Hyperspectral remote sensing image abnormal target detection method based on Fourier condition mask, storage medium and equipment

Abstract

The invention discloses a hyperspectral remote sensing image abnormal target detection method based on Fourier condition masks, a storage medium and equipment, and belongs to the technical field of target detection. In order to solve the problem that background and anomaly cannot be well separated due to the fact that spectral image feature extraction capacity is limited in hyperspectral remote sensing image abnormal target detection, a conditional mask network is used for processing a hyperspectral image X to obtain a conditional mask M, then the hyperspectral image X is spliced with the conditional mask M through channel dimensions, and the hyperspectral remote sensing image abnormal target detection method based on the conditional mask network is obtained. Forming a condition image Xcat; after a condition image Xcat is obtained, W is obtained through an attention weight filter; and multiplying W by Xcat, performing single-layer convolution operation on an obtained result image to obtain a reconstructed image # imgabs0 #, and performing subtraction on the reconstructed image # imgabs1 # and the original hyperspectral image X to obtain a final detection result.

Description

Technical Field

[0001] The present invention belongs to the technical field of target detection, and relates to a hyperspectral anomaly detection method, a storage medium, and a device. Background Art

[0002] Hyperspectral images with hundreds of spectral bands play a key role in remote sensing scenarios. Hyperspectral images contain rich spectral and spatial information, and thus have been widely used in many fields, such as mineral exploration, precision agriculture, and urban monitoring. In these applications, hyperspectral anomaly detection is a technique for detecting targets with spectral differences from the surrounding background without prior information. Currently, anomaly detection methods for hyperspectral images have been developed. However, these anomaly detection methods still have certain limitations in extracting features of hyperspectral images, resulting in the inability to well separate the background and anomalies. Summary of the Invention

[0003] The present invention aims to solve the problem that the ability to extract spectral image features in the anomaly target detection of hyperspectral remote sensing images is limited, resulting in the inability to well separate the background and anomalies.

[0004] A hyperspectral remote sensing image anomaly target detection method based on a Fourier conditional mask, comprising:

[0005] Processing the hyperspectral image X by a conditional mask network to obtain a conditional mask M, and then splicing the hyperspectral image X with the conditional mask M through the channel dimension to form a conditional image X cat ; Obtaining the conditional image X cat After that, obtaining W through an attention weight filter; multiplying W with X cat And performing a single-layer convolution operation on the obtained result image to obtain a reconstructed image

[0006] Performing a difference operation on the reconstructed image And the original hyperspectral image X to obtain a final detection result.

[0007] Further, the conditional mask M = CMASK(X) obtained by processing the hyperspectral image X by a function of the conditional mask network, where CMASK(·) is the conditional mask network; the processing process of the conditional mask network includes the following steps:

[0008] For the hyperspectral image Where H, W, and C respectively represent the height, width, and number of bands of the hyperspectral image, first passing through a convolutional layer that adjusts the number of channels to C 1 And then passing through a normalization layer to obtain an internal network input Then inputting M I Into a Fourier anomaly suppression filter to obtain

[0009] For M FASF , it is processed by a normalization layer and a spatio-spectral multi-layer perceptron module to obtain M SSMLP ; then, the original number of channels is regressed through a convolutional layer to obtain the final mask

[0010] Furthermore, for M FASF , the process of processing it by a normalization layer and a spatio-spectral multi-layer perceptron module to obtain M SSMLP is as follows:

[0011] M SSMLP = MLP(MLP(LN(M FASF ))) T ) T + LN(M FASF ))

[0012] where MLP(·) is the spatio-spectral multi-layer perceptron module; LN(·) is the normalization layer;

[0013] Furthermore, the process of inputting M I into the Fourier anomaly suppression filter to obtain M FASF includes the following steps:

[0014] For the input First, it is transformed by the fast Fourier transform to obtain M F = F(M I ); F(·) is the fast Fourier transform function; the features at different spatial positions in M F correspond to different frequency components of M I ;

[0015] Then, M F is subjected to coarse-level and fine-level anomaly suppression operations using two different branches:

[0016] In one branch, a low-pass filter LPF is used for filtering;

[0017] In the other branch, M F is split into a real part M real and an imaginary part M imag ; different processes are performed on the real part and the imaginary part respectively: the real part is convolved to obtain M' real ; for the imaginary part M imag , the vertical component of the image is smoothed by Gaussian convolution to obtain M' imag ; then, based on the real part processing result M' real and the imaginary part processing result M' imag , a complex feature M fine = M'real +i·M' imag ;

[0018] Finally, multiply the results of the two branches, and then perform an inverse fast Fourier transform to obtain the anomaly suppression result M of the Fourier anomaly suppression filter. FASF .

