A hyperspectral image real-time local anomaly detection method based on multi-line wave band processing

Abstract

The application discloses a hyperspectral image real-time local anomaly detection method based on multi-line wave band processing, which comprises the following steps: reading hyperspectral image data, initializing the number of wave bands n determined by the number of filters, initializing the number of lines j based on wave band extraction, and initializing the interval number k between different wave bands; converting the data into data required for real-time detection processing according to the values of the number of wave bands n, the number of lines j and the interval number k; establishing a state equation of a spectral vector correlation matrix R(n) of hyperspectral data pixels; updating the inverse matrix R(n) of the correlation matrix R(n) by using a block matrix inversion formula ‑1 ; detecting the hyperspectral data required for real-time detection processing by using a multi-line wave band correlation matrix real-time local anomaly RX operator, and finally obtaining the result of the hyperspectral image real-time local anomaly detection. The application realizes local anomaly detection of the hyperspectral image, avoids high-dimensional data storage and repeated calculation, and has a good real-time local anomaly detection effect.

Description

Technical Field

[0001] This invention relates to the field of hyperspectral image detection, and in particular to a method for real-time local anomaly detection of hyperspectral images based on multi-band processing. Background Technology

[0002] Anomaly detection is an important application area of ​​hyperspectral image processing. Based on the different statistical information of the operators, anomaly detection algorithms can be divided into global anomaly detection and local anomaly detection. Global anomaly detection algorithms use the background estimation statistics of the entire hyperspectral image, and in most cases, they can perform anomaly detection well. However, when the anomalous target is weak or only exists within a local area of ​​the image and is submerged in the global background, global anomaly detection algorithms cannot be applied. Local anomaly detection refers to constructing detection operators using local background statistical information to achieve hyperspectral anomaly detection; some algorithms can achieve fast anomaly detection.

[0003] In practical hyperspectral detection applications, as the spatial and spectral resolution of hyperspectral images continues to improve, the sheer volume of data increases information, leading to lower efficiency in anomaly detection algorithms. Furthermore, many anomalous targets have very short dwell times, appearing suddenly and disappearing rapidly. If data processing is severely delayed, it diminishes the advantages and efficiency of hyperspectral data applications. Even fast anomaly detection methods still cannot meet real-time requirements in terms of processing time. Therefore, it is essential to improve the timeliness of hyperspectral anomaly detection algorithms while ensuring their accuracy.

[0004] The development of spectral imaging technology has advanced real-time hyperspectral image processing. Currently, most real-time hyperspectral detection algorithms operate on a pixel-by-pixel, line-by-line, or band-by-band basis, without considering the real-time processing requirements in practical applications. For example, in actual industrial assembly line operations, to reduce equipment costs for acquiring hyperspectral information, filters are used to simultaneously acquire data from multiple lines and bands. Existing real-time hyperspectral detection methods cannot handle this data acquisition method. Therefore, developing a real-time hyperspectral local anomaly detection algorithm based on industrial assembly line detection processing is of greater practical value. Summary of the Invention

[0005] To address the problems existing in the prior art, this invention discloses a real-time local anomaly detection method for hyperspectral images based on multi-row band processing, specifically including the following steps:

[0006] S1: Read hyperspectral image data, initialize the number of bands n determined by the number of filters, initialize the number of rows j extracted based on the bands, and initialize the number of intervals k between different bands;

[0007] S2: Based on the above n, j and k values, convert the data into data that needs to be detected and processed in real time;

[0008] S3: Establish the state equation for the correlation matrix R(n) of the spectral vectors of hyperspectral data pixels.

[0009]

[0010] in, This represents an estimate of the state at the previous moment. Let U represent the observation value at the previous time step, and U represent the observation value at the current time step. Based on the observation value U at the current time step, the estimated value at the previous time step... and the observation value at the previous moment Update the current state estimate;

[0011] S4: Update the inverse matrix R(n) of the correlation matrix R(n) using the block matrix inversion formula. -1 ;

[0012] S5: The real-time local anomaly RX operator (RTMRB-CR-RXD) of the multi-band correlation matrix is ​​used to detect hyperspectral data that needs to be detected and processed in real time, and finally the results of real-time local anomaly detection of hyperspectral images are obtained.

[0013] Furthermore, the state equation for establishing the correlation matrix R(n) is... The calculation formula is:

[0014]

[0015] in This is the metadata of all pixels in the l-th band of the m-th row of the hyperspectral data that needs to be processed in real time, r l m It refers to the data of the l-th band of a specific pixel in the m-th row of hyperspectral data that needs to be processed in real time.

[0016] Furthermore, the inverse matrix R(n) of the correlation matrix R(n) is updated using the block matrix inversion formula. -1 The calculation formula is:

[0017]

[0018]

[0019]

[0020] in, This represents the inverse of the correlation matrix at the previous time step.

