A method for detecting and evaluating the effect of a concealed target by using an inside-outside window spectral unmixing mechanism

Abstract

The application discloses a kind of to adopt inside and outside window spectral unmixed mechanism's concealing target detection and effect evaluation method, comprising the following steps: step 1, establish inside and outside window detection mechanism: step 2, by VCA unsupervised algorithm respectively obtain the end member pure spectrum dictionary of inside window background area and outside window outside field area: step 3, by non-negative least square method NNLS, based on hyperspectral data and pure end member spectrum matrix, the abundance estimation matrix of each pixel is obtained: step 4, the end member spectrum dictionary expression of each pixel is obtained: step 5, abnormal target detection is carried out by detector: step 6, set detection threshold, obtain detection and analysis result.The method disclosed in the application, by designing inside and outside window detection mechanism, fully exploit the spectral characteristics of inside window background area and outside window environment area, and by the designed target detector, the purpose of accurate abnormal target detection is realized.

Description

Technical Field

[0001] This invention belongs to the technical fields of hyperspectral Earth remote sensing observation, occlusion target survey and assessment, and target characteristic analysis, and specifically relates to a method for occlusion target detection and effect assessment using an inner and outer window spectral unmixing mechanism. Background Technology

[0002] Hyperspectral remote sensing organically combines spectral and imaging technologies, enabling it to detect not only the two-dimensional geometric spatial information of targets but also the one-dimensional spectral information of targets. This allows for the detection of ground features that were previously undetectable by multispectral remote sensing.

[0003] Hyperspectral remote sensing offers superior spectral resolution compared to conventional remote sensing, enabling the acquisition of images across continuous bands in the visible to near-infrared region. Each pixel in the image can form a smooth spectral curve containing characteristic information. Hyperspectral images not only provide information about the spatial and geometric relationships of the target but also reveal its spectral characteristics. This feature makes it highly suitable for detecting ground-covered targets and evaluating their camouflage effectiveness.

[0004] Target detection using hyperspectral remote sensing images has gradually become an important tool for battlefield reconnaissance and surveying, and spectral-based target detection technology is receiving increasing attention. Currently, hyperspectral imaging technology has been successfully applied in remote sensing and aerospace military reconnaissance. Hyperspectral imagers can simultaneously image the same target across continuous spectral bands, directly reflecting the spectral characteristics of the observed object and even the composition of its surface material. Research shows that hyperspectral data can reveal a precise correlation between space-based detection information and actual ground targets. By analyzing the detected spectral characteristic curves, the material endmember spectrum and abundance estimate of each pixel can be calculated, thereby distinguishing between natural backgrounds and military targets and determining the nature and type of the target. Summary of the Invention

[0005] The technical problem to be solved by this invention is to overcome the shortcomings of existing algorithms that detect abnormal targets by means of the difference between the target and the background distribution, such as the high false alarm rate and the inaccuracy of manually labeled pure spectral training data in supervised deep learning methods. This invention provides a method for masked target detection and effect evaluation that adopts an inner and outer window spectral unmixing mechanism.

[0006] The present invention adopts the following technical solution:

[0007] An improved method for masked target detection and performance evaluation employing an inner and outer window spectral demixing mechanism includes the following steps:

[0008] Step 1, establish an internal and external window detection mechanism:

[0009] Step 11: Initialize the size of the inner and outer windows. By inverting the size of the inner window area using a mask, delete any abnormal target pixels and obtain the background area data cube of the size of the outer window.

[0010] Step 12: Obtain the environmental area data cube representing the size of the outer window;

[0011] Step 13: Combine the edge region padding 0 method in deep learning to complete the data filling of the edge region, set the step size, slide the detection box of the outer window size until the entire image is searched, and establish an unsupervised inner and outer window detection mechanism.

[0012] Step 2: Obtain the endmember pure spectral dictionaries for the inner window background region and the outer window outer field region using the VCA unsupervised algorithm.