[0019] Furthermore, the mean squared error MSE is used as the loss function L during the training process of the conditional mask network CMASK(·):

[0020]

[0021] where is the pixel vector at the position (i, j) of the input hyperspectral image, is the pixel vector at the position (i, j) of the background image reconstructed by the network.

[0022] Furthermore, the process of obtaining the final detection result by subtracting the reconstructed image from the original hyperspectral image X is as follows: The reconstruction error detection map R is obtained by calculating the difference between the reconstructed image and the original hyperspectral image.

[0023]

[0024] In the formula, R i,j,: represents the non - negative scalar error at the position (i, j), and X i,j,: are spectral vectors respectively;

[0025] Anomalies will be shown in the reconstruction error map R, thus realizing anomaly detection.

[0026] A computer storage medium stores at least one instruction, and the at least one instruction is loaded and executed by a processor to implement the above - mentioned method for detecting abnormal targets in hyperspectral remote sensing images based on Fourier conditional masking.

[0027] A device for detecting abnormal targets in hyperspectral remote sensing images based on Fourier conditional masking, the device includes a processor and a memory, and the memory stores at least one instruction, and the at least one instruction is loaded and executed by the processor to implement the above - mentioned method for detecting abnormal targets in hyperspectral remote sensing images based on Fourier conditional masking.

[0028] Beneficial effects:

[0029] The present invention optimizes the anomaly detection process and improves the spectral image feature extraction ability by generating a conditional mask to suppress abnormal high-frequency information and retain the low-frequency information of the frequency-domain background. In addition, in order to achieve fine-grained hyperspectral anomaly detection, the present invention also proposes a Fourier anomaly suppression filter, which uses Fourier technology to manage the background and anomalies and improves detection through precise frequency decoupling. In addition, the present invention also designs a conditional mask network to effectively suppress anomalies. The network integrates the Fourier anomaly suppression filter and the spatio-spectral multi-layer perceptron, enhancing the transformation and representation ability of the generated mask and also contributing to anomaly suppression. Finally, by subtracting the reconstruction result from the original image to obtain the anomaly detection result, the present invention can effectively separate the background and anomalies. Through experimental analysis, the method proposed by the present invention can obtain AUC values of 0.9905, 0.9960, 0.9933, 0.9972, and 0.9996 on five datasets respectively. Description of the Drawings

[0030] Figure 1 It is a flowchart of hyperspectral anomaly detection based on Fourier conditional mask;

[0031] Figure 2 It is a false-color image, ground-truth image, and detection result image of Dataset Ⅰ, where (a) is the false-color image, (b) is the ground-truth image, and (c) is the detection image;

[0032] Figure 3 It is a false-color image, ground-truth image, and detection result image of Dataset Ⅱ, where (a) is the false-color image, (b) is the ground-truth image, and (c) is the detection image;

[0033] Figure 4 It is a false-color image, ground-truth image, and detection result image of Dataset Ⅲ, where (a) is the false-color image, (b) is the ground-truth image, and (c) is the detection image;

[0034] Figure 5 It is a false-color image, ground-truth image, and detection result image of Dataset Ⅳ, where (a) is the false-color image, (b) is the ground-truth image, and (c) is the detection image;

[0035] Figure 6 It is a false-color image, ground-truth image, and detection result image of Dataset Ⅴ, where (a) is the false-color image, (b) is the ground-truth image, and (c) is the detection image. Detailed Embodiment

[0036] The present invention proposes a hybrid attention hyperspectral anomaly detection framework FCMMA based on Fourier conditional masking, which combines Fourier transform and deep learning techniques. By using Fourier conditional masking to guide the image to reconstruct more background pixels, the best performance can be achieved during the anomaly detection process. This design significantly improves the accuracy and robustness of hyperspectral image anomaly detection, bringing a novel and efficient solution to the field of hyperspectral image processing. The present invention can better extract the features of hyperspectral images, making it easier to detect anomaly targets. The following will explain the present invention in detail in combination with specific embodiments.