[0021] Furthermore, the calculation formula for the real-time local anomaly RX operator (RTMRB-CR-RXD) of the multi-row band correlation matrix is ​​as follows:

[0022]

[0023] The detection value represents the current state. The detection value represents the state at the previous moment. This represents the sample vector from the previous time step. This represents the estimated value of the correlation matrix of the state at the previous time step. Let U represent the observed data from the previous moment, U represent the observed value at the current state, α represent Equation 3, and β represent Equation 4. This represents the sample vector at the current moment.

[0024] By employing the aforementioned technical solution, this invention provides a real-time local anomaly detection method for hyperspectral images based on multi-band processing. This method derives a new RXD recursive equation using a multi-band correlation matrix. This equation consists only of the previous state and the currently processed data sample, thus avoiding extensive redundant calculations. By replacing the correlation matrix R with MRBCM, the proposed method can handle not only real-time detected moving targets but also subtle targets that might be overlooked or obscured by RXD detection in a single operation. This invention achieves local anomaly detection in hyperspectral images while simultaneously enabling real-time processing, achieving simultaneous transmission and detection. It avoids high-dimensional data storage and redundant calculations, thus addressing the real-time performance issues of industrial flow monitoring to some extent, and exhibits good local anomaly detection results. Attached Figure Description

[0025] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0026] Figure 1 A flowchart illustrating the method provided by the present invention;

[0027] Figure 2 This is a schematic diagram of the filter-splitting ray array push-broom imaging principle in this invention;

[0028] Figures 3a-3c This is a ground map and a schematic diagram of spectral reflectance for the abu-beach-4 dataset scene in this invention;

[0029] Figures 4a-4hThis is a schematic diagram illustrating the process of local anomaly detection in the abu-beach-4 dataset in this invention.

[0030] Figures 5a-5d This is a schematic diagram of the 3D ROC curve of the local anomaly detection results of the abu-beach-4 dataset in this invention and its three corresponding 2D ROC curves. Detailed Implementation

[0031] To make the technical solutions and advantages of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention:

[0032] like Figure 1 The method for real-time local anomaly detection in hyperspectral images based on multi-band processing, as shown, specifically includes the following steps:

[0033] (1): Initialization;

[0034] Read the hyperspectral image data, initialize the number of bands n determined by the number of filters, initialize the number of rows j extracted based on the bands, and initialize the interval k between different bands, such as... Figure 2 The filter-splitting beam array pushbroom imaging principle is shown.

[0035] (2): Conversion;

[0036] Based on the above n, j, and k values, the data is converted into data that needs to be detected and processed in real time;

[0037] (3): Establish the state equation of the correlation matrix R(n) of the spectral vector of the hyperspectral data pixel points;

[0038]

[0039] in, This represents an estimate of the state at the previous moment. Let U represent the observation value at the previous time step, and U represent the observation value at the current time step. Based on the observation value U at the current time step, the estimated value at the previous time step... and the observation value at the previous moment Update the current state estimate;

[0040] (4): Update the inverse matrix R(n) of the correlation matrix R(n) using the block matrix inversion formula. -1 ;

[0041] (5): The real-time local anomaly RX operator (RTMRB-CR-RXD) of the multi-band correlation matrix is ​​used to detect hyperspectral data that needs to be detected and processed in real time, and finally the results of real-time local anomaly detection of hyperspectral images are obtained.

[0042] Real hyperspectral data experiment

[0043] The following describes the method and steps described above, using a set of publicly available real hyperspectral image datasets to test and illustrate the real-time local anomaly detection method for hyperspectral images based on multi-band processing provided by this invention, as well as to analyze and evaluate its application effects.

[0044] 1. abu-beach-4 dataset and parameter settings

[0045] This section uses a real hyperspectral image scene from the Abu-Beach-4 sensor as the experimental object. The scene studied is a hyperspectral image of the city of Pavia, Italy, acquired by the ROSIS sensor, as shown in Figure 3(a). The scene size is a 150×150 pixel vector. The original ROSIS imager can acquire 102 bands, and 8 bands of data were used for subsequent experiments. The selected representative band subset is [30, 40, 50, 60, 70, 80, 90, 100]. The real ground map is shown in Figure 3(b), and the vehicles on the bridge are the targets to be measured. The spectral reflectance is shown in Figure 3(c).

[0046] 2. Experimental Evaluation Indicators

[0047] The proposed algorithm is evaluated using 3D-ROC curves. The AUC of (P) is calculated using three two-dimensional ROC curves. D ,P F ), AUC of(P D ,τ) and AUC of(P F The effectiveness of the proposed algorithm was verified by comparing the original RXD, RT-CR-RXD, and the proposed RTMRB-CR-RXD algorithm. The 3D-ROC curve and the corresponding three 2D-ROC curves are shown in Figures 4(a)-4(d).

