Hyperspectral spatial-spectral mask variational self-encoding anomaly detection method for local security and protection

Abstract

The invention belongs to the technical field of computer vision. The invention provides a hyperspectral spatial-spectral mask variational self-encoding anomaly detection method for local security and protection. According to the space-spectrum joint mask technology designed by the embodiment of the invention, the multi-dimensional information of the space and the spectrum is fully utilized, the difficulty of effectively reconstructing the abnormal region in the reconstruction process of the network is improved, and the network is enabled to pay more attention to normal space and spectrum characteristics when learning image data distribution; and the detection capability of a complex abnormal scene is obviously improved. The designed VAE reconstruction network and mask combination mechanism makes full use of the probability modeling advantage of VAE and the attention mechanism advantage of mask, the accuracy and generalization ability of anomaly detection are improved, and the interpretability of the model is enhanced. And the designed residual image processing method effectively alleviates the noise problem caused by a complex background through a domain transformation recursive filter, and realizes higher detection precision and robustness.

Description

Technical Field

[0001] The disclosed embodiments relate to the field of computer vision technology, and in particular to a hyperspectral space-spectrum masked variational autoencoder anomaly detection method for local security. Background Art

[0002] The current global security situation is becoming increasingly complex, and national and regional security is facing new challenges. The concept of vicinagearth security (VS) has emerged. Vicinagearth security is a set of diversified, cross-domain, three-dimensional, collaborative and intelligent technical systems built to meet the diverse needs of vicinagearth space, such as defense, protection, production, safety, and rescue. Its application scenarios involve low-altitude security, underwater security, and cross-domain security. The vicinity space covers waters, ground, and airspace from 1,000 meters below sea level to 10,000 meters above sea level. In the vicinityarth security system, hyperspectral images are an important input data. Compared with natural images, they have higher spectral resolution and can provide more spectral feature information in the fine identification of various types of objects. Hyperspectral image anomaly detection technology analyzes the spectral characteristics of each pixel in the hyperspectral image to identify abnormal areas that are significantly different from the normal background. It is not only of great significance in the fields of precision agriculture, urban management, and public safety, but also has broad application prospects in vicinityarth security tasks such as environmental monitoring, search and rescue, and military reconnaissance. Since the accuracy of anomaly detection directly affects the reliability of the results of hyperspectral image analysis, it is crucial to study efficient anomaly detection methods for hyperspectral images. Compared with natural image anomaly detection, hyperspectral image anomaly detection has unique task characteristics: (1) Hyperspectral images have high spectral resolution, complex spectral information, high correlation and redundancy. (2) Abnormal areas are usually sparsely distributed and have weak spectral performance, which is easily interfered by complex background and noise. The detection algorithm needs to have high-precision spectral resolution and anti-interference capabilities. (3) Due to the complex collection process of hyperspectral image anomaly detection datasets, difficult sample annotation, and high labor costs, high-quality hyperspectral anomaly detection datasets are relatively scarce. These characteristics make the hyperspectral image anomaly detection task highly complex, so it is of great research significance to develop efficient and accurate anomaly detection methods. At present, the mainstream hyperspectral image anomaly detection technology mainly relies on advanced spectral analysis and machine learning technology. According to different anomaly detection strategies, related research work can be divided into four categories based on statistics, data expression, data decomposition, and deep learning: The first category is statistical methods. However, statistical methods are usually simple and fast, but they need to assume that the data follows a Gaussian distribution, are less robust to background noise in the data, and have difficulty handling complex scenarios.

[0003] The second category is the method based on data expression. However, the method based on data expression introduces nonlinear ideas and local background calculations, which can reduce the interference of complex backgrounds. However, the amount of calculation is large, and it is more sensitive to background noise, resulting in a high false alarm rate.

[0004] The third category is the method based on data decomposition. However, the method based on data decomposition can extract background and abnormal information more finely and reduce the influence of background noise. However, it may ignore some details in the background, resulting in inaccurate decomposition results and high computational complexity.

