Hyperspectral anomaly detection method and system fusing low rank and smoothness
By combining weighted tensor correlation total variation regularization and significance weight tensor, the problems of difficult parameter selection and insufficient utilization of anomalous features in hyperspectral anomaly detection are solved, and efficient anomaly target detection is achieved.
Patent Information
- Application Number
- CN202610014235.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-01-07
- Publication Date
- 2026-02-06
AI Technical Summary
Existing hyperspectral anomaly detection methods suffer from problems such as difficulty in selecting optimal parameters due to complex regularization constraints and insufficient utilization of the salience features of anomaly targets.
We employ weighted tensor correlation total variation regularization and significant weight tensors, combined with tensor robust principal component analysis, and optimize the model using the alternating direction multiplier method to obtain the optimal outlier tensor.
It improves the flexibility and accuracy of anomaly detection, reduces the false negative rate, and enables the detection of anomalous targets of different scales in complex scenarios.
Smart Images
Figure CN121482616A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of optical image processing technology, and more specifically, to a method and system for detecting hyperspectral anomalies that integrates low rank and smoothness. Background Technology
[0002] As an important application of hyperspectral image interpretation, anomaly detection aims to detect anomalous pixels that have significant spectral differences from the surrounding background without any prior information. It has been widely used in fields such as camouflage recognition, military reconnaissance, environmental monitoring, and precision agriculture. However, traditional anomaly detection algorithms usually adopt a low-dimensional signal processing approach, which destroys the inherent spatial-spectral structure information. In order to effectively maintain the inherent multidimensional structural characteristics of hyperspectral images, researchers have proposed a large number of hyperspectral anomaly detection methods based on tensor representation. Although the hyperspectral anomaly detection methods based on tensor representation have achieved satisfactory detection results, there are still two key problems to be solved. (1) Most methods usually impose various prior regularization constraints on the background components. The combination of these complex regularizations adds additional trade-off parameters and cannot maximize their effectiveness, resulting in unavoidable biases in the estimation of background components. (2) Most methods only focus on the effective representation of background components, while ignoring the full mining of prior information of anomalous components, such as the salience prior of anomalous targets. They typically use some kind of sparse norm to constrain anomalous components without considering the visual attentional differences between the background and anomalous targets, which may fail to distinguish sparse non-anomalous targets, thus limiting detection performance.
[0003] Current research has explored the sparsity of the core tensor of the gradient graph and designed a gradient core sparsity (GCS) regularization that integrates low-rank and smoothness priors. However, GCS uses Tucker decomposition to characterize the low-rank nature of the gradient tensor. The Tucker rank is actually composed of the rank of the expansion matrix obtained along each modulus of the gradient tensor. This process involves matrix operations, which may destroy the intrinsic multidimensional structural information of the gradient tensor. Subsequently, an innovative prior-integrated regularization technique called tensor-related total variational (t-CTV) regularization was proposed and successfully applied to tensor restoration tasks. t-CTV couples the low-rank and smoothness priors of the background tensor, replacing the simple addition of two independent prior regularizations, providing a theoretically accurate restoration guarantee for tensor restoration. Unfortunately, t-CTV regularization is modeled by the tensor kernel norm (TNN) in the gradient domain, which treats each singular value equally. However, this ignores the prior knowledge of singular values associated with the gradient tensor, thus limiting its ability and flexibility in handling practical problems. In addition, some studies have attempted to enhance the contrast between background and anomalous targets by constructing saliency prior weights. However, these studies are all based on matrix operations, which may not fully utilize the inherent multidimensional structural information of hyperspectral images.
[0004] To address the aforementioned issues, this patent fully leverages prior knowledge of background and anomalous components, proposing a hyperspectral anomaly detection algorithm that integrates low-rank and smoothness properties to achieve effective separation of background and anomalous targets. Summary of the Invention
[0005] The purpose of this invention is to overcome the shortcomings of the prior art and provide a hyperspectral anomaly detection method and system that integrates low rank and smoothness. It aims to solve the problems of difficulty in selecting optimal parameters and insufficient utilization of the saliency features of anomalous targets caused by various complex regularization constraints in existing hyperspectral anomaly detection methods.
