Hyperspectral remote sensing image anomaly detection method, system, device and storage medium

Abstract

The application relates to the technical field of remote sensing image detection, and discloses a hyperspectral remote sensing image anomaly detection method, a system, equipment and a storage medium. The method comprises the following steps: acquiring a hyperspectral remote sensing image to be detected; inputting the hyperspectral remote sensing image to be detected into a pre-trained anomaly detection model to perform anomaly detection, and obtaining a detection result of the hyperspectral remote sensing image to be detected; and the anomaly detection model comprises a first convolution module, a second convolution module, a first Transform module, a first Resize layer and a first full connection layer which are sequentially connected. The application detects background and abnormal samples through clustering inspiration, effectively reduces the dependence of a network model on artificial labeling samples, and not only distinguishes abnormal information and background information during the training of the anomaly detection model, but also utilizes a reconstructed image of the hyperspectral image to improve the training effect of the model, so that the whole model has a local and global receptive field, and the detection precision of the anomaly detection model is effectively improved.

Description

Technical Field

[0001] This invention relates to the field of remote sensing image detection technology, and in particular to an anomaly detection method, system, device, and storage medium for hyperspectral remote sensing images. Background Technology

[0002] Anomaly detection in remote sensing images refers to extracting discrete points from multispectral images to indicate outliers in pixels. Each pixel in a hyperspectral image contains spectral information or features specific to a land cover, and different material compositions exhibit different spectral information. Therefore, regions containing the same type of material generally have highly similar spectra, while regions containing different land covers typically display locally statistically independent combinations of different spectral distributions. Thus, anomaly detection in hyperspectral remote sensing images plays a crucial role in remote sensing image processing and other applications, such as mineral crystal detection and oil and gas exploration.

[0003] With the rapid development of artificial intelligence, deep learning-based methods are increasingly being applied to anomaly detection in hyperspectral remote sensing images. However, existing deep learning-based anomaly detection models suffer from several drawbacks. Excellent detection results require a large amount of training data, but anomaly-labeled samples are often insufficient, hindering the effective suppression of background information and enhancement of anomaly information. Furthermore, single anomaly detection models struggle to simultaneously extract local and global discriminative features from both anomalies and background information in hyperspectral images, impacting the accuracy of anomaly detection results. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention provides an anomaly detection method, system, device, and storage medium for hyperspectral remote sensing images. This addresses the issues of insufficient labeled data and the difficulty in simultaneously extracting local and global discriminative features of anomalies and background information in hyperspectral images, thereby reducing reliance on manually labeled samples and improving anomaly detection accuracy.

[0005] In a first aspect, the present invention provides an anomaly detection method for hyperspectral remote sensing images, the method comprising:

[0006] Acquire the hyperspectral remote sensing image to be detected;

[0007] The hyperspectral remote sensing image to be detected is input into a pre-trained anomaly detection model for anomaly detection, and the detection result of the hyperspectral remote sensing image to be detected is obtained. The anomaly detection model includes a first convolutional module, a second convolutional module, a first Transformer module, a first Resize layer and a first fully connected layer connected in sequence.

[0008] Furthermore, the first convolutional module includes at least one first convolutional unit, and the first convolutional unit includes a first convolutional layer and a batch normalization layer connected in sequence.

[0009] The second convolutional module includes at least one second convolutional unit, and the second convolutional unit includes a first sub-convolutional module, a second sub-convolutional module, and a second convolutional layer connected in sequence; the output data of the second convolutional unit is the input data of the second convolutional unit and the output data of the second convolutional layer.

[0010] The first Transformer module includes at least one Transformer unit.

[0011] Furthermore, the training steps of the anomaly detection model include:

[0012] Detect false background and false anomaly samples in a hyperspectral remote sensing image set to generate a sample set;

[0013] Construct a decoder model whose structure is symmetrical to the anomaly detection model;

[0014] The sample set is input into the anomaly detection model for encoding to generate latent features, and the latent features are input into the decoder model for decoding to generate reconstructed samples;

[0015] The network parameters of the anomaly detection model are adjusted according to the first loss function to obtain the trained anomaly detection model.