[0037] Specific Embodiment 1: Combined with Figure 1 Explain this embodiment,

[0038] This embodiment is a method for detecting anomaly targets in hyperspectral remote sensing images based on Fourier conditional masking, including the following steps:

[0039] First, use the hybrid attention method based on Fourier conditional masking to guide the reconstruction of hyperspectral images. The specific process includes:

[0040] The hybrid attention method based on Fourier conditional masking can generate a conditional mask to guide the network to suppress abnormal high-frequency information and retain the low-frequency information of the background in the frequency domain, thereby guiding the anomaly detection process to achieve optimal performance. By performing different processes on the real and imaginary parts of the Fourier-transformed image in the frequency domain, fine distinction between abnormal pixels and background pixels can be achieved, while comprehensively and accurately suppressing anomalies and retaining more background information. Subsequently, use the hybrid attention weight discrimination (attention weight filter MADW) block to select the hybrid image conditions for filtering.

[0041] Given a hyperspectral image H, W, and C respectively represent the height, width, and number of bands of the hyperspectral image. For the hyperspectral cube X, reconstruction is performed. First, fuse the anomaly suppression conditional mask with the original image through channel connectivity to form a conditional image. Then, filter the conditional image, and the goal is to reconstruct a pure background image.

[0042] Conditional mask is a completely suppressed anomaly mask image obtained after passing the input X through the conditional mask network. The mask M is used to guide the image to pay more attention to background information during the reconstruction process and suppress the expression of anomalies. M is composed of:

[0043] M = CMASK(X) (1)

[0044] where CMASK(·) is the conditional mask network.

[0045] Then, splice the original hyperspectral image X with the conditional mask M through the channel dimension to form a conditional image X cat is used as a conditional image, which contains two important pieces of information. These are the original hyperspectral information and the mask information for suppressing anomalies. X cat The formula is:

[0046] X cat = concatenate(X, M) (2)

[0047] where concatenate(·) is the concatenation operation.

[0048] After obtaining the conditional image X cat it is passed through an attention weight filter to obtain W is used to filter out anomaly information and retain comprehensive background information. The formula for W is:

[0049] W = MAWD(X cat ) (3)

[0050] where MAWD (Multi - Attention Weighted Detector, MAWD) is the attention weight filter;

[0051] Multiply the output W of the attention weighted filter by X cat to obtain the result image after weighted channel selection. Then, adjust the channel size through a single - layer convolution operation to obtain the reconstructed image which produces better context and highly suppressed anomaly information. The formula is:

[0052]

[0053] where conv(·) is the single - layer convolution operation. ⊙ represents the Hadamard product, also known as element - wise multiplication.

[0054] During the training of CMASK(·), in order to optimize the network, the mean squared error MSE is used as the loss function L:

[0055]

[0056] where is the pixel vector at the position (i, j) of the input hyperspectral image, is the pixel vector at the position (i, j) of the background image reconstructed by the network. This loss is used to train the model. After repeating the training process on the dataset multiple times, the network can fully learn the background features and accurately reconstruct the background.

[0057] The conditional mask network CMASK can generate a conditional mask that sufficiently suppresses anomalies. This network precisely suppresses anomalous pixels in the frequency domain by using frequency-domain decoupling, generating a conditional mask that can sufficiently suppress anomalous pixels and enhance the representation of background pixels. This mask can effectively guide the reconstruction trends of different pixels in the image, enhance the reconstruction representation of background pixels, and simultaneously suppress the reconstruction of anomalous pixels.

[0058] As Figure 1 shown, the CMASK network mainly consists of two parts: a Fourier anomaly suppression filter and a spatio-spectral multi-layer perceptron block. The process of obtaining the mask through the CMASK network has a total of four steps:

[0059] First, X needs to pass through a convolutional layer that adjusts the number of channels to C 1 and then through a normalization layer to obtain the input to the internal network The formula is:

[0060] M I = LN(conv(X)) (6)

[0061] where conv(·) is the convolutional layer and LN(·) is layer normalization.

[0062] Secondly, to suppress the expression of anomalies, M I is input into the Fourier anomaly suppression filter to obtain the output

[0063] M FASF = FASF(M I ) (7)

[0064] where FASF(·) is the Fourier anomaly suppression filter. The Fourier anomaly suppression filter FASF is used to effectively suppress anomalies in the generated conditional mask.

[0065] After that, M FASF is input into the normalization layer and then fed into the spatio-spectral multi-layer perceptron (SSMLP) module to improve the expressiveness of the generated mask and further improve the performance of anomaly detection. The formula is:

[0066] M SSMLP = MLP(MLP(LN(M FASF ))) T + LN(M T )) (8) FASF where

[0067] MLP(·) is the function of the spatio-spectral multi-layer perceptron module.

