obtained, including: input each two-dimensional sequence feature to be processed into the first branch, so as to process each two-dimensional sequence feature to be processed respectively by using the second convolutional layer and the first preset activation function in the first branch to obtain the corresponding feature to be fused; scan each feature to be fused by using the preset four-way scanning strategy in the preset visual state space model, and fuse the scanned results to obtain the corresponding first processing result; the first processing result is the non-local feature corresponding to each spectral cluster to be reconstructed.

[0084] It is worth mentioning that The structural schematic diagram of is as shown in Figure 6 : The core mechanism of the network is to adopt a preset four-way scanning strategy on the basis of linear scanning of the Mamba network model, that is, scan from the four corners of the feature map to the relative positions and then fuse: from top left to bottom right, from bottom right to top left, from bottom left to top right, from top right to bottom left, so as to ensure that each element in the feature map integrates information from all other positions in different directions, thereby generating a global receptive field, and without increasing the linear computational complexity, realizing the extraction of non-local features.

[0085] Step S13, use the second local spatial and spectral feature compensation component and the preset sliding window spectral cluster folding component in the preset spectral cluster folding component to process each first reconstructed spectral cluster in the first reconstructed spectral cluster set in turn to obtain a second reconstructed spectral cluster set.

[0086] In this embodiment, the first reconstructed spectral cluster set is uniformly input into the preset spectral cluster folding component to obtain the second reconstructed spectral cluster set after background reconstruction learning , that is, fusing the spectral cluster set to obtain a hyperspectral image of the original size. Among them, the preset spectral cluster folding component is composed of a serially connected convolutional fusion module, a second local spatial and spectral feature compensation component, and a preset sliding window spectral cluster folding component. It is worth mentioning that the structures of the second local spatial and spectral feature compensation component and the convolutional fusion module are similar to those of the first local spatial and spectral feature compensation component, that is, the operations performed in the preset sliding window spectral cluster folding component correspond one by one to the operations performed in the preset spectral cluster unfolding component. Specifically, using the second local spatial and spectral feature compensation component and the preset sliding window spectral cluster folding component in the preset spectral cluster folding component to process each first reconstructed spectral cluster in the first reconstructed spectral cluster set in sequence to obtain a second reconstructed spectral cluster set may include: using the third convolutional layer and the second local spatial and spectral feature compensation component in the preset spectral cluster folding component to process each first reconstructed spectral cluster in the first reconstructed spectral cluster set respectively to obtain corresponding compensated spectral clusters; using the preset sliding window spectral cluster folding component to perform folding operations on each compensated spectral cluster respectively to obtain a second reconstructed spectral cluster set.

[0087] In this embodiment, it is necessary to the second reconstructed spectral cluster set and the spectral cluster set of the original hyperspectrum

[0088] In a specific implementation manner, the method used for reconstruction loss calculation is to use function, and the expression is as follows:

[0089] ;

[0090] Among them, is the loss calculation result, represents a custom function about , represents the learnable parameters of the network model, and N represents the number of spectral clusters.

[0091] Step S14: Determine an anomaly detection result map corresponding to the original hyperspectral image based on the second reconstructed spectral cluster set.

[0092] In this embodiment, after obtaining the network model after multiple iterative trainings, the embodiment of the present application needs to save the network model after multiple iterative trainings, and input the original hyperspectral image into the trained network model again to obtain the second reconstructed spectral cluster set , and use a preset reconstruction residual technique to obtain a reconstruction residual map corresponding to each spectral cluster:

[0093] ;

[0094] Among them, and respectively represent and the spectral vectors at the coordinate pixel . Specifically, determining an anomaly detection result map corresponding to the original hyperspectral image based on the second reconstructed spectral cluster set may include: using a preset reconstruction residual technique and based on the second reconstructed spectral cluster set, determining a reconstruction residual map corresponding to each reconstructed spectral cluster in the second reconstructed spectral cluster set. Finally, use a preset fusion technique to fuse each reconstruction residual map to obtain an anomaly detection result map corresponding to the original hyperspectral image.