[0016] Furthermore, the step of detecting false background samples and false anomaly samples in the hyperspectral remote sensing image set to generate a sample set includes:

[0017] Cluster analysis is performed on a set of hyperspectral remote sensing images to obtain specific classes, which are the classes with the largest variance in the clustering results and whose sample size meets the sample threshold;

[0018] Calculate the first distance between each sample in the specific class and the class center, select the maximum distance from each of the first distances, and divide the first distance by the maximum distance to obtain the quotient value corresponding to each sample;

[0019] Based on the quotient, the samples in the specific class are classified to obtain a pseudo-background sample set and a pseudo-anomaly sample set.

[0020] Further, the step of classifying samples in the specific class according to the quotient to obtain a pseudo-background sample set and a pseudo-anomaly sample set includes:

[0021] Determine whether the quotient is less than a first threshold. If it is less than the first threshold, then the sample corresponding to the quotient is used as a pseudo background sample. Otherwise, determine whether the quotient is greater than a second threshold.

[0022] If the value is greater than the second threshold, the sample corresponding to the quotient value will be regarded as a pseudo-abnormal sample.

[0023] Based on the pseudo-background samples and the pseudo-anomaly samples, a pseudo-background sample set and a pseudo-anomaly sample set are generated.

[0024] Furthermore, the decoder model includes a second fully connected layer, a second resize layer, a second Transformer module, a third convolutional module, a fourth convolutional module, and an activation layer connected in sequence.

[0025] The third convolutional module has the same structure as the second convolutional module, the fourth convolutional module has the same structure as the first convolutional module, and the second Transformer module has the same structure as the first Transformer module.

[0026] Furthermore, the step of inputting the latent features into the decoder model for decoding to generate reconstructed samples includes:

[0027] The network parameters of the decoder model are adjusted according to the second loss function;

[0028] The decoder model is then used to decode the latent features after they have been adjusted, generating reconstructed samples.

[0029] Secondly, the present invention provides an anomaly detection system for hyperspectral remote sensing images, the system comprising:

[0030] The image acquisition module is used to acquire the hyperspectral remote sensing image to be detected;

[0031] The image detection module is used to input the hyperspectral remote sensing image to be detected into a pre-trained anomaly detection model for anomaly detection, and obtain the detection result of the hyperspectral remote sensing image to be detected. The anomaly detection model includes a first convolutional module, a second convolutional module, a first Transformer module, a first Resize layer and a first fully connected layer connected in sequence.

[0032] Thirdly, embodiments of the present invention also provide a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the above-described method.

[0033] Fourthly, embodiments of the present invention also provide a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the above-described method.

[0034] This invention provides an anomaly detection method, system, computer device, and storage medium for hyperspectral remote sensing images. The invention utilizes clustering-inspired detection to identify background and anomalous samples, effectively reducing the network model's reliance on manually labeled samples. Furthermore, the anomaly detection model not only distinguishes between anomalous and background information during network training but also leverages reconstructed images from the hyperspectral image to enhance training performance. This imbues the model with both local and global receptive fields, thereby significantly improving the detection accuracy. Compared to supervised or semi-supervised anomaly detection models, the anomaly detection method provided by this invention has lower reliance on manually labeled samples and higher detection accuracy. Attached Figure Description

[0035] Figure 1 This is a flowchart illustrating the anomaly detection method for hyperspectral remote sensing images in an embodiment of the present invention;

[0036] Figure 2 yes Figure 1 Network structure diagram of the anomaly detection model;

[0037] Figure 3 This is a schematic diagram of the process of training the anomaly detection model in an embodiment of the present invention;

[0038] Figure 4 yes Figure 3 A schematic diagram of the network structure of the decoder model;

[0039] Figure 5 This is a schematic diagram of the process of using a trained anomaly detection model for anomaly detection in an embodiment of the present invention;

[0040] Figure 6 This is a diagram showing the detection results of an anomaly detection experiment using an anomaly detection model in an embodiment of the present invention.