[0006] To achieve the above objectives, the present invention adopts the following technical solution:
[0007] A hyperspectral anomaly detection method integrating low-rank and smoothness, the method comprising:
[0008] Step S1: Obtain hyperspectral image tensor data of the region to be detected;
[0009] Step S2: Perform band-by-band normalization on the acquired hyperspectral image tensor data;
[0010] Step S3: Decompose the hyperspectral image tensor data into background tensor and anomaly tensor;
[0011] Step S4: Design a weighted tensor-related total variation regularization and apply it to the background tensor to simultaneously capture the low-rank and smoothness priors of the background.
[0012] Step S5, construct the saliency weight tensor, in the tensor - Norm constraint is used to weight the anomalous tensor to enhance the contrast between anomalous targets and the background;
[0013] Step S6: Integrate the weighted tensor correlation total variation regularization and significance weight tensor into the tensor robust principal component analysis framework to establish an anomaly detection model;
[0014] Step S7: Optimize the established model using an efficient iterative update algorithm based on the alternating direction multiplier method to obtain the optimal anomaly tensor;
[0015] Step S8: Detect the obtained optimal anomaly tensor and generate an anomaly detection result map.
[0016] Furthermore, in step S4, the design method for weighted tensor correlation total variation regularization is as follows:
[0017]
[0018]
[0019] In the formula, Represents the background tensor The weighted tensor-related total variation regularization term; express The weighted tensor kernel norm, Represents the background tensor The gradient tensor along the nth modulus, where n is the modulus index of the tensor; yes The result of performing a discrete Fourier transform along the third dimension; Represents each front slice The singular values, where l is the index of the singular value; The weighting parameters are represented by h, v, and b, which represent the height, width, and number of bands of the hyperspectral image, respectively. ; This represents the nuclear norm.
[0020] Furthermore, in step S5, the construction steps of the saliency weight tensor are as follows:
[0021] Step S51: Calculate the adaptive weight tensor based on the spectral angular distance;
[0022] Step S52: Calculate the iterative weight tensor based on sparsity priors;
[0023] Step S53: Perform a Hadamard product on the adaptive weight tensor and the iterative weight tensor to form a salient weight tensor.
[0024]
[0025] In the formula, For significance weight tensors; For adaptive weight tensors; For iterative weight tensors; For Hadama accumulation.
[0026] Furthermore, in step S51, the adaptive weight tensor The calculation process is as follows:
[0027] Step S511, using the original hyperspectral image tensor data The adaptive weighted graph is calculated using spectral angular distance. The spectral angular distance is calculated as follows:
[0028]
[0029] In the formula, Representing hyperspectral image tensor data The Middle One pixel to be tested; Representing hyperspectral image tensor data The spectral mean vector of all pixels in the array; Indicates the first The weight of each pixel to be tested; This is a transpose operation;
[0030] Step S512, multiple adaptive weight maps Stacked into a tensor ;
[0031] Step S513, convert the tensor Taking the reciprocal yields the adaptive weight tensor. .
[0032] Furthermore, in step S52, the weight tensor is iterated. Based on anomalous tensors The calculation is performed using the following formula:
[0033]
[0034] In the formula, This represents the number of iterations. It is the first The anomalous tensor after the next iteration; It is a constant.
[0035] Furthermore, in step S6, the anomaly detection model is as follows:
[0036]
[0037]
[0038]
[0039] In the formula, , and These are hyperspectral image tensor data, background tensor, and anomaly tensor, respectively. Represents the background tensor The weighted tensor-related total variation regularization term, Represents the background tensor The gradient tensor along the nth modulus, where n is the modulus index of the tensor. Indicates the weighted parameters. Represents the nuclear norm; Representing anomaly tensors With significance weight tensor Hadama's Norm, For the significance weight tensor, For Hadamah accumulation; For regularization parameters; Background tensor The operator for calculating the gradient along the nth modulus.