[0095] It is worth mentioning that Figure 7 (a), (b), (c), and (d) in Figure 7 are the anomaly detection result maps obtained by performing anomaly detection on the Pavia hyperspectral image using the PCA-TLRSR anomaly detection method, the KIFD anomaly detection method, the DFAN anomaly detection method based on a deep neural network, and the anomaly detection method proposed in the embodiment of the present application, respectively. Among them, from

[0096] As can be seen from the above, before performing hyperspectral image anomaly detection based on the Mamba network model in the embodiments of the present application, it is necessary to use the preset sliding window spectral cluster expansion component and the first local spatial and spectral feature compensation component in the preset spectral cluster expansion component to process the original hyperspectral image in sequence to obtain a set of to-be-reconstructed spectral clusters corresponding to the original hyperspectral image; use the preset non-local reconstruction feature extraction component to perform non-local feature extraction and background reconstruction feature learning on each to-be-reconstructed spectral cluster in the set of to-be-reconstructed spectral clusters respectively to obtain a first set of reconstructed spectral clusters; use the second local spatial and spectral feature compensation component and the preset sliding window spectral cluster folding component in the preset spectral cluster folding component to process each first reconstructed spectral cluster in the first set of reconstructed spectral clusters in sequence to obtain a second set of reconstructed spectral clusters; determine an anomaly detection result map corresponding to the original hyperspectral image based on the second set of reconstructed spectral clusters. In this way, during the hyperspectral image anomaly detection process based on the Mamba network model, the problem of degradation of detection performance caused by limited non-local feature extraction and abnormal target contamination in the background reconstruction learning mode can be prevented.

[0097] Correspondingly, as shown in Figure 8 the present application also provides a hyperspectral image anomaly detection system based on the Mamba network model, which is applied to the Mamba network model including a preset spectral cluster expansion component, a preset non-local reconstruction feature extraction component, and a preset spectral cluster folding component. The system includes:

[0098] A first spectral cluster set determination module 11, configured to use the preset sliding window spectral cluster expansion component and the first local spatial and spectral feature compensation component in the preset spectral cluster expansion component to process the original hyperspectral image in sequence to obtain a set of to-be-reconstructed spectral clusters corresponding to the original hyperspectral image;

[0099] A second spectral cluster set determination module 12, configured to use the preset non-local reconstruction feature extraction component to perform non-local feature extraction and background reconstruction feature learning on each to-be-reconstructed spectral cluster in the set of to-be-reconstructed spectral clusters respectively to obtain a first set of reconstructed spectral clusters;

[0100] A third spectral cluster set determination module 13, configured to use the second local spatial and spectral feature compensation component and the preset sliding window spectral cluster folding component in the preset spectral cluster folding component to process each first reconstructed spectral cluster in the first set of reconstructed spectral clusters in sequence to obtain a second set of reconstructed spectral clusters;

[0101] A detection result map determination module 14, configured to determine an anomaly detection result map corresponding to the original hyperspectral image based on the second set of reconstructed spectral clusters.

[0102] As can be seen from the above, before performing hyperspectral image anomaly detection based on the Mamba network model in the embodiments of the present application, it is necessary to process the original hyperspectral image by using a preset sliding window spectral cluster expansion component and a first local spatial and spectral feature compensation component in sequence to obtain a set of spectral clusters to be reconstructed; then. The preset non-local reconstruction feature extraction component is used to perform non-local feature extraction and background reconstruction feature learning on each spectral cluster to be reconstructed in the set of spectral clusters to be reconstructed, so as to obtain a first set of reconstructed spectral clusters; subsequently. The second local spatial and spectral feature compensation component in the preset spectral cluster folding component and the preset sliding window spectral cluster folding component are used to process each first reconstructed spectral cluster in the first set of reconstructed spectral clusters in sequence to obtain a second set of reconstructed spectral clusters; finally. An anomaly detection result map corresponding to the original hyperspectral image is determined based on the second set of reconstructed spectral clusters. In this way, during the hyperspectral image anomaly detection based on the Mamba network model, the problem of degradation of detection performance caused by limited non-local feature extraction and abnormal target contamination in the background reconstruction learning mode is prevented.

