Hyperspectral anomaly detection method and system based on plug-and-play tensor low-rank approximation

Abstract

The present invention relates to a hyperspectral anomaly detection method and system based on plug-and-play tensor low-rank approximation. The detection method comprises the following steps: inputting a detected image data set in a data input module and storing it in a database; extracting the detected image data set in the database, and extracting data features from the detected image data set using a plug-and-play tensor low-rank approximation method in a data feature extraction module; inputting the extracted data features into a data analysis module, and using an alternating direction multiplier method to perform data analysis and processing; inputting the results of the analysis and processing into a result analysis module, and performing image anomaly analysis in the result analysis module. The detection system comprises a data input module, a data feature extraction module, a data analysis module and a result analysis module. Aiming at the problem that hyperspectral anomaly detection has poor noise resistance, the present invention proposes a PnP‑TLRA method, which has better detection capability under the conditions of complex background and noise interference.

Description

Technical Field

[0001] The present invention relates to the technical field of image processing, and in particular to a hyperspectral anomaly detection method and system based on plug-and-play tensor low-rank approximation. Background Art

[0002] Hyperspectral images contain rich spectral and spatial information, so they have been widely used in many fields, such as agricultural monitoring, mineral exploration and military defense. In these applications, hyperspectral anomaly detection is a technology that detects pixels with significant spectral differences from the surrounding background without prior information. Anomalies in hyperspectral images are defined as a small number of other pixels with significant spectral characteristics different from the vast majority of background pixels.

[0003] In traditional abnormal target detection methods based on statistical models, low-order statistics have limitations in accurately describing the background and abnormal targets. Therefore, abnormal target detection methods based on matrix decomposition are more effective, such as methods based on low-rank sparse matrix decomposition.

[0004] However, when the existing hyperspectral imaging anomaly detection methods are interfered by noise, the detection performance will be seriously reduced, and they have shortcomings such as poor noise resistance, sensitivity to parameters, and poor scalability. Summary of the invention

[0005] In view of some or all of the problems in the prior art, the present invention provides a hyperspectral anomaly detection method based on plug-and-play tensor low-rank approximation, which comprises the following steps:

[0006] Input the detected image data set into the data input module and store it in the database;

[0007] Extracting a detected image data set from a database, and using a plug-and-play tensor low-rank approximation method in a data feature extraction module to extract data features from the detected image data set;

[0008] Inputting the extracted data features into a data analysis module and using an alternating direction multiplier method to perform data analysis and processing; and

[0009] The result of the analysis process is input into a result analysis module, and abnormality analysis of the image is performed in the result analysis module.

[0010] Further, extracting data features from the detected image data set using a plug-and-play tensor low-rank approximation method in the data feature extraction module includes:

[0011] The detected image dataset is regarded as a third-order tensor. Combining sparse representation and tensor decomposition, the image is separated into three aspects: background, anomaly, and noise. A plug-and-play tensor low-rank approximation model is constructed. The model is expressed as:

[0012]

[0013]

[0014] Among them, min represents the minimum value of the function, X is a tensor, A is a dictionary, Z is a coefficient, E is anomaly, and N is noise; represents the plug-and-play prior, ||ε|| F，0 represents the F0 norm of the anomaly tensor ε; It represents the square of the Frobenius norm of the noise N, which can distinguish noise and filter out abnormal noise; λ and μ are both trade-off parameters.

[0015] Furthermore, extracting data features from the detected image data set using a plug-and-play tensor low-rank approximation method in the data feature extraction module also includes:

[0016] A denoising network using a pre-trained convolutional neural network is used to extract data features.

[0017] Furthermore, the data analysis and processing using the alternating direction multiplier method includes:

[0018] Based on the constructed plug-and-play tensor low-rank approximation model, an augmented Lagrangian function is constructed, and the function is expressed as follows:

[0019]

[0020]

[0021] Among them, X is a tensor, A is a dictionary, Z is a representation coefficient, E is anomaly, and N is noise; represents the plug-and-play prior; ||ε|| 2，0 represents the 2,0 norm of the anomaly tensor ε; W is the Lagrange multiplier; λ and μ are both trade-off parameters; the penalty parameter β>0, <·> represents the matrix inner product, Represents the square of the F-norm of the matrix.