[0040] Furthermore, in step S7, the anomaly detection model is optimized using an efficient iterative update algorithm based on the alternating direction multiplier method. The calculation method is as follows:
[0041]
[0042]
[0043] In the formula, and Let these represent the Lagrange multipliers and the penalty parameter, respectively. , and These are hyperspectral image tensor data, background tensor, and anomaly tensor, respectively. Represents the background tensor The gradient tensor along the nth modulus, where n is the modulus index of the tensor. Indicates the weighted parameters. Represents the nuclear norm; Representing anomaly tensors With significance weight tensor Hadama's Norm, For the significance weight tensor, For Hadamah accumulation; Background tensor The operator for calculating the gradient along the nth modulus; It is the Frobenius norm.
[0044] Furthermore, in step S8, the anomaly detection structure diagram is generated as follows:
[0045]
[0046] In the formula, Diagram showing the anomaly detection structure The Middle One element; Represents the optimal anomaly tensor The Middle One element; and For spatial location index; For band indexing; This represents the number of bands in the hyperspectral image.
[0047] This invention also proposes a hyperspectral anomaly detection system that integrates low rank and smoothness, the system comprising:
[0048] The hyperspectral image acquisition unit is used to acquire hyperspectral image tensor data of the region to be detected.
[0049] The preprocessing unit is used to perform band-by-band normalization preprocessing on hyperspectral image tensor data;
[0050] Background modeling units are used to capture the global low rank and local smoothness of background components;
[0051] Anomaly modeling unit, used to enhance the contrast between anomalous targets and the background;
[0052] The anomaly detection model establishment and optimization unit is used to establish and optimize a hyperspectral anomaly detection model that integrates low rank and smoothness, and obtain the optimal anomaly tensor.
[0053] The detection unit is used to construct a detection graph based on the optimal anomaly tensor to obtain an anomaly detection result graph.
[0054] This invention also proposes a hyperspectral anomaly detection device that integrates low rank and smoothness, comprising:
[0055] Memory, used to store computer programs and data;
[0056] A processor is used to implement the steps of the above-described hyperspectral anomaly detection method that integrates low rank and smoothness when executing a computer program.
[0057] The beneficial effects of this invention are:
[0058] This invention designs a weighted tensor-related total variational regularization to simultaneously characterize the global low-rank and local smoothness priors of the background tensor, avoiding the need for additional regularization parameters to balance these two prior regularizations. It fully considers the prior distribution information of singular values in the gradient tensor, thereby improving the ability and flexibility to handle practical problems. Simultaneously, in the tensor... Under the norm constraint, a saliency weight tensor is constructed to weight the anomaly tensor, enhancing the contrast between the anomaly target and the background. This improves the detection rate while reducing the false negative rate, enabling the detection of anomaly targets of different scales in complex scenes, providing reliable technical support for applications such as environmental monitoring and camouflage identification. Attached Figure Description
[0059] Figure 1 This is a flowchart of a hyperspectral anomaly detection method that integrates low rank and smoothness in this embodiment;
[0060] Figure 2 This is a structural framework diagram of a hyperspectral anomaly detection system that integrates low rank and smoothness in this embodiment;
[0061] Figure 3 This is a structural framework diagram of a hyperspectral anomaly detection device that integrates low rank and smoothness in this embodiment.
[0062] Figure labels: 1. Hyperspectral image acquisition unit; 2. Preprocessing unit; 3. Background modeling unit; 4. Anomaly modeling unit; 5. Anomaly detection model establishment and optimization unit; 6. Detection unit; 7. Memory; 8. Processor. Detailed Implementation
[0063] 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, and 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.