[0103] In some specific embodiments, the first spectral cluster set determination module 11 may specifically include:

[0104] A spectral cluster expansion unit, configured to perform spectral cluster expansion on the original hyperspectral image by using a preset sliding window spectral cluster expansion component in the preset spectral cluster expansion component, so as to obtain a plurality of spectral clusters to be processed corresponding to the original hyperspectral image; the sizes of the spectral clusters to be processed are the same;

[0105] A feature extraction unit, configured to use a plurality of depthwise separable convolution branches in the first local spatial and spectral feature compensation component to perform feature extraction on each spectral cluster to be processed respectively, so as to obtain corresponding first local spatial and spectral neighborhood features; the convolution kernel sizes corresponding to the depthwise separable convolution branches are all different;

[0106] A first splicing feature acquisition unit, configured to splice and convolve each first local spatial and spectral neighborhood feature by using a preset splicing and convolution rule, so as to obtain a first splicing feature corresponding to each first local spatial and spectral neighborhood feature respectively;

[0107] A second splicing feature acquisition unit, configured to use a plurality of depthwise separable convolution branches in the first local spatial and spectral feature compensation component to extract corresponding second local spatial and spectral neighborhood features from each first splicing feature respectively, and splice and convolve each second local spatial and spectral neighborhood feature by using a preset splicing and convolution rule, so as to obtain a second splicing feature corresponding to each second local spatial and spectral neighborhood feature respectively;

[0108] A spectral cluster and feature fusion unit is used to fuse each of the to-be-processed spectral clusters with the corresponding second spliced feature by using a preset residual connection to obtain a number of to-be-reconstructed spectral clusters and a set of to-be-reconstructed spectral clusters including each of the to-be-reconstructed spectral clusters.

[0109] In some specific embodiments, the second spectral cluster set determination module 12 may specifically include:

[0110] A spectral cluster conversion unit is used to perform a conversion operation on each of the to-be-reconstructed spectral clusters in the to-be-reconstructed spectral cluster set in the preset non-local reconstruction feature extraction component to obtain corresponding initial two-dimensional sequence features;

[0111] A sequence feature determination unit is used to sequentially process each of the initial two-dimensional sequence features by using a first convolutional layer, a root mean square normalization layer, and a linear transformation layer in the preset non-local reconstruction feature extraction component to obtain corresponding to-be-processed two-dimensional sequence features;

[0112] A first processing result acquisition unit is used to input each of the to-be-processed two-dimensional sequence features into a first branch so as to process each of the to-be-processed two-dimensional sequence features by using a second convolutional layer, a first preset activation function, and a preset visual state space model in the first branch to obtain corresponding first processing results;

[0113] A second processing result acquisition unit is used to input each of the to-be-processed two-dimensional sequence features into a second branch so as to process each of the to-be-processed two-dimensional sequence features by using a second preset activation function in the second branch to obtain corresponding second processing results;

[0114] A result inverse conversion unit is used to fuse the first processing result, the second processing result, and the initial two-dimensional sequence features by using a bottleneck residual connection in the preset non-local reconstruction feature extraction component to obtain corresponding fusion results, and perform an inverse conversion operation on each of the fusion results to obtain a first set of reconstructed spectral clusters.

[0115] Correspondingly, in some specific embodiments, the second spectral cluster set determination module 12 may specifically include:

[0116] A to-be-fused feature determination unit is used to input each of the to-be-processed two-dimensional sequence features into a first branch so as to process each of the to-be-processed two-dimensional sequence features by using a second convolutional layer and a first preset activation function in the first branch to obtain corresponding to-be-fused features;

[0117] A feature scanning unit, configured to scan each of the to-be-fused features in the preset visual state space model by using a preset four-way scanning strategy, and fuse the scanning results to obtain a corresponding first processing result; the first processing result is non-local features corresponding to each of the to-be-reconstructed spectral clusters.

[0118] In some specific embodiments, the third spectral cluster set determination module 13 may specifically include:

[0119] A spectral cluster processing unit, configured to sequentially process each of the first reconstructed spectral clusters in the first reconstructed spectral cluster set by using a third convolutional layer in the preset spectral cluster folding component and a second local spatial and spectral feature compensation component to obtain corresponding compensated spectral clusters;

[0120] A spectral cluster folding unit, configured to perform a folding operation on each of the compensated spectral clusters by using a preset sliding window spectral cluster folding component to obtain a second reconstructed spectral cluster set.