[0022] Furthermore, the data analysis and processing using the alternating direction multiplier method also includes:

[0023] For the augmented Lagrangian function, a program is designed using an alternating direction multiplier method to solve it, and the calculation iterations are performed in sequence to update the values ​​of noise N, Lagrangian multiplier W, anomaly E, and background Z; and

[0024] If the relative error between the results of two calculation iterations is less than 10^-3 or the number of iterations reaches 200, the calculation iteration is stopped.

[0025] Further, performing abnormal analysis of the image in the result analysis module includes:

[0026] The results of the analysis were qualitatively and quantitatively analyzed using ROC curves and AUC values.

[0027] The present invention also provides a hyperspectral anomaly detection system based on plug-and-play tensor low-rank approximation, which includes the following modules:

[0028] A data input module is configured to input a detected image data set and store it in a database;

[0029] A data feature extraction module is configured to extract a detected image dataset from a database, wherein a plug-and-play tensor low-rank approximation method is used to extract data features from the detected image dataset;

[0030] A data analysis module, configured to input the extracted data features and perform data analysis using an alternating direction multiplier method; and

[0031] The result analysis module is configured to input the result of the analysis process and perform abnormality analysis of the image therein.

[0032] The present invention also provides a computer system, comprising:

[0033] a processor configured to execute machine-readable instructions; and

[0034] A memory stores machine-readable instructions, which, when executed by a processor, perform the steps of the hyperspectral anomaly detection method.

[0035] The present invention also provides a computer-readable storage medium having machine-readable instructions stored thereon, wherein the machine-readable instructions, when executed by a processor, execute the steps of the hyperspectral anomaly detection method.

[0036] Compared with the prior art, the present invention has the following advantages:

[0037] 1. The algorithm used in the present invention uses a pre-trained convolutional neural network, which can automatically extract features.

[0038] 2. The present invention develops a unified first-order optimization algorithm based on the framework of alternating direction multiplier method.

[0039] 3. The present invention uses a new tensor sparsity to represent local anomalies, which can accurately extract anomalies. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] To further illustrate the above and other advantages and features of various embodiments of the present invention, a more specific description of various embodiments of the present invention will be presented with reference to the accompanying drawings. It will be understood that these drawings only depict typical embodiments of the present invention and are therefore not to be considered as limiting the scope thereof. In the accompanying drawings, for clarity, identical or corresponding parts will be represented by identical or similar reference numerals.

[0041] Figure 1 A schematic diagram of a process of a hyperspectral anomaly detection method according to an embodiment of the present invention;

[0042] Figure 2 A schematic diagram of a flow chart of solving an augmented Lagrangian function using an alternating direction multiplier method according to an embodiment of the present invention; and

[0043] Figure 3 Schematic diagram of a hyperspectral anomaly detection system according to an embodiment of the present invention. DETAILED DESCRIPTION

[0044] In the following description, the present invention is described with reference to various embodiments. However, those skilled in the art will recognize that various embodiments may be implemented without one or more specific details or with other replacement and / or additional methods or components. In other cases, well-known structures or operations are not shown or described in detail to avoid obscuring the inventive point of the present invention. Similarly, for the purpose of explanation, specific numbers and configurations are set forth to provide a comprehensive understanding of embodiments of the present invention. However, the present invention is not limited to these specific details.

[0045] In this specification, reference to "one embodiment" or "the embodiment" means that a particular feature, structure, or characteristic described in conjunction with the embodiment is included in at least one embodiment of the present invention. The phrase "in one embodiment" appearing in various places in this specification does not necessarily all refer to the same embodiment.

[0046] It should be noted that the embodiments of the present invention describe the method steps in a specific order, but this is only for the purpose of illustrating the specific embodiment, rather than limiting the order of the steps. On the contrary, in different embodiments of the present invention, the order of the steps can be adjusted according to actual needs.

[0047] In the present invention, each module of the system according to the present invention can be implemented using software, hardware, firmware or a combination thereof. When the module is implemented using software, the function of the module can be implemented by a computer program flow, for example, the module can be implemented by a code segment (such as a code segment of a language such as C, C++) stored in a storage device (such as a hard disk, a memory, etc.), wherein the corresponding function of the module can be implemented when the code segment is executed by a processor. When the module is implemented using hardware, the function of the module can be implemented by setting a corresponding hardware structure, for example, the function of the module can be implemented by hardware programming a programmable device such as a field programmable gate array (FPGA), or the function of the module can be implemented by designing an application-specific integrated circuit (ASIC) including electronic devices such as a plurality of transistors, resistors and capacitors. When the module is implemented using firmware, the function of the module can be written into a read-only memory such as an EPROM or EEPROM of the device in the form of a program code, and the corresponding function of the module can be implemented when the program code is executed by a processor. In addition, certain functions of the module may need to be implemented by separate hardware or by collaboration with the hardware, for example, the detection function is implemented by corresponding sensors (such as proximity sensors, acceleration sensors, gyroscopes, etc.), the signal transmission function is implemented by corresponding communication devices (such as Bluetooth devices, infrared communication devices, baseband communication devices, Wi-Fi communication devices, etc.), the output function is implemented by corresponding output devices (such as displays, speakers, etc.), and so on.