[0005] The fourth category is deep learning-based methods. However, deep learning-based methods can automatically extract effective features and effectively handle complex scenes, but most current methods only focus on the inhibitory effect of single-dimensional masks on abnormal reconstruction, ignoring the relationship between spatial and background mask structures, which may cause image details to be lost and reduce detection efficiency.

[0006] Therefore, it is necessary to improve one or more problems existing in the above-mentioned related technical solutions.

[0007] It should be noted that this section is intended to provide background or context for the technical solutions of the present disclosure stated in the claims. The description herein is not admitted to be prior art by virtue of being included in this section. Summary of the invention

[0008] The purpose of the embodiments of the present disclosure is to provide a hyperspectral spatial-spectral masked variational autoencoder anomaly detection method for local security, thereby overcoming one or more problems caused by the limitations and defects of related technologies at least to a certain extent.

[0009] According to an embodiment of the present disclosure, a hyperspectral space-spectrum mask variational autoencoder anomaly detection method for local security is provided, the method comprising: Performing spatial mask processing on the hyperspectral image and filling the pixel values ​​that obey Gaussian distribution to obtain a hyperspectral image with spatial mask applied; Perform dimensionality reduction and segmentation on the hyperspectral image with spatial mask applied to obtain several image blocks of equal size; According to each target band in each image block, its adjacent band is obtained as input, and a spectral mask is applied to the target band to obtain a spatial-spectral mask band; A VAE reconstruction network consisting of two layers of encoders and two layers of decoders is used to reconstruct the spatial-spectral mask bands in combination with adjacent bands to obtain a reconstructed image. The first layers of the encoder and decoder of the VAE reconstruction network both use exponential linear unit activation functions, and the reconstruction loss is calculated jointly by the mean square error and Kullback-Leibler divergence.

[0010] A preliminary residual image is generated according to the comparison between the reconstructed image and the hyperspectral image, and the residual image is subjected to domain transformation recursive filtering to obtain a filtered residual image; The LRX method is used to filter the residual image to obtain the abnormal pixel distribution map.

[0011] Further, the step of performing spatial mask processing on the hyperspectral image and filling the pixel values ​​obeying Gaussian distribution to obtain the hyperspectral image with spatial mask applied includes: The hyperspectral image is divided into several equal-sized square blocks, some blocks are randomly selected, and the mask area is determined in the selected blocks by four-way random iterative expansion until the mask area reaches the preset range; The pixels in the mask area are replaced with random pixel values ​​that obey Gaussian distribution to obtain a hyperspectral image with a spatial mask applied.

[0012] Furthermore, the preset spatial mask regions are 30 to 50 groups, and the number of mask pixels in each region ranges from 15 to 32; The mean of the Gaussian distribution is 0, the variance is 1, and the upper and lower limits are the median and minimum values ​​of the original image pixels, respectively.

[0013] Furthermore, the step of reducing the dimension and segmenting the hyperspectral image with the spatial mask applied thereto to obtain a number of image blocks of equal size includes: K-means dimensionality reduction is performed on the hyperspectral image with spatial mask, and the reduced image is cropped into several image blocks of equal size using a sliding window.

[0014] Furthermore, the total loss of the VAE reconstruction network is expressed as: (1) in, p is the number of square blocks, MSE is the mean square error between the original pixel and the reconstructed pixel, KL is the original pixel and the reconstructed pixel KL Divergence; Further, the step of generating a preliminary residual image based on the comparison between the reconstructed image and the hyperspectral image, and performing domain transformation recursive filtering on the residual image to obtain a filtered residual image includes: By comparing the hyperspectral image and the reconstructed image, a preliminary residual image is obtained, and the preliminary residual image is converted into a transform domain by domain transformation to obtain a transformed residual image; Applying a recursive filter to the transformed residual image in the transform domain to obtain a recursively filtered result; The result of recursive filtering is converted back to the original space through inverse transformation to obtain the filtered residual image.

[0015] Furthermore, the process of converting the preliminary residual image to the transform domain through domain transformation to obtain the transformed residual image is: (2) in, is the transformed residual image, is the domain transformation operation; The recursive filtering process is: (3) in, For the The result after iterations is is a recursive filter, represents the convolution operation, For the -The result after 1 iteration, the initial condition is ; The result of recursive filtering is converted back to the original space through inverse transformation: (4) in, is the residual image after filtering, is the inverse transform operation.