[0041] Figure 7 This is a schematic diagram of the anomaly detection system for hyperspectral remote sensing images in an embodiment of the present invention;

[0042] Figure 8 This is an internal structural diagram of the computer device in an embodiment of the present invention. Detailed Implementation

[0043] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0044] Please see Figure 1 The first embodiment of the present invention proposes an anomaly detection method for hyperspectral remote sensing images, comprising steps S10 to S20:

[0045] Step S10: Acquire the hyperspectral remote sensing image to be detected;

[0046] Step S20: Input the hyperspectral remote sensing image to be detected into a pre-trained anomaly detection model for anomaly detection to obtain the detection result of the hyperspectral remote sensing image to be detected. The anomaly detection model includes a first convolutional module, a second convolutional module, a first Transformer module, a first Resize layer, and a first fully connected layer connected in sequence.

[0047] In existing technologies, the neural networks used for anomaly detection in hyperspectral remote sensing images are relatively simple and cannot simultaneously extract local and global discriminative features of anomalies and background information in hyperspectral images. Therefore, this invention provides a novel anomaly detection model, the network structure of which is as follows: Figure 2 As shown, the model includes a first convolutional module, a second convolutional module, a first Transformer module, a first resize layer, and a first fully connected layer. The first convolutional module comprises N0 first convolutional units, each consisting of a first convolutional layer and a batch normalization layer connected sequentially. The second convolutional module comprises N1 second convolutional units, each consisting of a first sub-convolutional module, a second sub-convolutional module, and a second convolutional layer connected sequentially. The first Transformer module comprises N2 Transformer units. N0, N1, and N2 are all greater than or equal to 1. It should be noted that in the anomaly detection model, N... i The value can be set according to the actual situation. For example, when the server performance is good and the input data is complex, a multi-layer network structure can be set for anomaly detection. There are no restrictions here. The following only uses a single-layer network structure as an example.

[0048] In this embodiment, the first convolutional unit performs preliminary local feature extraction on the input hyperspectral remote sensing image, and the second convolutional unit performs secondary extraction on the extracted local features to extract more local details. The output data of the second convolutional unit is not only the output data of the second convolutional layer, but also includes the data input to the second convolutional unit. That is, the input data of the second convolutional unit and the output data of the second convolutional layer are used together as the output data of the second convolutional unit and input into the Transformer unit. The Transformer unit is used to extract global features. That is, the anomaly detection model provided by this invention can simultaneously extract local and global features. By obtaining more comprehensive detailed features, the accuracy of the detection results can be effectively improved. After the extracted global features are adjusted by the first resize layer, the latent representation learned by this model, i.e., the latent features, is finally output by the first fully connected layer, and the final anomaly detection result is obtained based on the latent features.

[0049] To improve the detection performance of the above anomaly detection model, please refer to [link / reference needed]. Figure 3 The present invention also provides a method for training a network model, wherein the steps for training an anomaly detection model using this method include:

[0050] Detect false background and false anomaly samples in a hyperspectral remote sensing image set to generate a sample set;

[0051] Construct a decoder model whose structure is symmetrical to the anomaly detection model;

[0052] The hyperspectral samples in the sample set are input into the anomaly detection model for encoding to generate latent features, and the latent features are input into the decoder model for decoding to generate reconstructed samples;

[0053] The network parameters of the anomaly detection model are adjusted according to the first loss function to obtain the trained anomaly detection model.