[0048] The technical solutions in the embodiments of the present invention are described clearly and completely below in conjunction with the accompanying drawings in the embodiments of the present invention. Figure 1 FIG. 1 is a flow chart of a hyperspectral anomaly detection method according to an embodiment of the present invention. Figure 1 As shown, the hyperspectral anomaly detection method based on plug-and-play tensor low-rank approximation includes the following steps:

[0049] First, the detected image dataset is input into the data input module and stored in the database. The detected image dataset can be a public dataset, such as the commonly used hyperspectral anomaly detection dataset ABU.

[0050] Next, extract the detected image dataset from the database, and use the plug-and-play tensor low-rank approximation method in the data feature extraction module to extract data features from the detected image dataset. The plug-and-play tensor low-rank approximation method, namely PnP-TLRA (plug-and-play Tensor Low-Rank Approximation), regards the detected image dataset as a third-order tensor, and combines sparse representation and tensor decomposition to separate the image into three aspects: background, anomaly, and noise. PnP-TLRA uses a pre-trained convolutional neural network denoising network for data feature extraction. A pre-trained convolutional neural network refers to a convolutional neural network model that is trained in advance. Once the pre-trained convolutional neural network is trained, it can be used as a general feature extractor and used directly in the task. This denoising network based on convolutional neural network (CNN) is called FFDNet. FFDNet can not only effectively learn the prior structure of CNN, but also can be effectively solved by the existing pre-trained neural network. Therefore, when processing HSI (Hyperspectral Image), the PnP-TLRA proposed in the present invention is superior to the existing methods in terms of background representation and robustness to noise.

[0051] Build a plug-and-play tensor low-rank approximation model, which is represented as follows,

[0052]

[0053]

[0054] Among them, min represents the minimum value of the function, X is a tensor, A is a dictionary, Z is a coefficient, E is anomaly, and N is noise; represents the plug-and-play prior, ||ε|| F，0 represents the F0 norm of the anomaly tensor ε; It represents the square of the Frobenius norm of the noise N, which can distinguish noise and filter out abnormal noise; λ and μ are both trade-off parameters.

[0055] Noise usually satisfies Gaussian distribution, so it can be represented by Frobenius norm, but anomalies cannot. Existing methods usually adopt approximate strategies. The tensor sparse representation proposed in the present invention is a more essential characterization of anomalies. Therefore, the proposed plug-and-play tensor low-rank approximation can better extract the abnormal part in the hyperspectral image.

[0056] Next, the extracted data features are input into a data analysis module, and the data analysis and processing is performed using the alternating direction multiplier method. The data analysis and processing using the alternating direction multiplier method includes:

[0057] Based on the constructed plug-and-play tensor low-rank approximation model, an augmented Lagrangian function is constructed, and the function is expressed as follows:

[0058]

[0059] Among them, X is a tensor, A is a dictionary, Z is a representation coefficient, E is anomaly, and N is noise; represents the plug-and-play prior; ||ε|| 2，0 represents the 2,0 norm of the anomaly tensor ε; W is the Lagrange multiplier; λ and μ are both trade-off parameters; the penalty parameter β>0, <·> represents the matrix inner product, Represents the square of the F-norm of the matrix.

[0060] Figure 2 FIG. 1 is a flow chart of solving the augmented Lagrangian function using the alternating direction multiplier method according to an embodiment of the present invention. Figure 2 As shown, for the augmented Lagrangian function, the alternating direction multiplier method is used to design a program for solving, and the iterations are calculated in sequence to update the values ​​of noise N, Lagrangian multiplier W, anomaly E, and background Z; and if the relative error of the results of two calculation iterations is less than 10^-3 or the number of iterations reaches 200, the calculation iteration is stopped.

[0061] Finally, the results of the analysis and processing are input into the result analysis module, and the abnormality analysis of the image is performed in the result analysis module. The E obtained by solving the problem using the alternating direction multiplier method is the abnormality. The ROC curve and AUC value are used in the result analysis module to perform qualitative and quantitative analysis on the results of the analysis and processing.