[0016] Furthermore, the number of iterations of the recursive filter is 3.

[0017] The technical solution provided by the embodiments of the present disclosure may have the following beneficial effects: In the embodiments of the present disclosure, through the above-mentioned hyperspectral space-spectrum mask variational autoencoder anomaly detection method for on-site security, on the one hand, the designed spatial-spectral joint masking technology makes full use of the multi-dimensional information of space and spectrum, improves the difficulty of the network to effectively reconstruct abnormal areas during the reconstruction process, and makes the network pay more attention to normal spatial and spectral characteristics when learning image data distribution, significantly improving the detection ability of complex abnormal scenes. On the other hand, the designed VAE reconstruction network and mask combination mechanism fully utilizes the probabilistic modeling advantages of VAE and the attention mechanism advantages of mask, improves the accuracy and generalization ability of anomaly detection, and enhances the interpretability of the model. And the designed residual image processing method effectively alleviates the noise problem caused by complex background through domain transformation recursive filter, and achieves higher detection accuracy and robustness. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The accompanying drawings herein are incorporated into the specification and constitute a part of the specification, illustrate embodiments consistent with the present disclosure, and together with the specification are used to explain the principles of the present disclosure. Obviously, the accompanying drawings described below are only some embodiments of the present disclosure, and for ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without creative work.

[0019] Figure 1 A step diagram showing a hyperspectral space-spectrum mask variational autoencoder anomaly detection method for on-site security in an exemplary embodiment of the present disclosure; Figure 2 A specific flow chart of the hyperspectral space-spectrum mask variational autoencoder anomaly detection method for local security in an exemplary embodiment of the present disclosure is shown; Figure 3 The overall framework diagram of the hyperspectral space-spectrum mask variational autoencoder anomaly detection method for local security in the exemplary embodiment of the present disclosure is shown; Figure 4 A comparison of ROC curves of the method proposed in the present application and other methods on all data sets in an exemplary embodiment of the present disclosure is shown; Figure 5 A visual comparison diagram of the present application on all data sets in an exemplary embodiment of the present disclosure is shown. DETAILED DESCRIPTION

[0020] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in a variety of forms and should not be construed as limited to the examples set forth herein; rather, these embodiments are provided so that the disclosure will be more comprehensive and complete and to fully convey the concepts of the example embodiments to those skilled in the art. The described features, structures, or characteristics may be combined in any suitable manner in one or more embodiments.

[0021] In addition, the accompanying drawings are only schematic illustrations of the embodiments of the present disclosure and are not necessarily drawn to scale. The same reference numerals in the figures represent the same or similar parts, and their repeated descriptions will be omitted. Some of the block diagrams shown in the accompanying drawings are functional entities and do not necessarily correspond to physically or logically independent entities.

[0022] This example implementation provides a hyperspectral space-spectrum mask variational autoencoder anomaly detection method for local security. Figure 1 As shown in , the hyperspectral space-spectrum mask variational autoencoder anomaly detection method for local security may include: Step S101: performing spatial mask processing on the hyperspectral image and filling the pixel values ​​that obey the Gaussian distribution to obtain a hyperspectral image with a spatial mask applied; Step S102: performing dimensionality reduction and segmentation on the hyperspectral image with the spatial mask applied thereto to obtain a number of image blocks of equal size; Step S103: according to each target band in each image block, obtain its adjacent band as input, and apply a spectral mask to the target band to obtain a spatial-spectral mask band; Step S104: using a VAE reconstruction network including two layers of encoders and two layers of decoders, the spatial-spectral mask bands are reconstructed in combination with adjacent bands to obtain a reconstructed image; wherein the first layers of the encoder and decoder of the VAE reconstruction network both use an exponential linear unit activation function, and the reconstruction loss is jointly calculated by the mean square error and the Kullback-Leibler divergence; Step S105: generating a preliminary residual image according to the comparison between the reconstructed image and the hyperspectral image, and performing domain transformation recursive filtering on the residual image to obtain a filtered residual image; Step S106: Using the LRX method to filter the residual image to obtain an abnormal pixel distribution map.