[0054] In training anomaly detection models, the first problem to be solved is the dependence on labeled data. To address this, this embodiment provides a clustering-inspired approach to process the dataset, that is, to detect background and anomalous samples through clustering. The specific steps are as follows:

[0055] Cluster analysis is performed on a set of hyperspectral remote sensing images to obtain specific classes, which are the classes with the largest variance in the clustering results and whose sample size meets the sample threshold;

[0056] Calculate the first distance between each sample in the specific class and the class center, select the maximum distance from each of the first distances, and divide the first distance by the maximum distance to obtain the quotient value corresponding to each sample;

[0057] Based on the quotient, the samples in the specific class are classified to obtain a pseudo-background sample set and a pseudo-anomaly sample set.

[0058] Specifically, taking the HYDICE Urban hyperspectral remote sensing image set as an example, K-means clustering is first performed on the HYDICE Urban hyperspectral images through a while loop until a suitable clustering result and a specific class C are obtained. s The selection of a specific class is based on the special characteristics of anomalous samples in anomaly detection, namely, the small number of anomalies and their difference from the surrounding environment. Therefore, in this embodiment, the specific class C is selected. s It refers to C s The variance of is the largest among all clusters, and C s The number of samples in a cluster is relatively small, such as the smallest, second smallest, or third smallest. A category that satisfies these two conditions is a specific class.

[0059] After obtaining the specific class, the next step is to select abnormal samples from the specific class. In this embodiment, the criterion for judging abnormal samples is the distance relationship between each sample in the specific class and the class center, specifically:

[0060] Determine whether the quotient is less than a first threshold. If it is less than the first threshold, then the sample corresponding to the quotient is used as a pseudo background sample. Otherwise, determine whether the quotient is greater than a second threshold.

[0061] If the value is greater than the second threshold, the sample corresponding to the quotient value will be regarded as a pseudo-abnormal sample.

[0062] Based on the pseudo-background samples and the pseudo-anomaly samples, a pseudo-background sample set and a pseudo-anomaly sample set are generated.

[0063] In this embodiment, the distances between each sample of a specific class and the class center are first obtained, and the maximum distance is selected from these distances. Then, the ratio between each sample's distance and the maximum distance is calculated. If the ratio is less than a first threshold, the sample is determined to be a pseudo-background sample; if the ratio is greater than a second threshold, the sample is determined to be a pseudo-abnormal sample. Finally, a pseudo-background sample set and a pseudo-abnormal sample set are generated. The first threshold is preferably 0.4, and the second threshold is preferably 0.88. Other thresholds can also be set, and no further restrictions are imposed here. After generating the pseudo-abnormal sample set and the pseudo-background sample set, the other images in the image set are used as the normal sample set, thus obtaining the data sample set used for model training, which includes both normal and abnormal samples.

[0064] The data preprocessing method provided in this embodiment can not only effectively reduce the dependence of network model training on manually labeled samples, but also improve the separability of anomalies through the preliminary extraction of background and anomaly information, thereby improving the training effect of the model.

[0065] To further improve the training performance of the anomaly detection model, this invention also creates a decoder model for decoding training of the anomaly detection model. The structure of the decoder model is symmetrical to that of the anomaly detection model. Please refer to [link / reference]. Figure 4 The decoder model comprises a second fully connected layer, a second resize layer, a second Transformer module, a third convolutional module, a fourth convolutional module, and an activation layer, all connected in sequence. The third convolutional module has the same structure as the second convolutional module, the fourth convolutional module has the same structure as the first convolutional module, and the second Transformer module has the same structure as the first Transformer module. In other words, the unit structures and internal structures of each module in the decoder model are identical to those of the corresponding modules in the anomaly detection model. For example, if the first convolutional module of the anomaly detection model contains N0 first convolutional units, each consisting of a first convolutional layer and a batch normalization layer, then the fourth convolutional module of the decoder model also contains N0 first convolutional units, each consisting of a first convolutional layer and a batch normalization layer. The roles of each module in the decoder model are the same as those of their corresponding modules in the anomaly detection model, while the activation layer in the decoder model is used for non-linear activation.