[0062] ROC curve is the receiver operating characteristics curve. ROC curve is a common indicator for evaluating model performance and judging the detection ability of various methods. The AUC value is the area under the ROC curve. The ROC curve is drawn by a series of target detection rate and false alarm rate point pairs, which can directly reflect the relationship between the two detection probabilities. Whenever a discrimination threshold is given, a corresponding set of detection rate and false alarm rate values ​​can be accurately calculated, and a series of point pairs are obtained by setting different thresholds. In order to conduct accurate quantitative analysis, the AUC value can be obtained by integrating the area under the ROC curve to measure the pros and cons of the method and intuitively evaluate the performance of the anomaly detection method. The AUC value ranges from 0 to 1. The closer the value is to 1, the better the performance of the detection method.

[0063] The effect of the present invention can be further illustrated by the following experimental results of the abnormality detection method.

[0064] The commonly used hyperspectral anomaly detection data ABU data set is used as the detected image data set to compare the hyperspectral anomaly detection performance of the PnP-TLRA (plug-and-play tensor low-rank approximation method) proposed in the present invention with that of the current mainstream methods. The ABU data set contains 13 hyperspectral images. The current mainstream methods include: GRXD (global-RX detection, global RX detection method), RPCA-RX (robust principal component analysis-RX, RX method based on robust principal component analysis), LSMAD (LRaSMD-based Mahalanobis distance, RX method based on low-rank sparse matrix decomposition), LRASR (low-rank and sparse representation, low-rank and sparse representation), GTVLRR (graph and total variation regularized low-rank representation, graph and total variation regularized low-rank representation), TPCA (tensor principal component analysis, tensor principal component analysis), PTA (prior-based tensor approximation, prior tensor approximation). Table 1 shows the anomaly detection AUC scores of the PnP-TLRA method proposed in the present invention and the current mainstream methods for the ABU dataset. As shown in Table 1, the PnP-TLRA method proposed in the present invention obtains the best AUC score, that is, the PnP-TLRA method proposed in the present invention achieves the best anomaly detection.

[0065] Table 1 Anomaly detection AUC scores for the ABU dataset

[0066]

[0067] The plug-and-play tensor low-rank approximation method proposed in the present invention, namely the PnP-TLRA method, regards the detected image dataset as a third-order tensor, and combines sparse representation and tensor decomposition to separate the image into three aspects: background, anomaly, and noise. Aiming at the problem of anomaly detection in hyperspectral imaging, the plug-and-play tensor low-rank approximation method proposed in the present invention has better detection capability under the conditions of complex background and noise interference.

[0068] The present invention also provides a hyperspectral anomaly detection system based on plug-and-play tensor low-rank approximation. Figure 3 FIG. 1 is a schematic diagram of a hyperspectral anomaly detection system according to an embodiment of the present invention. Figure 3 As shown, the system includes the following modules:

[0069] A data input module is configured to input a detected image data set and store it in a database;

[0070] A data feature extraction module is configured to extract a detected image dataset from a database, wherein a plug-and-play tensor low-rank approximation method is used to extract data features from the detected image dataset;

[0071] A data analysis module, configured to input the extracted data features and perform data analysis using an alternating direction multiplier method; and

[0072] The result analysis module is configured to input the result of the analysis process and perform abnormality analysis of the image therein.

[0073] In one embodiment, a computer system is also provided, including a memory and a processor, the memory storing machine-readable instructions, and the processor implementing the following processing steps when executing the machine-readable instructions: inputting a detected image data set into a data input module and storing it in a database; extracting the detected image data set from the database, and using a plug-and-play tensor low-rank approximation method in a data feature extraction module to extract data features from the detected image data set; inputting the extracted data features into a data analysis module, and using an alternating direction multiplication method to perform data analysis and processing; and inputting the results of the analysis and processing into a result analysis module, and performing image abnormality analysis in the result analysis module.

[0074] The memory includes: a USB flash drive, a read-only memory (ROM), a random access memory (RAM), a mobile hard disk, a magnetic disk or an optical disk, and other media that can store machine-readable instructions.

[0075] It can be understood that in addition to the memory and processor mentioned above, the above-mentioned computer system also includes other software and hardware components not listed in this specification. The specific components can be determined according to the model of specific data processing equipment in different application scenarios, and this specification will not list them one by one in detail.