[0023] Through the above-mentioned hyperspectral space-spectrum mask variational autoencoder anomaly detection method for on-site security, on the one hand, the designed spatial-spectral joint masking technology makes full use of the multi-dimensional information of space and spectrum, which improves the difficulty of the network to effectively reconstruct abnormal areas during the reconstruction process, so that the network pays more attention to normal spatial and spectral features when learning image data distribution, and significantly improves the detection ability of complex abnormal scenes. On the other hand, the designed VAE reconstruction network and mask combination mechanism fully utilizes the probabilistic modeling advantages of VAE and the attention mechanism advantages of mask, improves the accuracy and generalization ability of anomaly detection, and enhances the interpretability of the model. And the designed residual image processing method effectively alleviates the noise problem caused by complex background through domain transformation recursive filter, and achieves higher detection accuracy and robustness.

[0024] Next, we will refer to Figures 1 to 5 The various steps of the hyperspectral space-spectrum masked variational autoencoder anomaly detection method for local security in this example implementation are described in more detail.

[0025] In step S101 , a spatial mask process is performed on the hyperspectral image, and pixel values ​​that obey a Gaussian distribution are filled in to obtain a hyperspectral image with a spatial mask applied thereto.

[0026] Specifically, the hyperspectral image is divided into a number of square blocks of equal size, some blocks are randomly selected, and the mask area is determined in the selected blocks by a four-way random iterative expansion method until the mask area reaches a preset range; The pixels in the mask area are replaced with random pixel values ​​that obey Gaussian distribution to obtain a hyperspectral image with a spatial mask applied.

[0027] Furthermore, the preset spatial mask regions are 30 to 50 groups, and the number of mask pixels in each region ranges from 15 to 32; The mean of the Gaussian distribution is 0, the variance is 1, and the upper and lower limits are the median and minimum values ​​of the original image pixels, respectively.

[0028] More specifically, when performing spatial masking, the image is divided into square areas of equal size, randomly select Then, a starting pixel is randomly selected from each selected region, and the mask pixel is determined by four-way random iterative expansion until the mask area in the region reaches , Randomly selected from integer interval To prevent the model from identifying the spatial mask area as a new abnormal area, it is necessary to use a Gaussian distribution Fill the mask area with random pixel values.

[0029] In step S102, the hyperspectral image with the spatial mask applied thereto is subjected to dimensionality reduction and segmentation to obtain a number of image blocks of equal size.

[0030] Specifically, K-means dimensionality reduction is performed on the hyperspectral image with spatial mask applied, and the image after dimensionality reduction is cropped into several image blocks of equal size using a sliding window.

[0031] More specifically, when performing spectral masking, the spatial mask image is first subjected to K-means dimensionality reduction, and after obtaining the dimensionality-reduced image, the image is segmented into several patches through a sliding window. Then, each band of each patch is reconstructed in turn. During the reconstruction process, the adjacent bands of the band to be reconstructed are used as network inputs, while the band to be reconstructed itself will be masked and will not participate in the reconstruction process.

[0032] In step S103 and step S104, according to each target band in each image block, its adjacent band is obtained as input, and a spectral mask is applied to the target band to obtain a spatial-spectral mask band; a VAE reconstruction network including two layers of encoders and two layers of decoders is used to reconstruct the spatial-spectral mask band in combination with the adjacent bands to obtain a reconstructed image; wherein the first layers of the encoder and decoder of the VAE reconstruction network both use an exponential linear unit activation function, and the reconstruction loss is calculated jointly by the mean square error and the Kullback-Leibler divergence.

[0033] Specifically, the network contains two layers of encoders and two layers of decoders to capture and extract the characteristic information of the data more deeply. The first layer of the encoder and decoder uses the exponential linear unit activation function. The reconstruction loss of VAE is calculated using the mean square error and Kullback-Leibler divergence, and the total loss formula is: (1) In step S105 and step S106, a preliminary residual image is generated based on the comparison between the reconstructed image and the hyperspectral image, and the residual image is subjected to domain transform recursive filtering to obtain a filtered residual image; the filtered residual image is processed using the LRX method to obtain an abnormal pixel distribution map.