[0068] Finally, the final mask​ It is obtained by adding another convolutional layer to regress the original number of channels. The formula is as follows:

[0069] M = conv(M SSMLP )) (9)

[0070] The Fourier Anomaly Suppression Filter (FASF) can effectively and accurately process the abnormal components in the frequency domain. This filter uses frequency decoupling to separately process the real and imaginary parts of the image, filtering out all abnormal information while retaining a large amount of background information. The Fourier Anomaly Suppression Filter (FASF) can effectively and accurately suppress the anomalies in the real and imaginary parts respectively using frequency decoupling. The low-frequency part of the image contains the overall structure, contour, and large-scale changes of the image. In contrast, the high-frequency part contains image details, textures, and edge information. Anomalies usually appear as high-frequency components in the frequency domain because anomalies usually manifest as sharp local changes or mutations in the image, and these changes correspond to high-frequency information in the frequency domain.

[0071] Therefore, anomaly suppression is to screen and process the high-frequency part of the image in the frequency domain. The processing process of FASF is as follows:

[0072] Given the input of the Fourier Anomaly Suppression Filter First, M is obtained through the Fast Fourier Transform F = F(M I ). F(·) is the Fast Fourier Transform function, expressed as:

[0073]

[0074] As can be seen from the above formula, the features at different spatial positions in M F correspond to different frequency components of M I , and this process converts M I to M F , where the feature M F is actually a complex feature.

[0075] Next, coarse and fine anomaly suppression operations are performed on M F by two different branches. In one branch, a low-pass filter (LPF) is used to filter out the high-frequency components of the intermediate frequency. This step is to coarsely filter the anomalies in M F . The coarse suppression result of this branch is:

[0076] M coarse = LPF(M F ) (11)

[0077] In the other branch, the complex feature M F is split into the real part M real and the imaginary part Mimag The formula is:

[0078] M F = M real + i·M imag (12)

[0079] In the frequency domain, the real part is usually associated with the low-frequency components of the image, which can provide information about the image structure, such as the smoothness of the image. The imaginary part reflects the frequency-related phase information of the image in the frequency domain and describes the phase characteristics of the image. The imaginary part usually represents the vertical component of the image. Therefore, when extracting depth features while suppressing the expression of abnormal features, different treatments need to be performed on the real part and the imaginary part respectively. Perform a convolution operation on the real part and fuse global features to extract depth background features. The formula is:

[0080] M' real = K * M real (13)

[0081] where K is a common convolution kernel.

[0082] The imaginary part M imag uses a Gaussian convolution to smooth the vertical component of the image and effectively suppress abnormal information. The formula is:

[0083] M' imag = G * M imag (14)

[0084] Here G is a two-dimensional Gaussian convolution kernel.

[0085] A better abnormal suppression result is obtained in this branch. The formula is:

[0086] M fine = M' real + i·M' imag (15)

[0087] Finally, multiply the rough and fine abnormal suppression results of the two branches, and then perform an inverse fast Fourier transform to obtain the abnormal suppression result of the Fourier abnormal suppression filter. The final Fourier abnormal suppression filter (FASF) is calculated as:

[0088] M FASF = F -1 (M coarse ⊙M fine ) (16)

[0089] The present invention obtains the final detection result by subtracting the reconstruction result from the original image.

[0090] After the training phase, in the detection phase, the difference between the reconstructed image and the original image is calculated as the detection result. The detection result is reflected in the form of a reconstruction error detection map R. The principle of obtaining the reconstruction error detection map R is based on the reconstruction ability of the deep learning model for normal data. Normal data can be reconstructed well, while abnormal data will generate large errors during the reconstruction process. These errors are highlighted in the reconstruction error map R, thus realizing anomaly detection. The formula for the reconstruction error detection map R is:

[0091]

[0092] In the formula, R i,j,: represents the non - negative scalar error at position (i, j), and X i,j,: are spectral vectors respectively.