[0076] In one embodiment, a computer-readable storage medium is also provided, on which machine-readable instructions are stored, and the machine-readable instructions implement the following processing steps when executed by a processor: inputting a detected image data set into a data input module and storing it in a database; extracting the detected image data set in the database, and using a plug-and-play tensor low-rank approximation method in a data feature extraction module to extract data features from the detected image data set; inputting the extracted data features into a data analysis module, and using an alternating direction multiplication method to perform data analysis and processing; and inputting the results of the analysis and processing into a result analysis module, and performing image abnormality analysis in the result analysis module.

[0077] Although various embodiments of the present invention are described above, it should be understood that they are presented as examples only and not as limitations. It is obvious to those skilled in the relevant art that various combinations, modifications and changes can be made thereto without departing from the spirit and scope of the present invention. Therefore, the breadth and scope of the present invention disclosed herein should not be limited by the exemplary embodiments disclosed above, but should only be defined according to the attached claims and their equivalents.

Claims

1. A hyperspectral anomaly detection method based on plug-and-play tensor low-rank approximation, characterized in that: The steps include: Input the detected image data set into the data input module and store it in the database; Extracting a detected image data set from a database, and using a plug-and-play tensor low-rank approximation method in a data feature extraction module to extract data features from the detected image data set; Inputting the extracted data features into a data analysis module and using an alternating direction multiplier method to perform data analysis and processing; as well as Inputting the result of the analysis process into a result analysis module, and performing abnormality analysis of the image in the result analysis module; Wherein, using the plug-and-play tensor low-rank approximation method in the data feature extraction module to extract data features from the detected image data set includes: The detected image dataset is regarded as a third-order tensor. Combining sparse representation and tensor decomposition, the image is separated into three aspects: background, anomaly, and noise. A plug-and-play tensor low-rank approximation model is constructed. The model is expressed as: Among them, min represents the minimum value of the function, X is a tensor, A is a dictionary, Z is the coefficient, ε is anomaly, and N is noise; represents the plug-and-play prior, ||ε|| F,0 represents the F0 norm of the anomaly tensor ε; It represents the square of the Frobenius norm of the noise N, which can distinguish noise and filter out abnormal noise; λ and μ are both trade-off parameters; A denoising network using a pre-trained convolutional neural network is used to extract data features; Among them, the analysis and processing of data using the alternating direction multiplier method includes: Based on the constructed plug-and-play tensor low-rank approximation model, an augmented Lagrangian function is constructed, and the function is expressed as follows: Among them, X is a tensor, A is a dictionary, Z is the representation coefficient, ε is anomaly, and N is noise; represents the plug-and-play prior; ||ε|| 2，0 L represents the abnormal tensor ε 2,0 norm; W is the Lagrange multiplier; λ and μ are both trade-off parameters; penalty parameter β>0, <·> represents the inner product of the matrix, represents the square of the F norm of the matrix; For the augmented Lagrangian function, an alternating direction multiplier method is used to design a program for solving, and the calculation iterations are performed in sequence to update the values ​​of noise N, Lagrangian multiplier W, anomaly ε, and background Z; and If the relative error between the results of two calculation iterations is less than 10^-3 or the number of iterations reaches 200, the calculation iteration is stopped.

2. The hyperspectral anomaly detection method based on plug-and-play tensor low-rank approximation according to claim 1, characterized in that: The abnormal analysis of images performed in the result analysis module includes: The results of the analysis were qualitatively and quantitatively analyzed using ROC curves and AUC values.

3. A system using the hyperspectral anomaly detection method based on plug-and-play tensor low-rank approximation as claimed in claim 1 or 2, characterized in that: Includes the following modules: A data input module is configured to input a detected image data set and store it in a database; A data feature extraction module is configured to extract a detected image dataset from a database, wherein a plug-and-play tensor low-rank approximation method is used to extract data features from the detected image dataset; A data analysis module, configured to input the extracted data features and perform data analysis and processing using an alternating direction multiplier method; as well as The result analysis module is configured to input the result of the analysis process and perform abnormality analysis of the image therein.

4. A computer system, characterized in that: include: a processor configured to execute machine-readable instructions; as well as A memory storing machine-readable instructions, which, when executed by a processor, perform the steps of the method according to claim 1 or 2.

5. A computer-readable storage medium, characterized in that: Machine readable instructions are stored thereon, and when the machine readable instructions are executed by a processor, the steps of the method according to claim 1 or 2 are performed.

Images