[0034] Specifically, the residual image processing method is: by comparing the original image and the reconstructed image, a preliminary residual image is obtained. , and transform it to the transform domain by domain transformation: (2) in, is the residual image after transformation, is the domain transformation operation. Then a recursive filter is applied in the transform domain. The recursive filtering process is: (3) in, For the The result after iterations is is the filter, represents the convolution operation, and the initial condition is The result of recursive filtering is converted back to the original space by inverse transformation: (4) in, is the residual image after filtering, is the inverse transform operation. Finally, the LRX algorithm is used to detect the filtered residual image and generate an abnormal pixel distribution map.

[0035] In a specific embodiment, Figure 2 As shown in FIG. , it is a specific flow chart of the hyperspectral space-spectrum mask variational autoencoder anomaly detection method for local security; Figure 3 As shown, it is the overall framework diagram of the hyperspectral space-spectrum mask variational autoencoder anomaly detection method for local security. The implementation steps of this application are as follows: 1. Prepare the dataset Create the datasets required for training and testing: (1) Airport-Beach-Urban (ABU) dataset: collected by the Airborne Visible / Infrared Imaging Spectrometer (AVIRIS) and the Reflection Optical System Imaging Spectrometer (ROSIS-03) sensors. Six representative images were used in the experiment, with sizes of 100 × 100 × 205, 150 × 150 × 188, 100 × 100 × 193, 100 × 100 × 188, 150 × 150 × 102, and 100 × 100 × 191, respectively.

[0036] (2) Salinas dataset: collected by the AVIRIS sensor. The spatial resolution is 3.7 meters, with a total of 204 spectral bands and a size of 180 × 180 pixels.

[0037] (3) Lakeshore dataset: collected by the SAMSON sensor. The spatial resolution is 10 meters, with a total of 156 spectral bands and a size of 200 × 200 pixels.

[0038] The model is trained in an unsupervised manner, and all datasets are used as both training and testing datasets.

[0039] 2. Design of VAE reconstruction network model based on spatial-spectral joint mask The spatial-spectral joint mask, VAE reconstruction network and residual image processing method are designed to constitute a VAE network model based on spatial-spectral joint mask.

[0040] Spatial-spectral joint masking: When performing spatial masking, the image is divided into square areas of equal size, randomly select Then, a starting pixel is randomly selected from each selected region, and the mask pixel is determined by four-way random iterative expansion until the mask area in the region reaches , Randomly selected from integer interval To prevent the model from identifying the spatial mask area as a new abnormal area, it is necessary to use a Gaussian distribution The mask area is filled with random pixel values; when performing spectral masking, the spatial mask image is first subjected to K-means dimensionality reduction, and after obtaining the reduced dimensionality image, the image is segmented into several patches through a sliding window. Then, each band of each patch is reconstructed in turn. During the reconstruction process, the adjacent bands of the band to be reconstructed are used as network inputs, and the band to be reconstructed itself will be masked and will not participate in the reconstruction process.

[0041] VAE reconstruction network: The network contains two layers of encoders and two layers of decoders to capture and extract the characteristic information of the data more deeply. The first layer of the encoder and decoder uses the exponential linear unit activation function. The reconstruction loss of VAE is calculated using the mean square error and Kullback-Leibler divergence, and the total loss formula is: (1) Residual image processing method: By comparing the original image and the reconstructed image, a preliminary residual image is obtained. , and transform it to the transform domain by domain transformation: (2) in, is the residual image after transformation, is the domain transformation operation. Then a recursive filter is applied in the transform domain. The recursive filtering process is: (3) in, For the The result after iterations is is the filter, represents the convolution operation, and the initial condition is The result of recursive filtering is converted back to the original space by inverse transformation: (4) in, is the residual image after filtering, is the inverse transform operation. Finally, the LRX algorithm is used to detect the filtered residual image and generate an abnormal pixel distribution map.