[0013] Step 21: Using the endmember extraction algorithm, set the number of endmembers to be extracted, and extract the pure spectral dictionary E of the background region after mask inversion. 背景 ;

[0014] Step 22: Using the endmember extraction algorithm, set the number of endmembers to be extracted, and extract the pure spectral dictionary E of the environmental region. 环境 ;

[0015] Step 3: Using nonnegative least squares (NNLS), obtain the abundance estimation matrix for each pixel based on the hyperspectral data and the pure endmember spectral matrix.

[0016] Step 31: Using a spectral unmixing algorithm, calculate the abundance estimation matrix A of each pixel in the sliding window region relative to the background pure spectral matrix, based on the masked background region and the corresponding pure spectral matrix. 背景 ;

[0017] Step 32: Using a spectral unmixing algorithm, calculate the abundance estimation matrix A of each pixel in the sliding window region relative to the environmental pure spectral matrix, based on the environmental region and the corresponding pure spectral matrix. 环境 ;

[0018] Step 4: Obtain the endmember spectral dictionary representation for each pixel:

[0019] Step 41, based on the obtained background region abundance estimation matrix A 背景 and the corresponding pure spectrum dictionary E 背景 The endmember spectral dictionary representation D of each pixel in the sliding window region is calculated. 背景重组 ;

[0020] Step 42, based on the obtained environmental area abundance estimation matrix A 环境 and the corresponding pure spectrum dictionary E 环境The endmember spectral dictionary representation D of each pixel in the sliding window region is calculated. 环境重组 ;

[0021] Step 5: Construct a detector to detect abnormal targets:

[0022] Step 51: Multiply the endmember spectral dictionary representation and abundance estimation matrix corresponding to each pixel in the background and environment regions to obtain the endmember vector of that pixel;

[0023] D 背景重组i,j =E 背景 *A 背景i,j

[0024] H 环境重组i,j =E 环境 *A 环境i,j

[0025] In the above formula, i represents the i-th row and j represents the j-th column;

[0026] Step 52: Subtract the reflectance of the corresponding pixel in the original preprocessed hyperspectral data from the endmember vectors of the background and environment regions to obtain the difference in expressed noise.

[0027]

[0028] In the above formula, x i,j This represents the feature representation of the element in the i-th row and j-th column of the original data; β represents the adjustment factor.

[0029] Step 6: Set the detection threshold and obtain the detection and analysis results.

[0030] Step 61: Perform unsupervised abnormal target detection on the entire image to obtain detection results with a value range of 0 to 1. Set the segmentation threshold to obtain a segmented binary image of the target and background.

[0031] Step 62: By using different feature segmentation thresholds, the correspondence between camouflage level and feature segmentation threshold is obtained, and then the camouflage level analysis results of abnormally occluded targets are obtained.

[0032] Furthermore, in step 11, the inner and outer window sizes are initialized to odd numbers.

[0033] Furthermore, in steps 21 and 22, the endmember extraction algorithm includes the VCA vertex component analysis algorithm.

[0034] The beneficial effects of this invention are:

[0035] The method disclosed in this invention, by designing an inner and outer window detection mechanism, fully explores the spectral characteristics of the background region of the inner window and the environmental region of the outer window, and uses a designed target detector to express each pixel in a dictionary, thereby achieving the goal of accurate abnormal target detection. It solves the shortcomings of traditional algorithms that detect abnormal targets by the difference between the target and the background, such as the high false alarm rate, and the shortcomings of supervised deep learning methods, such as the inaccuracy of manually labeled pure spectral training data and the high manual cost. Compared with traditional methods, the results are more accurate and scientific. Attached Figure Description

[0036] Figure 1 This is a flowchart illustrating the method of the present invention;

[0037] Figure 2 This is a detailed flowchart illustrating the process of the method of the present invention. Detailed Implementation

[0038] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0039] Example 1 discloses a method for detecting and evaluating the effectiveness of masked targets using an inner and outer window spectral unmixing mechanism. This method draws on the idea of ​​the RX detector to establish an inner and outer window spectral unmixing mechanism. It achieves the detection of abnormal targets by recursively accessing the entire image. Through parameter adjustment, the detection results are made more accurate and scientific. Figure 1 As shown, it includes the following steps:

[0040] Step 1, establish an internal and external window detection mechanism:

[0041] Step 11: Initialize the size of the inner and outer windows. The inner window serves as a target detection buffer. By inverting the size of the inner window area through a mask, any abnormal target pixels that may exist are deleted, and the background area data cube of the size of the outer window is obtained.