[0066] The reason for designing the decoder model in this invention is to generate a reconstructed sample corresponding to the input hyperspectral sample by decoding the latent features output by the anomaly detection model. Then, the anomaly detection model is trained based on the similarity between the reconstructed sample and the hyperspectral sample. During the training of the anomaly detection model, the training of the decoder model is also carried out simultaneously. The training of the decoder model is based on adjusting the network parameters of the decoder model based on the latent features learned from the sample and pseudo-background sample.

[0067] Specifically, this invention updates the network parameters of the anomaly detection model and the decoder model using the following loss function:

[0068]

[0069]

[0070] In the formula, x (i) and Let L be the i-th input hyperspectral sample and its corresponding reconstructed sample. (i)Let C be the latent representation, or latent feature, learned by the anomaly detection model, and let C be the average value of the latent features learned by the anomaly detection model from the pseudo-background samples (i.e., the center of the latent feature hypersphere). (i) ={-1,+1} represents the label of the i-th sample, when l (i) When =+1, x(i) belongs to the background sample set; when l (i) When x = -1, (i) It belongs to the abnormal sample set.

[0071] The purpose of the first loss function, Loss1, is to train the reconstructed samples... Compared with the input hyperspectral sample x (i) The goal is to make the features of the background samples as similar as possible, while the purpose of the second loss function, Loss2, is to train the features L of the background samples to be as similar as possible. (i) The features of the anomalous samples are mapped to the center C of the hypersphere, while the features of the anomalous samples are far from the center C of the hypersphere.

[0072] Without loss of generality, in this embodiment, let x∈R 1×175 For hyperspectral samples, L∈R 1×175 and Let L be the learned latent representation and the reconstructed hyperspectral sample, respectively. Then L can be expressed as:

[0073]

[0074] In the formula, and Θ Encoder These represent the anomaly detection model and its corresponding parameters, respectively.

[0075] and It can be represented as:

[0076]

[0077] In the formula, and Θ decoder These represent the decoder model and its corresponding parameters, respectively.

[0078] In this embodiment, the anomaly detection model The function of the decoder model is to map the input data to a low-dimensional representation. The function of this is to map this low-dimensional representation back to the original input space, and to reconstruct the sample Its function is to learn a compact representation of the input hyperspectral sample x using an unsupervised learning algorithm.

[0079] After obtaining the trained anomaly detection model through the above training steps, the remote sensing image to be detected can be input into the anomaly detection model to obtain the corresponding latent representation and the final anomaly detection result. Please refer to [link to relevant documentation]. Figure 5 For a given test sample y∈R located at point (m,n) 1×175 Where 0≤m≤80, 0≤n≤100, this is input into the pre-trained anomaly detection model. And obtain the corresponding latent feature L y And the final anomaly detection result D(m,n), where L y Specifically, it is expressed as follows:

[0080]

[0081] D(m,n) is specifically represented as:

[0082] D(m,n)=||L y -C|| 2

[0083] To verify the detection effectiveness of the anomaly detection method provided by this invention, the original hyperspectral remote sensing image is tested using this method, and the detection results are as follows. Figure 6 As shown, Figure 6 (a) is a false-color image composed of three bands of the original hyperspectral remote sensing image, and its corresponding Ground Truth is as follows: Figure 6 As shown in (b). Figure 6 (c) shows the detection results obtained by the anomaly detection method provided by the present invention. It can be seen that the anomaly detection method provided by the present invention can effectively detect discrete points in hyperspectral remote sensing images, thus proving that the anomaly detection method of the present invention has extremely high detection accuracy.

[0084] This embodiment provides an anomaly detection method for hyperspectral remote sensing images. Compared with traditional methods that rely on manually labeled samples and have low detection accuracy, this invention uses clustering-inspired detection to detect background and anomaly samples, effectively reducing the dependence on manually labeled samples for model training. By creating a decoder model to train the anomaly detection model, the network training not only distinguishes between anomalies and background information but also reconstructs the hyperspectral image, giving the entire anomaly detection model both local and global receptive fields, thereby improving the detection efficiency and accuracy of the anomaly detection model.