[0042] 3. Experimental conditions All experiments in this application were completed on a server equipped with an Intel(R) Xeon(R) Silver 4210R CPU and a 24GB NVIDIA Geforce RTX 4090 GPU for model training and testing. The system environment was Ubuntu 22.04.2. The algorithm was implemented in Python and relied on the Pytorch library. The Python version used was 3.9.19 and the Pytorch version was 1.13.1.

[0043] The training and test data used in the experiment come from the public data sets Airport-Beach-Urban, Salinas, and Lakeshore. The specific content of the data sets has been introduced above.

[0044] 4. Experimental content According to the steps given in the specific implementation method, the results of the network model are shown in the following table.

[0045] Table 1 AUC comparison results of this application on all datasets

[0046] Table 1 is the AUC comparison results of the method proposed in this application and other methods on all data sets. The bold font represents the best result. Figure 4 This is a comparison chart of ROC curves of the method proposed in this application and other methods on all data sets. Figure 5 This is a visualization comparison chart of this application on all data sets. As shown in the chart, the detection performance of the hyperspectral space-spectrum mask variational autoencoder anomaly detection method for local security proposed in this application has reached an advanced level.

[0047] Through the above-mentioned hyperspectral space-spectrum mask variational autoencoder anomaly detection method for on-site security, on the one hand, the designed spatial-spectral joint masking technology makes full use of the multi-dimensional information of space and spectrum, which improves the difficulty of the network to effectively reconstruct abnormal areas during the reconstruction process, so that the network pays more attention to normal spatial and spectral features when learning image data distribution, and significantly improves the detection ability of complex abnormal scenes. On the other hand, the designed VAE reconstruction network and mask combination mechanism fully utilizes the probabilistic modeling advantages of VAE and the attention mechanism advantages of mask, improves the accuracy and generalization ability of anomaly detection, and enhances the interpretability of the model. And the designed residual image processing method effectively alleviates the noise problem caused by complex background through domain transformation recursive filter, and achieves higher detection accuracy and robustness.

[0048] It should be understood that the terms "center", "longitudinal", "lateral", "length", "width", "thickness", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside", "clockwise", "counterclockwise", etc. in the above description indicate orientations or positional relationships based on the orientations or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the embodiments of the present disclosure and simplifying the description, and do not indicate or imply that the referred device or element must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be understood as a limitation on the embodiments of the present disclosure.

[0049] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the features. In the description of the embodiments of the present disclosure, the meaning of "multiple" is two or more, unless otherwise clearly and specifically defined.

[0050] In the embodiments of the present disclosure, unless otherwise clearly specified and limited, the terms "installed", "connected", "connected", "fixed" and the like should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection or an indirect connection through an intermediate medium, it can be the internal connection of two elements or the interaction relationship between two elements. For ordinary technicians in this field, the specific meanings of the above terms in the present disclosure can be understood according to specific circumstances.

[0051] In the embodiments of the present disclosure, unless otherwise clearly specified and limited, a first feature being "above" or "below" a second feature may include the first and second features being in direct contact, or may include the first and second features not being in direct contact but being in contact through another feature between them. Moreover, a first feature being "above", "above" and "above" a second feature includes the first feature being directly above and obliquely above the second feature, or simply indicates that the first feature is higher in level than the second feature. A first feature being "below", "below" and "below" a second feature includes the first feature being directly below and obliquely below the second feature, or simply indicates that the first feature is lower in level than the second feature.

[0052] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example" or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present disclosure. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification.

[0053] Those skilled in the art will readily appreciate other embodiments of the present disclosure after considering the specification and practicing the invention disclosed herein. This application is intended to cover any modification, use or adaptation of the present disclosure, which follows the general principles of the present disclosure and includes common knowledge or customary techniques in the art that are not disclosed in the present disclosure. The specification and examples are intended to be exemplary only, and the true scope and spirit of the present disclosure are indicated by the appended claims.