[0042] Step 12: The outer window is used as a mixed area containing the target and the background. The environmental area data cube with the size of the outer window is obtained.

[0043] Step 13: Combine the edge region padding 0 method in deep learning to complete the data filling of the edge region, set the step size, slide the detection box of the outer window size until the entire image is searched, and establish an unsupervised inner and outer window detection mechanism.

[0044] Step 2: Obtain the endmember pure spectral dictionaries for the inner window background region and the outer window outer field region using the VCA unsupervised algorithm.

[0045] Step 21: Using an endmember extraction algorithm, including but not limited to the VCA vertex component analysis algorithm, set the number of endmembers to be extracted, and extract the pure spectral dictionary E of the background region after mask inversion. 背景 ;

[0046] Step 22: Using an endmember extraction algorithm, including but not limited to the VCA vertex component analysis algorithm, set the number of endmembers to be extracted and extract the pure spectral dictionary E of the environmental region. 环境 ;

[0047] Step 3: Using nonnegative least squares (NNLS), obtain the abundance estimation matrix for each pixel based on the hyperspectral data and the pure endmember spectral matrix.

[0048] Step 31: Using spectral unmixing algorithms, including but not limited to NNLS nonnegative least squares, calculate the abundance estimation matrix A of each pixel in the sliding window region relative to the background pure spectral matrix, based on the masked background region and the corresponding pure spectral matrix. 背景 ;

[0049] Step 32: Using spectral unmixing algorithms, including but not limited to NNLS nonnegative least squares, calculate the abundance estimation matrix A of each pixel in the sliding window region relative to the environmental pure spectral matrix, based on the environmental region and the corresponding pure spectral matrix. 环境 ;

[0050] Step 4: Obtain the endmember spectral dictionary representation for each pixel:

[0051] Step 41, based on the obtained background region abundance estimation matrix A 背景 and the corresponding pure spectrum dictionary E 背景 The endmember spectral dictionary representation D of each pixel in the sliding window region is calculated. 背景重组 ;

[0052] Step 42, based on the obtained environmental area abundance estimation matrix A 环境 and the corresponding pure spectrum dictionary E 环境 The endmember spectral dictionary representation D of each pixel in the sliding window region is calculated. 环境重组 ;

[0053] Step 5: Construct a detector to detect abnormal targets:

[0054] Step 51: Multiply the endmember spectral dictionary representation and abundance estimation matrix corresponding to each pixel in the background and environment regions to obtain the endmember vector of that pixel;

[0055] D 背景重组i,j =E 背景 *A 背景i,j

[0056] H 环境重组i,j =E 环境 *A 环境i,j

[0057] In the above formula, i represents the i-th row and j represents the j-th column;

[0058] Step 52: Subtract the reflectance of the corresponding pixel in the original preprocessed hyperspectral data from the endmember vectors of the background and environment regions to obtain the difference in expressed noise.

[0059]

[0060] In the above formula, x i,j This represents the feature representation of the element in the i-th row and j-th column of the original data; β represents the adjustment factor, used to control the calculation result within the identifiable range.

[0061] Step 6: Set the detection threshold and obtain the detection and analysis results.

[0062] Step 61: Perform unsupervised abnormal target detection on the entire image to obtain detection results with a value range of 0 to 1. Set the segmentation threshold to obtain a segmented binary image of the target and background.