[0064] Example: A hyperspectral anomaly detection method integrating low-rank and smoothness. This method first acquires hyperspectral image tensor data of the region to be detected, then performs band-by-band normalization processing, decomposing it into a background tensor and an anomaly tensor. Next, a designed weighted tensor correlation total variation (wt-CTV) regularization is applied to constrain the background tensor, simultaneously capturing the global low-rank and local smoothness priors of the background tensor. Meanwhile, in the tensor... Under norm constraints, a weighted anomaly tensor is constructed using a saliency weight tensor to enhance the contrast between the anomaly target and the background. Finally, the weighted tensor correlated total variation (wt-CTV) regularization and the saliency weight tensor are integrated into a unified tensor robust principal component analysis framework to establish an anomaly detection model. The model is optimized using an efficient iterative update algorithm based on the alternating direction multiplier method (ADMM) to obtain the optimal anomaly tensor. A detection map is constructed based on the optimal anomaly tensor, and the anomaly detection result map is output.
[0065] This method improves anomaly detection performance by fully leveraging prior information from both background and anomalous components. For example... Figure 1 As shown, specifically:
[0066] Step S1: Acquire hyperspectral image tensor data of the area to be detected using a hyperspectral imaging device;
[0067] Step S2: Perform band-by-band normalization on the acquired hyperspectral image tensor data.
[0068] The method for band-by-band normalization is as follows:
[0069]
[0070] In the formula, For hyperspectral tensor data The A frontal slice, , , Let h, v, and b be the set of real numbers, representing the height, width, and number of bands of the hyperspectral image, respectively. and These are operations for finding the maximum and minimum values, respectively.
[0071] Step S3: Decompose the hyperspectral image tensor data into a background tensor and an anomaly tensor, the expressions of which are as follows:
[0072]
[0073] In the formula, , and These are the original hyperspectral image tensor data, the background tensor, and the anomaly tensor, respectively. , .
[0074] Step S4: Design a weighted tensor-related total variational regularization and apply it to the background tensor to simultaneously capture the low-rank and smoothness priors of the background.
[0075] The design method for weighted tensor-related total variational regularization is as follows:
[0076]
[0077]
[0078] In the formula, Represents the background tensor The weighted tensor-related total variation regularization term; express The weighted tensor kernel norm, Represents the background tensor The gradient tensor along the nth modulus, where n is the modulus index of the tensor; yes The result of performing a discrete Fourier transform along the third dimension; Represents each front slice The singular values, where l is the index of the singular value; Indicates the weighted parameters; This represents the nuclear norm.
[0079] By designing a weighted tensor correlation total variation regularization, a single term is used to simultaneously capture the global low rank and local smoothness of the background, thus avoiding the complex parameter trade-offs caused by using multiple independent regularization terms in traditional methods. At the same time, by introducing weights to weight the singular values of the tensor, the complex structure of the background can be characterized more flexibly and realistically, improving the adaptability to real hyperspectral scenes and the detection accuracy.
[0080] Step S5, construct the saliency weight tensor, in the tensor - Norm constraint is used to weight the anomalous tensor to enhance the contrast between anomalous targets and the background.
[0081] In step 5, a weight tensor that integrates spectral saliency and iterative sparsity information is constructed to provide adaptive and precise weighted constraints for anomalous components, thereby significantly increasing the separability between anomalous targets and the background in the optimization model. Specifically, the construction steps of the saliency weight tensor are as follows:
[0082] Step S51: Calculate the adaptive weight tensor based on the spectral angular distance;
[0083] Step S52: Calculate the iterative weight tensor based on sparsity priors;
[0084] Step S53: Perform a Hadamard product on the adaptive weight tensor and the iterative weight tensor to form a salient weight tensor.
[0085]
[0086] In the formula, For significance weight tensors; For adaptive weight tensors; For iterative weight tensors; For Hadama accumulation.