[0085] Please see Figure 7 Based on the same inventive concept, the second embodiment of this invention proposes an anomaly detection system for hyperspectral remote sensing images, comprising:

[0086] Image acquisition module 10 is used to acquire the hyperspectral remote sensing image to be detected;

[0087] The image detection module 20 is used to input the hyperspectral remote sensing image to be detected into a pre-trained anomaly detection model for anomaly detection, and obtain the detection result of the hyperspectral remote sensing image to be detected. The anomaly detection model includes a first convolutional module, a second convolutional module, a first Transformer module, a first Resize layer and a first fully connected layer connected in sequence.

[0088] The technical features and effects of the hyperspectral remote sensing image anomaly detection system proposed in this embodiment of the invention are the same as those of the method proposed in this embodiment of the invention, and will not be repeated here. Each module in the above-mentioned hyperspectral remote sensing image anomaly detection system can be implemented entirely or partially through software, hardware, or a combination thereof. Each module can be embedded in or independent of the processor in a computer device in hardware form, or it can be stored in the memory of a computer device in software form, so that the processor can call and execute the operations corresponding to each module.

[0089] Please see Figure 8 The diagram illustrates the internal structure of a computer device in one embodiment. This computer device can specifically be a terminal or a server. The computer device includes a processor, memory, network interface, display, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores an operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The network interface is used to communicate with external terminals via a network connection. When the computer program is executed by the processor, it implements an anomaly detection method for hyperspectral remote sensing images. The display screen can be a liquid crystal display (LCD) or an e-ink display. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad mounted on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0090] Those skilled in the art will understand that Figure 8 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computing devices may include more or fewer components than shown in the diagram, or combine certain components, or have the same component arrangement.

[0091] Furthermore, embodiments of the present invention also propose a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the above-described method.

[0092] Furthermore, embodiments of the present invention also propose a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of the above-described method.

[0093] In summary, the anomaly detection method, system, device, and storage medium for hyperspectral remote sensing images proposed in this invention involve acquiring a remote sensing image to be detected; inputting the image into a pre-trained anomaly detection model for detection; and obtaining the detection result of the anomaly detection image. The anomaly detection model comprises a first convolutional module, a second convolutional module, a first Transformer module, a first Resize layer, and a first fully connected layer connected in sequence. This invention uses clustering heuristics to detect background and anomalous samples, effectively reducing the network model's dependence on manually labeled samples. Furthermore, for the created anomaly detection model, its network training not only distinguishes between anomalous and background information but also utilizes reconstructed images from hyperspectral images to improve the model's training effect, giving the entire model both local and global receptive fields, thus effectively improving the detection accuracy of the anomaly detection model.

[0094] The various embodiments in this specification are described in a progressive manner. For directly identical or similar parts of the embodiments, refer to each other. Each embodiment focuses on its differences from other embodiments. In particular, the system embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions in the method embodiments. It should be noted that the technical features of the above embodiments can be combined arbitrarily. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as the combination of these technical features does not contradict each other, it should be considered within the scope of this specification.

[0095] The embodiments described above are merely preferred embodiments of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various improvements and substitutions without departing from the technical principles of this invention, and these improvements and substitutions should also be considered within the scope of protection of this application. Therefore, the scope of protection of this patent application should be determined by the scope of the claims.