[0121] In some specific embodiments, the hyperspectral image anomaly detection system based on the Mamba network model may further include:

[0122] A model parameter adjustment unit, configured to calculate a reconstruction loss between the second reconstructed spectral cluster set and the to-be-reconstructed spectral cluster set by using a preset reconstruction loss calculation function to obtain a corresponding reconstruction loss result, and use the reconstruction loss result to adjust the model parameters of the Mamba network model.

[0123] In some specific embodiments, the detection result map determination module 14 may specifically include:

[0124] A residual map determination unit, configured to determine a reconstruction residual map corresponding to each of the reconstructed spectral clusters in the second reconstructed spectral cluster set based on the second reconstructed spectral cluster set by using a preset reconstruction residual technique;

[0125] A residual map fusion unit, configured to fuse each of the reconstruction residual maps by using a preset fusion technique to obtain an anomaly detection result map corresponding to the original hyperspectral image.

[0126] Furthermore, the various embodiments in this specification are described in a progressive manner. Each embodiment focuses on the differences from other embodiments. The same or similar parts among the various embodiments can be referred to each other. For the systems disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple, and the relevant parts can be referred to the description of the method part.

[0127] Those skilled in the art may further realize that the units and algorithm steps of each example described in combination with the embodiments disclosed herein can be implemented by electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described according to functions in the above description. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. Those skilled in the art 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 this application.

[0128] The steps of the methods or algorithms described in combination with the embodiments disclosed herein can be directly implemented by hardware, software modules executed by a processor, or a combination of both. The software modules can be placed in a random access memory (RAM), internal memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, removable disk, CD-ROM, or any other form of storage medium well-known in the technical field.

[0129] Finally, it should also be noted that in this document, relational terms such as "first" and "second" are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device comprising a series of elements not only includes those elements, but also includes other elements not expressly listed, or elements inherent to such process, method, article or device. Without further limitation, an element defined by the statement "comprising a..." does not exclude the existence of additional identical elements in the process, method, article or device comprising the element.

[0130] The technical solutions provided in this application have been introduced in detail above. Specific examples are used herein to illustrate the principles and implementation manners of this application. The description of the above embodiments is only used to help understand the method and its core idea of this application; at the same time, for those of ordinary skill in the art, according to the idea of this application, there will be changes in the specific implementation manners and application scopes. In summary, the content of this specification should not be construed as a limitation to this application.

Claims

1. A hyperspectral image anomaly detection method based on the Mamba network model, characterized in that, Applied to a Mamba network model including a preset spectral cluster unfolding component, a preset non-local reconstruction feature extraction component, and a preset spectral cluster folding component, the method includes: Using the preset sliding window spectral cluster unfolding component in the preset spectral cluster unfolding component and the first local spatial and spectral feature compensation component to process the original hyperspectral image in sequence, to obtain a set of spectral clusters to be reconstructed corresponding to the original hyperspectral image; specifically: using the preset sliding window spectral cluster unfolding component in the preset spectral cluster unfolding component to perform spectral cluster unfolding on the original hyperspectral image, to obtain a number of spectral clusters to be processed with the same size corresponding to the original hyperspectral image; using a number of depthwise separable convolution branches in the first local spatial and spectral feature compensation component to perform feature extraction on each of the spectral clusters to be processed, to obtain first local spatial and spectral neighborhood features; the convolution kernel sizes of each of the depthwise separable convolution branches are different; using a preset splicing and convolution rule to splice and convolve each of the first local spatial and spectral neighborhood features, to obtain a first spliced feature corresponding to each of the first local spatial and spectral neighborhood features; using a number of depthwise separable convolution branches in the first local spatial and spectral feature compensation component again to extract second local spatial and spectral neighborhood features from each of the first spliced features, using the preset splicing and convolution rule to splice and convolve each of the second local spatial and spectral neighborhood features, to obtain a second spliced feature; using a preset residual connection to fuse each of the spectral clusters to be processed with the corresponding second spliced feature, to obtain a number of spectral clusters to be reconstructed and a set of spectral clusters to be reconstructed including each of the spectral clusters to be reconstructed; Using the preset non-local reconstruction feature extraction component to perform non-local feature extraction and background reconstruction feature learning on each of the spectral clusters to be reconstructed in the set of spectral clusters to be reconstructed, to obtain a first set of reconstructed spectral clusters; Using the second local spatial and spectral feature compensation component and the preset sliding window spectral cluster folding component in the preset spectral cluster folding component to process each of the first reconstructed spectral clusters in the first set of reconstructed spectral clusters in sequence, to obtain a second set of reconstructed spectral clusters; specifically: using the third convolutional layer and the second local spatial and spectral feature compensation component in the preset spectral cluster folding component to process each of the first reconstructed spectral clusters in the first set of reconstructed spectral clusters respectively, to obtain compensated spectral clusters; using the preset sliding window spectral cluster folding component to perform a folding operation on each of the compensated spectral clusters, to obtain a second set of reconstructed spectral clusters; Determining an anomaly detection result map corresponding to the original hyperspectral image based on the second set of reconstructed spectral clusters.