[0087] Furthermore, in step S51, the adaptive weight tensor The calculation process is as follows:
[0088] Step S511, using the original hyperspectral image tensor data The adaptive weighted graph is calculated using spectral angular distance. The spectral angular distance is calculated as follows:
[0089]
[0090] In the formula, Representing hyperspectral image tensor data The Middle One pixel to be tested ; Representing hyperspectral image tensor data The spectral mean vector of all pixels in the array; Indicates the first The weight of each pixel to be tested; This is the transpose operation; through this formula, an adaptive weight graph can be obtained. ;
[0091] Step S512, Application Multiple adaptive weight graphs Stacked into a tensor , ;
[0092] Step S513: Considering that anomalous pixels are far from the mean vector, pixels with large spectral angular distances can be regarded as anomalous pixels that need to be retained. Therefore, for the tensor... Taking the reciprocal yields the adaptive weight tensor. As shown below:
[0093]
[0094] In the formula, 1 in the molecule is a tensor in which all elements are 1.
[0095] Furthermore, in step S52, the weight tensor is iterated. Based on anomalous tensors Calculations are performed to enhance the sparsity of anomalous targets; the calculation formula is as follows:
[0096]
[0097] In the formula, This represents the number of iterations. It is the first The anomalous tensor after the next iteration; It is a constant, set to 0.5.
[0098] Step S6: Integrate the weighted tensor correlation total variation regularization and significance weight tensor into the tensor robust principal component analysis framework to establish an anomaly detection model.
[0099] By integrating the weighted tensor correlation total variation regularization constraint for the background with the significance weighted sparsity constraint for the anomalies into the tensor robust principal component analysis framework, we can avoid the complex parameter trade-offs caused by multiple prior regularizations and achieve joint and collaborative optimization separation of the background and anomaly tensors.
[0100] The anomaly detection model is as follows:
[0101]
[0102]
[0103]
[0104] In the formula, , and These are hyperspectral image tensor data, background tensor, and anomaly tensor, respectively. Represents the background tensor The weighted tensor-related total variation regularization term, Represents the background tensor The gradient tensor along the nth modulus, where n is the modulus index of the tensor. Indicates the weighted parameters. Represents the nuclear norm; Representing anomaly tensors With significance weight tensor Hadama's Norm, For the significance weight tensor, For Hadamah accumulation; For regularization parameters; Background tensor The operator for calculating the gradient along the nth modulus.
[0105] Step S7: The established model is optimized using an efficient iterative update algorithm based on the alternating direction multiplier method (ADMM) to obtain the optimal anomaly tensor.
[0106] The anomaly detection model, optimized using an efficient iterative update algorithm based on ADMM, is calculated as follows:
[0107]
[0108]
[0109] In the formula, and Let these represent the Lagrange multipliers and the penalty parameter, respectively. , and These are hyperspectral image tensor data, background tensor, and anomaly tensor, respectively. Represents the background tensor The gradient tensor along the nth modulus, where n is the modulus index of the tensor. Indicates the weighted parameters. Represents the nuclear norm; Representing anomaly tensors With significance weight tensor Hadama's Norm, For the significance weight tensor, For Hadamah accumulation; Background tensor The operator for calculating the gradient along the nth modulus; It is the Frobenius norm.
[0110] The optimization problem in the above equation can be decomposed into several subproblems and solved using ADMM; this method updates individual variables one by one while keeping other variables fixed, and finally obtains the optimal anomaly tensor. .
[0111] Step S8: Detect the obtained optimal anomaly tensor and generate an anomaly detection result map.
[0112] The anomaly detection structure diagram is generated as follows:
[0113]
[0114] In the formula, Diagram showing the anomaly detection structure The Middle One element; Represents the optimal anomaly tensor The Middle One element; and For spatial location index; For band indexing; This represents the number of bands in the hyperspectral image.
[0115] The proposed method employs a weighted tensor-related total variational (wt-CTV) regularization to simultaneously characterize the global low-rank and local smoothness priors of the background tensor. This avoids the need for additional regularization parameters to balance these two prior regularizations, fully considering the prior distribution information of singular values in the gradient tensor, thereby improving the ability and flexibility to handle practical problems. Furthermore, in the tensor... Under the norm constraint, a saliency weight tensor is constructed to weight the anomaly tensor, enhancing the contrast between the anomaly target and the background. This improves the detection rate while reducing the false negative rate, enabling the detection of anomaly targets of different scales in complex scenes, providing reliable technical support for applications such as environmental monitoring and camouflage identification.