Claims

1. A hyperspectral space-spectrum masked variational autoencoder anomaly detection method for on-site security, characterized in that: The method includes: Performing spatial mask processing on the hyperspectral image and filling the pixel values ​​that obey Gaussian distribution to obtain a hyperspectral image with spatial mask applied; Perform dimensionality reduction and segmentation on the hyperspectral image with spatial mask applied to obtain several image blocks of equal size; According to each target band in each image block, its adjacent band is obtained as input, and a spectral mask is applied to the target band to obtain a spatial-spectral mask band; A VAE reconstruction network consisting of two layers of encoders and two layers of decoders is used to reconstruct the spatial-spectral mask bands in combination with adjacent bands to obtain a reconstructed image. The first layers of the encoder and decoder of the VAE reconstruction network both use an exponential linear unit activation function, and the reconstruction loss is calculated jointly by the mean square error and the Kullback-Leibler divergence. A preliminary residual image is generated according to the comparison between the reconstructed image and the hyperspectral image, and the residual image is subjected to domain transformation recursive filtering to obtain a filtered residual image; The LRX method is used to filter the residual image to obtain the abnormal pixel distribution map.

2. According to claim 1, the hyperspectral space-spectrum masked variational autoencoder anomaly detection method for local security is characterized by: The step of performing spatial mask processing on the hyperspectral image and filling the pixel values ​​obeying the Gaussian distribution to obtain the hyperspectral image with the spatial mask applied includes: The hyperspectral image is divided into several equal-sized square blocks, some blocks are randomly selected, and the mask area is determined in the selected blocks by four-way random iterative expansion until the mask area reaches the preset range; The pixels in the mask area are replaced with random pixel values ​​that obey Gaussian distribution to obtain a hyperspectral image with a spatial mask applied.

3. According to claim 2, the hyperspectral space-spectrum masked variational autoencoder anomaly detection method for local security is characterized in that: The preset spatial mask regions are 30 to 50 groups, and the number of mask pixels in each region ranges from 15 to 32; The mean of the Gaussian distribution is 0, the variance is 1, and the upper and lower limits are the median and minimum values ​​of the original image pixels, respectively.

4. The hyperspectral space-spectrum masked variational autoencoder anomaly detection method for local security according to claim 3 is characterized in that: The step of reducing the dimension and segmenting the hyperspectral image with the spatial mask applied thereto to obtain a number of image blocks of equal size includes: K-means dimensionality reduction is performed on the hyperspectral image with spatial mask, and the reduced image is cropped into several image blocks of equal size using a sliding window.

5. According to claim 4, the hyperspectral space-spectrum masked variational autoencoder anomaly detection method for local security is characterized in that: The total loss of the VAE reconstruction network is expressed as: （1） in, p is the number of pixels in the square block, MSE is the mean square error between the original pixel and the reconstructed pixel, KL is the original pixel and the reconstructed pixel KL Divergence.

6. The hyperspectral space-spectrum masked variational autoencoder anomaly detection method for local security according to claim 5 is characterized in that: The steps of generating a preliminary residual image according to the comparison between the reconstructed image and the hyperspectral image, and performing domain transformation recursive filtering on the residual image to obtain a filtered residual image include: By comparing the hyperspectral image and the reconstructed image, a preliminary residual image is obtained, and the preliminary residual image is converted into a transform domain by domain transformation to obtain a transformed residual image; Applying a recursive filter to the transformed residual image in the transform domain to obtain a recursively filtered result; The result of recursive filtering is converted back to the original space through inverse transformation to obtain the filtered residual image.

7. The hyperspectral space-spectrum masked variational autoencoder anomaly detection method for local security according to claim 6 is characterized in that: The process of converting the preliminary residual image to the transform domain through domain transformation to obtain the transformed residual image is: （2） in, is the residual image after transformation, is the domain transformation operation; The recursive filtering process is: （3） in, For the The result after iterations is is a recursive filter, represents the convolution operation, For the -The result after 1 iteration, the initial condition is ; The result of recursive filtering is converted back to the original space through inverse transformation: （4） in, is the residual image after filtering, is the inverse transform operation.

8. The hyperspectral space-spectrum masked variational autoencoder anomaly detection method for local security according to claim 6 is characterized in that: The number of iterations of the recursive filter is 3.

Images