Claims

1. An anomaly detection method for hyperspectral remote sensing images, characterized in that, include: Acquire the hyperspectral remote sensing image to be detected; The hyperspectral remote sensing image to be detected is input into a pre-trained anomaly detection model for anomaly detection, and the detection result of the hyperspectral remote sensing image to be detected is obtained. The anomaly detection model includes a first convolutional module, a second convolutional module, a first Transformer module, a first Resize layer and a first fully connected layer connected in sequence. The training steps of the anomaly detection model include: Detect false background and false anomaly samples in a hyperspectral remote sensing image set to generate a sample set; Construct a decoder model whose structure is symmetrical to the anomaly detection model; The sample set is input into the anomaly detection model for encoding to generate latent features, and the latent features are input into the decoder model for decoding to generate reconstructed samples; The network parameters of the anomaly detection model are adjusted according to the first loss function to obtain the trained anomaly detection model; The step of inputting the latent features into the decoder model for decoding to generate reconstructed samples includes: The network parameters of the decoder model are adjusted according to the second loss function; The decoder model is then used to decode the latent features after they have been adjusted to generate reconstructed samples. The first loss function is expressed by the following formula: In the formula, x (i) This represents the i-th input hyperspectral sample. This represents the reconstructed sample corresponding to the i-th input hyperspectral sample; The second loss function is expressed by the following formula: In the formula, L (i) Let represent the i-th latent representation learned by the anomaly detection model, and C represent the average value of the latent features learned by the anomaly detection model from the pseudo-background samples. (i) This represents the label of the i-th sample.

2. The anomaly detection method for hyperspectral remote sensing images according to claim 1, characterized in that, The first convolutional module includes at least one first convolutional unit, and the first convolutional unit includes a first convolutional layer and a batch normalization layer connected in sequence. The second convolutional module includes at least one second convolutional unit, and the second convolutional unit includes a first sub-convolutional module, a second sub-convolutional module, and a second convolutional layer connected in sequence; the output data of the second convolutional unit is the input data of the second convolutional unit and the output data of the second convolutional layer. The first Transformer module includes at least one Transformer unit.

3. The anomaly detection method for hyperspectral remote sensing images according to claim 1, characterized in that, The step of detecting false background and false anomaly samples in the hyperspectral remote sensing image set to generate a sample set includes: Cluster analysis is performed on a set of hyperspectral remote sensing images to obtain specific classes, which are the classes with the largest variance in the clustering results and whose sample size meets the sample threshold; Calculate the first distance between each sample in the specific class and the class center, select the maximum distance from each of the first distances, and divide the first distance by the maximum distance to obtain the quotient value corresponding to each sample; Based on the quotient, the samples in the specific class are classified to obtain a pseudo-background sample set and a pseudo-anomaly sample set.

4. The anomaly detection method for hyperspectral remote sensing images according to claim 3, characterized in that, The step of classifying samples in the specific class according to the quotient to obtain a pseudo-background sample set and a pseudo-anomaly sample set includes: Determine whether the quotient is less than a first threshold. If it is less than the first threshold, then the sample corresponding to the quotient is used as a pseudo background sample. Otherwise, determine whether the quotient is greater than a second threshold. If the value is greater than the second threshold, the sample corresponding to the quotient value will be regarded as a pseudo-abnormal sample. Based on the pseudo-background samples and the pseudo-anomaly samples, a pseudo-background sample set and a pseudo-anomaly sample set are generated.

5. The anomaly detection method for hyperspectral remote sensing images according to claim 2, characterized in that, The decoder model includes a second fully connected layer, a second resize layer, a second Transformer module, a third convolutional module, a fourth convolutional module, and an activation layer connected in sequence. The third convolutional module has the same structure as the second convolutional module, the fourth convolutional module has the same structure as the first convolutional module, and the second Transformer module has the same structure as the first Transformer module.

6. An anomaly detection system for hyperspectral remote sensing images, said system being applied to the method described in any one of claims 1 to 5, characterized in that, include: The image acquisition module is used to acquire the hyperspectral remote sensing image to be detected; The image detection module is used to input the hyperspectral remote sensing image to be detected into a pre-trained anomaly detection model for anomaly detection, and obtain the detection result of the hyperspectral remote sensing image to be detected. The anomaly detection model includes a first convolutional module, a second convolutional module, a first Transformer module, a first Resize layer and a first fully connected layer connected in sequence.

7. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 5.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 5.

Images