[0116] This embodiment also proposes a hyperspectral anomaly detection system that integrates low-rank and smoothness, such as Figure 2 As shown, the system includes a hyperspectral image acquisition unit 1, a preprocessing unit 2, a background modeling unit 3, an anomaly modeling unit 4, an anomaly detection model establishment and optimization unit 5, and a detection unit 6.
[0117] Among them, the hyperspectral image acquisition unit 1 is used to acquire hyperspectral image tensor data of the region to be detected.
[0118] Preprocessing unit 2 is used to perform band-by-band normalization preprocessing on hyperspectral image tensor data.
[0119] Background modeling unit 3 is used to capture the global low rank and local smoothness of background components.
[0120] Anomaly modeling unit 4 is used to enhance the contrast between the anomaly target and the background.
[0121] The anomaly detection model establishment and optimization unit 5 is used to establish and optimize a hyperspectral anomaly detection model that integrates low rank and smoothness, and obtain the optimal anomaly tensor.
[0122] Detection unit 6 is used to construct a detection map based on the optimal anomaly tensor to obtain an anomaly detection result map.
[0123] This embodiment also proposes a hyperspectral anomaly detection device that integrates low-rank and smoothness, such as... Figure 3As shown, the device includes a memory 7 and a processor 8, wherein the memory 7 is used to store computer programs and data; and the processor 8 is used to implement the steps of the above-described hyperspectral anomaly detection method that integrates low rank and smoothness when executing the computer program.
[0124] The above description is merely a preferred embodiment of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for those skilled in the art, any improvements and modifications made without departing from the principles of the present invention should also be considered within the scope of protection of the present invention.
Claims
1. A hyperspectral anomaly detection method integrating low-rank and smoothness, characterized in that, The method includes: Step S1: Obtain hyperspectral image tensor data of the region to be detected; Step S2: Perform band-by-band normalization on the acquired hyperspectral image tensor data; Step S3: Decompose the hyperspectral image tensor data into background tensor and anomaly tensor; Step S4: Design a weighted tensor-related total variation regularization and apply it to the background tensor to simultaneously capture the low-rank and smoothness priors of the background. Step S5, construct the saliency weight tensor, in the tensor - Norm constraint is used to weight the anomalous tensor to enhance the contrast between anomalous targets and the background; Step S6: Integrate the weighted tensor correlation total variation regularization and significance weight tensor into the tensor robust principal component analysis framework to establish an anomaly detection model; Step S7: Optimize the established model using an efficient iterative update algorithm based on the alternating direction multiplier method to obtain the optimal anomaly tensor; Step S8: Detect the obtained optimal anomaly tensor and generate an anomaly detection result map.
2. The hyperspectral anomaly detection method integrating low rank and smoothness according to claim 1, characterized in that, In step S4, the design method for weighted tensor correlation total variation regularization is as follows: In the formula, Represents the background tensor The weighted tensor-related total variation regularization term; express The weighted tensor kernel norm, Represents the background tensor The gradient tensor along the nth modulus, where n is the modulus index of the tensor; yes The result of performing a discrete Fourier transform along the third dimension; Represents each front slice The singular values, where l is the index of the singular value; The weighting parameters are represented by h, v, and b, which represent the height, width, and number of bands of the hyperspectral image, respectively. ; This represents the nuclear norm.
3. The hyperspectral anomaly detection method integrating low rank and smoothness according to claim 1, characterized in that, In step S5, the construction steps of the significance weight tensor are as follows: Step S51: Calculate the adaptive weight tensor based on the spectral angular distance; Step S52: Calculate the iterative weight tensor based on sparsity priors; Step S53: Perform a Hadamard product on the adaptive weight tensor and the iterative weight tensor to form a salient weight tensor. In the formula, For significance weight tensors; For adaptive weight tensors; For iterative weight tensors; For Hadama accumulation.