[0048] 3. Analysis and Evaluation of Experimental Results

[0049] The results of the experiment using a set of real hyperspectral image data provided by the present invention for real-time local anomaly detection based on multi-band processing are shown in Table 1, and the corresponding detection result process images are shown in Figures 3(a)-3(h).

[0050] Based on the real-time local anomaly detection results, the following conclusions can be drawn:

[0051] (1) The proposed method directly uses the acquired data for anomaly detection, realizing data acquisition and processing at the same time. It eliminates the need for full data storage and calculation of duplicate information, reducing processing time and the required storage space, thus providing support for real-time processing.

[0052] (2) The proposed method has the ability to detect and process local anomalies. When performing anomaly detection for each band of each row, it does not need to recalculate the correlation matrix and its inverse matrix. It only needs to use the current state and the band information of the new row to perform recursive solution, that is, to realize real-time local anomaly detection by using the autocorrelation matrix of multiple bands.

[0053] (3) For weak targets and local anomalies in the background, by continuously acquiring data from multiple rows and multiple bands, a real-time anomaly detection operator based on statistical characteristics is adopted to prevent weak targets from being overwhelmed by strong targets detected later.

[0054] (4) The recursive equation developed for real-time local anomaly detection was used to update innovation information, paving the way for the hardware implementation of industrial-scale hyperspectral image detection.

[0055] Table 1 Comparison of AUC values ​​for anomaly detection results on the abu-beach-4 dataset.

[0056]

[0057] This invention addresses the principle of multi-band imaging in filter imaging spectrometers, proposing a recursive update approach based on multi-band data acquisition. This approach is then applied to local anomaly detection, resulting in a real-time local anomaly detection operator (RTMRB-CR-RXD) based on multi-band recursion. This operator enables the filter imaging spectrometer to simultaneously acquire information from multiple bands across multiple rows while simultaneously obtaining the detection results for existing bands in the current row. Furthermore, by introducing the block matrix inversion lemma, the operator's band recursive update eliminates the need to repeatedly calculate information from existing bands when acquiring each band. Instead, the multi-band recursive update operator recursively acquires the detection results for all current bands. During algorithm execution, only the previous state information and the current pixel information need to be stored, eliminating the need to store the entire high-dimensional spectral data. This significantly reduces data storage space and avoids repetitive matrix inversion operations. Simulation experiments demonstrate that the proposed algorithm maintains high performance in local anomaly detection while enabling real-time data acquisition and processing, providing a theoretical foundation for real-time local anomaly detection systems in industrial flow monitoring. Experimental results from a set of real, publicly available hyperspectral datasets demonstrate the effectiveness of the real-time local anomaly detection method for hyperspectral images based on multi-band processing provided in this invention.

[0058] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for real-time local anomaly detection in hyperspectral images based on multi-band processing, characterized in that... include: Read the hyperspectral image data, initialize the number of bands n determined by the number of filters, initialize the number of rows j extracted based on the bands, and initialize the number of intervals k between different bands; Based on the values ​​of the number of bands n, the number of rows j, and the number of intervals k, the data is converted into data that needs to be detected and processed in real time; Establish the spectral vector correlation matrix of hyperspectral data pixels The equation of state; Update the correlation matrix using the block matrix inversion formula inverse matrix ; The real-time local anomaly RX operator of the multi-band correlation matrix is ​​used to detect hyperspectral data that needs real-time detection and processing, and finally the results of real-time local anomaly detection of hyperspectral images are obtained. The spectral vector correlation matrix The state equation is: in, This represents the estimated value of the correlation matrix of the state at the previous time step. This represents the observation value at the previous moment. The observation represents the current state, based on the observation of the current state. The correlation matrix estimate of the previous time step and the observation value at the previous moment Update the correlation matrix estimate for the current state; Establish a correlation matrix In the state equation The calculation formula is as follows: in This is the first hyperspectral data that needs to be processed in real time. The first line All image metadata for each band, This is the first hyperspectral data that needs to be processed in real time. The first row of a specific cell Data for each band; Update the correlation matrix using the block matrix inversion formula inverse matrix The calculation formula is as follows: in, Represents the correlation matrix of the previous time step. The inverse matrix.

2. The real-time local anomaly detection method for hyperspectral images based on multi-band processing according to claim 1, characterized in that: The real-time local anomaly RX operator for the multi-band correlation matrix is ​​given by the following formula: The detection value represents the current state. The detection value represents the state at the previous moment. This represents the sample vector from the previous time step. This represents the estimated value of the correlation matrix of the state at the previous time step. This represents the observation data from the previous moment. The observation representing the current state. This represents the sample vector at the current moment.

Images