2. The hyperspectral image anomaly detection method based on the Mamba network model according to claim 1, wherein The using the preset non-local reconstruction feature extraction component to perform non-local feature extraction and background reconstruction feature learning on each of the spectral clusters to be reconstructed in the set of spectral clusters to be reconstructed, to obtain a first set of reconstructed spectral clusters, includes: In the preset non-local reconstruction feature extraction component, perform a conversion operation on each of the to-be-reconstructed spectral clusters in the to-be-reconstructed spectral cluster set to obtain corresponding initial two-dimensional sequence features; Use the first convolutional layer, root mean square normalization layer, and linear transformation layer in the preset non-local reconstruction feature extraction component to process each of the initial two-dimensional sequence features in sequence to obtain corresponding to-be-processed two-dimensional sequence features; Input each of the to-be-processed two-dimensional sequence features into the first branch, so as to use the second convolutional layer, first preset activation function, and preset visual state space model in the first branch to process each of the to-be-processed two-dimensional sequence features respectively to obtain corresponding first processing results; Input each of the to-be-processed two-dimensional sequence features into the second branch, so as to use the second preset activation function in the second branch to process each of the to-be-processed two-dimensional sequence features respectively to obtain corresponding second processing results; Use the bottleneck residual connection in the preset non-local reconstruction feature extraction component to fuse the first processing result, the second processing result, and the initial two-dimensional sequence features to obtain corresponding fusion results, and perform inverse conversion operations on each of the fusion results respectively to obtain the first reconstructed spectral cluster set.

3. The hyperspectral image anomaly detection method based on the Mamba network model according to claim 2, wherein The step of inputting each of the to-be-processed two-dimensional sequence features into the first branch, so as to use the second convolutional layer, first preset activation function, and preset visual state space model in the first branch to process each of the to-be-processed two-dimensional sequence features respectively to obtain corresponding first processing results, includes: Input each of the to-be-processed two-dimensional sequence features into the first branch, so as to use the second convolutional layer, first preset activation function in the first branch to process each of the to-be-processed two-dimensional sequence features respectively to obtain corresponding to-be-fused features; In the preset visual state space model, use a preset four-way scanning strategy to scan each of the to-be-fused features, and fuse the scanned results to obtain corresponding first processing results; the first processing results are non-local features corresponding to each of the to-be-reconstructed spectral clusters.

4. The hyperspectral image anomaly detection method based on the Mamba network model according to claim 1, characterized in that After using the second local spatial and spectral feature compensation component and the preset sliding window spectral cluster folding component in the preset spectral cluster folding component to process each of the first reconstructed spectral clusters in the first reconstructed spectral cluster set to obtain the second reconstructed spectral cluster set, further includes: Use a preset reconstruction loss calculation function to calculate the reconstruction loss between the second reconstructed spectral cluster set and the to-be-reconstructed spectral cluster set to obtain corresponding reconstruction loss results, so as to use the reconstruction loss results to adjust the model parameters of the Mamba network model.

5. The hyperspectral image anomaly detection method based on the Mamba network model according to any one of claims 1 to 4, characterized in that, The step of determining an anomaly detection result map corresponding to the original hyperspectral image based on the second reconstructed spectral cluster set includes: Use a preset reconstruction residual technique and based on the second reconstructed spectral cluster set to determine reconstruction residual maps corresponding to each of the reconstructed spectral clusters in the second reconstructed spectral cluster set; Fuse each of the reconstructed residual maps using a preset fusion technique to obtain an anomaly detection result map corresponding to the original hyperspectral image.