4. The hyperspectral anomaly detection method integrating low rank and smoothness according to claim 3, characterized in that, In step S51, the adaptive weight tensor The calculation process is as follows: Step S511, using the original hyperspectral image tensor data The adaptive weighted graph is calculated using spectral angular distance. The spectral angular distance is calculated as follows: In the formula, Representing hyperspectral image tensor data The Middle One pixel to be measured; Representing hyperspectral image tensor data The spectral mean vector of all pixels in the array; Indicates the first The weight of each pixel to be tested; This is a transpose operation; Step S512, multiple adaptive weight maps Stacked into a tensor ; Step S513, convert the tensor Taking the reciprocal yields the adaptive weight tensor. .
5. The hyperspectral anomaly detection method integrating low rank and smoothness according to claim 3, characterized in that, In step S52, the weight tensor is iterated. Based on anomalous tensors The calculation is performed using the following formula: In the formula, This represents the number of iterations. It is the first The anomalous tensor after the next iteration; It is a constant.
6. The hyperspectral anomaly detection method integrating low rank and smoothness according to claim 1, characterized in that, In step S6, the anomaly detection model is as follows: In the formula, , and These are hyperspectral image tensor data, background tensor, and anomaly tensor, respectively. Represents the background tensor The weighted tensor-related total variation regularization term, Represents the background tensor The gradient tensor along the nth modulus, where n is the modulus index of the tensor. Indicates the weighted parameters. Represents the nuclear norm; Representing anomaly tensors With significance weight tensor Hadama's Norm, For the significance weight tensor, For Hadamah accumulation; For regularization parameters; Background tensor The operator for calculating the gradient along the nth modulus.
7. The hyperspectral anomaly detection method integrating low rank and smoothness according to claim 1, characterized in that, In step S7, the anomaly detection model is optimized using an efficient iterative update algorithm based on the alternating direction multiplier method. The calculation method is as follows: In the formula, and Let these represent the Lagrange multipliers and the penalty parameter, respectively. , and These are hyperspectral image tensor data, background tensor, and anomaly tensor, respectively. Represents the background tensor The gradient tensor along the nth modulus, where n is the modulus index of the tensor. Indicates the weighted parameters. Represents the nuclear norm; Representing anomaly tensors With significance weight tensor Hadama's Norm, For the significance weight tensor, For Hadamah accumulation; Background tensor The operator for calculating the gradient along the nth modulus; It is the Frobenius norm.
8. The hyperspectral anomaly detection method integrating low rank and smoothness according to claim 1, characterized in that, In step S8, the anomaly detection structure diagram is generated as follows: In the formula, Diagram showing the anomaly detection structure The Middle One element; Represents the optimal anomaly tensor The Middle One element; and For spatial location index; For band indexing; This represents the number of bands in the hyperspectral image.
9. A hyperspectral anomaly detection system for implementing the method of claim 1, characterized in that, The system includes: The hyperspectral image acquisition unit (1) is used to acquire hyperspectral image tensor data of the region to be detected; Preprocessing unit (2) is used to perform band-by-band normalization preprocessing on hyperspectral image tensor data; Background modeling unit (3) is used to capture the global low rank and local smoothness of background components; Anomaly modeling unit (4) is used to enhance the contrast between the anomaly target and the background; Anomaly detection model establishment and optimization unit (5) is used to establish a hyperspectral anomaly detection model that integrates low rank and smoothness and optimize it to obtain the optimal anomaly tensor. The detection unit (6) is used to construct a detection graph based on the optimal anomaly tensor to obtain an anomaly detection result graph.
10. A hyperspectral anomaly detection device integrating low rank and smoothness, characterized in that, include: Memory (7) is used to store computer programs and data; The processor (8) is configured to implement the steps of the hyperspectral anomaly detection method that integrates low rank and smoothness as described in any one of claims 1-8 when executing a computer program.