[0093] Example:

[0094] The present invention proposes a hyperspectral anomaly detection method based on Fourier conditional masking for hyperspectral anomaly detection. This method generates a conditional mask to suppress abnormal high - frequency information and retain the low - frequency information of the frequency - domain background, optimizing the anomaly detection process. In addition, in order to achieve fine - grained hyperspectral anomaly detection, a Fourier anomaly suppression filter is proposed. This filter uses Fourier techniques to manage the background and anomalies, improving detection through precise frequency decoupling. Then, a conditional mask network is designed to effectively suppress anomalies. This network integrates the Fourier anomaly suppression filter and the spatio - spectral multi - layer perceptron, enhancing the transformation and representation ability of the generated mask and also contributing to suppressing anomalies. Finally, the anomaly detection result is obtained by taking the difference between the reconstruction result and the original image. Through experimental analysis, the method proposed by the present invention can obtain AUC values of 0.9905, 0.9960, 0.9933, 0.9972, and 0.9996 on five datasets respectively.

[0095] This part illustrates the effect of the hyperspectral anomaly detection method based on Fourier conditional masking proposed by the present invention using five hyperspectral data. The detailed information of the five data used is listed in Table 1. The experimental results use the area under the receiver operating characteristic curve (AUC) as the evaluation index. The higher the value of AUC, the better the detection effect.

[0096] Table 1 Details of the hyperspectral images used

[0097]

[0098] For different data, the AUC values of the method of the present invention are shown in Table 2. The language environment of the method of the present invention is PYTHON, the experimental hardware platform: NVIDIA GeForce GTX 2080Ti gpu, and the memory: 8G.

[0099] AUC values on five groups of experimental data in Table 2

[0100] Specific Embodiment 2:

[0102] This embodiment is a computer storage medium, in which at least one instruction is stored, and the at least one instruction is loaded and executed by a processor to implement the method for detecting abnormal targets in hyperspectral remote sensing images based on Fourier conditional masks.

[0103] It should be understood that the instructions include computer program products, software, or computerized methods corresponding to any method described in the present invention; the instructions can be used to program a computer system or other electronic devices. A computer storage medium may include a readable medium on which instructions are stored, which may include, but are not limited to, magnetic storage media and optical storage media; magneto-optical storage media include read-only memory ROM, random access memory RAM, erasable programmable memory (e.g., EPROM and EEPROM), and flash memory layers, or other types of media suitable for storing electronic instructions. Specific Embodiment 3:

[0105] This embodiment is a device for detecting abnormal targets in hyperspectral remote sensing images based on Fourier conditional masks. The device includes a processor and a memory. It should be understood that it includes any device including a processor and a memory described in the present invention. The device may also include other units and modules for display, interaction, processing, control, etc. through signals or instructions, and other functions.

[0106] At least one instruction is stored in the memory, and the at least one instruction is loaded and executed by the processor to implement the method for detecting abnormal targets in hyperspectral remote sensing images based on Fourier conditional masks.

[0107] Those skilled in the art should understand that the stored at least one instruction is a computer program product corresponding to the method or system. Therefore, the present application can adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk memories, CD-ROMs, optical memories, etc.) containing computer-usable program codes. The solutions in the embodiments of the present application can be implemented in various computer languages, for example, object-oriented programming languages such as Java and interpreted scripting languages such as JavaScript.

[0108] This application is described with reference to the flowcharts and / or block diagrams of methods, systems, and computer program products according to embodiments of the present application, and can also be used for corresponding devices. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, such that the instructions executed by the processor of the computer or other programmable data processing devices generate a device for implementing the functions specified in one process Figure 1 or more processes and / or blocks Figure 1 or a device for implementing the functions specified in one block or more blocks.

[0109] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, such that the instructions stored in the computer-readable memory generate a manufactured article including an instruction device that implements the functions specified in one process Figure 1 or more processes and / or blocks Figure 1 or a device for implementing the functions specified in one block or more blocks.

[0110] These computer program instructions can also be loaded onto a computer or other programmable data processing device, such that a series of operation steps are executed on the computer or other programmable device to generate a computer-implemented process, so that the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in one process Figure 1 or more processes and / or blocks Figure 1 or a device for implementing the functions specified in one block or more blocks.

[0111] Although the preferred embodiments of the present application have been described, those skilled in the art can make additional changes and modifications once they learn the basic creative concepts. Therefore, the appended claims are intended to be construed to include the preferred embodiments as well as all changes and modifications falling within the scope of the present application.

[0112] Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalent technologies, the present application is also intended to include these changes and modifications.

[0113] The above numerical examples of the present invention are only for explaining in detail the calculation model and calculation process of the present invention, rather than limiting the implementation manners of the present invention. For those of ordinary skill in the art, other different forms of changes or modifications can be made on the basis of the above description. It is impossible to enumerate all the implementation manners here. Any obvious changes or modifications derived from the technical solutions of the present invention still fall within the protection scope of the present invention.