6. A hyperspectral image anomaly detection system based on the Mamba network model, characterized in that, Applied to a Mamba network model including a preset spectral cluster unfolding component, a preset non-local reconstruction feature extraction component, and a preset spectral cluster folding component, the system includes: A first spectral cluster set determination module, configured to sequentially process an original hyperspectral image using a preset sliding window spectral cluster unfolding component and a first local spatial and spectral feature compensation component in the preset spectral cluster unfolding component to obtain a spectral cluster set to be reconstructed corresponding to the original hyperspectral image; specifically: use the preset sliding window spectral cluster unfolding component in the preset spectral cluster unfolding component to perform spectral cluster unfolding on the original hyperspectral image to obtain a plurality of to-be-processed spectral clusters of the same size corresponding to the original hyperspectral image; in the first local spatial and spectral feature compensation component, use a plurality of depthwise separable convolution branches to perform feature extraction on each of the to-be-processed spectral clusters to obtain first local spatial and spectral neighborhood features; the convolution kernel sizes of each of the depthwise separable convolution branches are different; use a preset splicing and convolution rule to splice and convolve each of the first local spatial and spectral neighborhood features to obtain a first splicing feature corresponding to each of the first local spatial and spectral neighborhood features; in the first local spatial and spectral feature compensation component, use a plurality of depthwise separable convolution branches again to extract second local spatial and spectral neighborhood features from each of the first splicing features, use a preset splicing and convolution rule to splice and convolve each of the second local spatial and spectral neighborhood features to obtain a second splicing feature; use a preset residual connection to fuse each of the to-be-processed spectral clusters with the corresponding second splicing feature to obtain a plurality of to-be-reconstructed spectral clusters and a spectral cluster set to be reconstructed including each of the to-be-reconstructed spectral clusters; A second spectral cluster set determination module, configured to perform non-local feature extraction and background reconstruction feature learning on each of the to-be-reconstructed spectral clusters in the spectral cluster set to be reconstructed using the preset non-local reconstruction feature extraction component to obtain a first reconstructed spectral cluster set; A third spectral cluster set determination module, configured to sequentially process each of the first reconstructed spectral clusters in the first reconstructed spectral cluster set using a second local spatial and spectral feature compensation component and a preset sliding window spectral cluster folding component in the preset spectral cluster folding component to obtain a second reconstructed spectral cluster set; specifically: use a third convolutional layer and a second local spatial and spectral feature compensation component in the preset spectral cluster folding component to sequentially process each of the first reconstructed spectral clusters in the first reconstructed spectral cluster set to obtain compensated spectral clusters; use a preset sliding window spectral cluster folding component to perform a folding operation on each of the compensated spectral clusters to obtain a second reconstructed spectral cluster set; An anomaly detection result map determination module, configured to determine an anomaly detection result map corresponding to the original hyperspectral image based on the second reconstructed spectral cluster set.

7. The hyperspectral image anomaly detection system based on the Mamba network model according to claim 6, wherein The second spectral cluster set determination module includes: A spectral cluster conversion unit, configured to perform a conversion operation on each of the to-be-reconstructed spectral clusters in the to-be-reconstructed spectral cluster set in the preset non-local reconstruction feature extraction component, so as to obtain corresponding initial two-dimensional sequence features; A sequence feature determination unit, configured to sequentially process each of the initial two-dimensional sequence features by using a first convolutional layer, a root mean square normalization layer, and a linear transformation layer in the preset non-local reconstruction feature extraction component to obtain corresponding to-be-processed two-dimensional sequence features; A first processing result acquisition unit, configured to input each of the to-be-processed two-dimensional sequence features into a first branch, so as to process each of the to-be-processed two-dimensional sequence features by using a second convolutional layer, a first preset activation function, and a preset visual state space model in the first branch to obtain corresponding first processing results; A second processing result acquisition unit, configured to input each of the to-be-processed two-dimensional sequence features into a second branch, so as to process each of the to-be-processed two-dimensional sequence features by using a second preset activation function to obtain corresponding second processing results; A result inverse conversion unit, configured to fuse the first processing result, the second processing result, and the initial two-dimensional sequence features by using a bottleneck residual connection in the preset non-local reconstruction feature extraction component to obtain corresponding fusion results, and perform an inverse conversion operation on each of the fusion results to obtain a first set of reconstructed spectral clusters.

Images