Claims

1. A method for abnormal target detection in hyperspectral remote sensing images based on Fourier conditional masking, characterized in that: include: The conditional mask network is used to process the hyperspectral image X to obtain the conditional mask M, and then the hyperspectral image X is concatenated with the conditional mask M through the channel dimension to form the conditional image X cat ; Get the conditional image X cat After that, we get W by paying attention to the weight filter; we add W to X cat Multiply, and then perform a single-layer convolution operation on the resulting image to obtain the reconstructed image. The reconstructed image The final detection result is obtained by subtracting the original hyperspectral image X.

2. The method for detecting abnormal targets in hyperspectral remote sensing images based on Fourier conditional mask according to claim 1, characterized in that: The conditional mask M=CMASK(X) is obtained by processing the hyperspectral image X using the function of the conditional mask network, where CMASK(·) is the conditional mask network; the processing process of the conditional mask network includes the following steps: For hyperspectral images H, W, and C represent the height, width, and number of frequency bands of the hyperspectral image, respectively. The image first passes through a convolutional layer that adjusts the number of channels to C1, and then passes through a normalization layer to obtain the internal network input. Then M I Input Fourier anomaly suppression filter to get For M FASF , and processed using the normalization layer and the spatial-spectral multilayer perceptron module to obtain M SSMLP ; Then a convolutional layer is used to regress the original number of channels to obtain the final mask 3. The method for detecting abnormal targets in hyperspectral remote sensing images based on Fourier conditional mask according to claim 2, characterized in that: For M FASF , and processed using the normalization layer and the spatial-spectral multilayer perceptron module to obtain M SSMLP The process is as follows: M SSMLP =MLP(MLP(LN(M FASF ) T ) T +LN(M FASF )) Among them, MLP(·) is a spatial-spectral multilayer perceptron module; LN(·) is a normalization layer.

4. The method for detecting abnormal targets in hyperspectral remote sensing images based on Fourier conditional mask according to claim 2, characterized in that: M I Input Fourier anomaly suppression filter to get M FASF The process includes the following steps: For input First, we use fast Fourier transform to get M F =F(M I );F(·) is the fast Fourier transform function; M F The features at different spatial positions in M I Different frequency components of Then use two different branches to F Perform coarse and fine level exception suppression operations: In one branch, a low-pass filter LPF is used for filtering; In the other branch, M F Split into real part M real and the imaginary part M imag ; Perform different processing on the real part and the imaginary part: Perform convolution operation on the real part to obtain M' real ; For the imaginary part M imag , use Gaussian convolution to smooth the vertical component of the image to get M' imag ; Then process the result M' based on the real part real And the imaginary part processing result M' imag Get the complex feature M fine =M' real +i·M' imag ; Finally, the results of the two branches are multiplied, and then the inverse fast Fourier transform is performed to obtain the anomaly suppression result M of the Fourier anomaly suppression filter. FASF .

5. The method for abnormal target detection in hyperspectral remote sensing images based on Fourier conditional masking according to any one of claims 2 to 4, characterized in that: The mean square error MSE is used as the loss function L during the training of the conditional mask network CMASK(·): in, is the pixel vector at position (i, j) of the input hyperspectral image, is the pixel vector at position (i, j) of the background image reconstructed by the network.

6. The method for abnormal target detection in hyperspectral remote sensing images based on Fourier conditional mask according to claim 1, characterized in that: The reconstructed image The process of obtaining the final detection result by subtracting the original hyperspectral image X is as follows: The reconstruction error detection map R is obtained by calculating the difference between the reconstructed image and the original hyperspectral image In the formula, R i,j,: represents the non-negative scalar error at position (i,j), and X i,j,: are spectral vectors respectively; The anomalies will be displayed in the reconstruction error map R, thus achieving anomaly detection.

7. A computer storage medium, characterized in that: The storage medium stores at least one instruction, and the at least one instruction is loaded and executed by the processor to implement the method for abnormal target detection in hyperspectral remote sensing images based on Fourier conditional mask as described in any one of claims 1 to 6.

8. A device for detecting abnormal targets in hyperspectral remote sensing images based on Fourier conditional mask, characterized in that: The device includes a processor and a memory, wherein the memory stores at least one instruction, and the at least one instruction is loaded and executed by the processor to implement a method for detecting abnormal targets in hyperspectral remote sensing images based on Fourier conditional masks as described in any one of claims 1 to 6.

Images