[0063] Step 62: By using different feature segmentation thresholds, the correspondence between camouflage level and feature segmentation threshold is obtained, and then the camouflage level analysis results of abnormally occluded targets are obtained.

[0064] To facilitate understanding, the method of the present invention will be further described in detail below, such as... Figure 2 As shown, the specific steps include the following:

[0065] Step 201: Establish an inner and outer window detection mechanism. The specific implementation process is as follows:

[0066] (11): Drawing on the idea of ​​RX anomaly detection, the size of the inner and outer windows is initialized and set to an odd number. The inner window is used as a target detection buffer. The inner window size region is multiplied with the original outer window region data matrix by "mask inversion" to obtain the background region matrix with all zeros in the inner window data, and the background region data cube of the outer window size is obtained.

[0067] (12): Initialize the size of the inner and outer windows and set them to odd numbers. The outer window is used as a mixed area containing the target and the background. Obtain the environmental area data cube of the outer window size.

[0068] (13): Combining the edge region padding 0 method in deep learning, the edge region data is filled, the step size is set, and the detection box of the outer window size is slid until the entire image is searched, thus establishing an unsupervised inner and outer window detection mechanism.

[0069] Step 202: Obtain the pure spectral matrix corresponding to the inner and outer windows. The specific implementation process is as follows:

[0070] (21): Using endmember extraction algorithms, including but not limited to VCA vertex component analysis, the number of endmembers to be extracted is set to extract the pure spectral matrix E of the background region after "mask inversion". 背景 ;

[0071] (22): Using endmember extraction algorithms, including but not limited to VCA vertex component analysis, the number of endmembers to be extracted is set to extract the pure spectral matrix E of the environmental region. 环境 .

[0072] Step 203: Calculate the abundance estimate of the pure spectrum corresponding to the background region. The specific implementation process is as follows:

[0073] (31): Using spectral unmixing algorithms, including but not limited to NNLS nonnegative least squares, the abundance estimation matrix A of each pixel in the sliding window region relative to the background pure spectral matrix is ​​calculated based on the background region data and the corresponding pure spectral matrix obtained in (21). 背景 ;

[0074] Step 204: Calculate the abundance estimate of the pure spectrum corresponding to the environmental region. The specific implementation process is as follows:

[0075] (41): Using spectral unmixing algorithms, including but not limited to NNLS nonnegative least squares, the abundance estimation matrix A of each pixel in the sliding window region relative to the environmental pure spectral matrix is ​​calculated based on the environmental region and the corresponding pure spectral matrix obtained in (22). 环境 .

[0076] Step 205: Calculate the dictionary representation of each pixel in the background region. The specific implementation process is as follows:

[0077] (51): Based on the obtained background region abundance estimation matrix A 背景 and the corresponding pure spectrum dictionary E 背景 The endmember spectral dictionary representation D of each pixel in the sliding window region is calculated. 背景重组 ;

[0078] (52): Multiply the endmember dictionary representation and abundance estimation matrix of each pixel in the background region, and obtain the endmember vector of the pixel according to the following formula.

[0079] D 背景重组i,j =E 背景 *A 背景i,j

[0080] Step 206: Calculate the dictionary representation of each pixel in the environment region. The specific implementation process is as follows:

[0081] (61): Based on the obtained environmental area abundance estimation matrix A 环境 and the corresponding pure spectrum dictionary E 环境 The endmember spectral dictionary representation D of each pixel in the sliding window region is calculated. 环境重组 ;

[0082] (62): Multiply the endmember dictionary representation and abundance estimation matrix of each pixel in the environment region, and obtain the endmember vector of the pixel according to the following formula.

[0083] H 环境重组i,j =E 环境 *A 环境i,j

[0084] Step 207: Construct an anomaly target detector. The specific implementation process is as follows:

[0085] (71): The difference between the reflectance of the corresponding pixel in the original preprocessed hyperspectral data and the endmember vectors of the background and environment regions is obtained to obtain the difference in the expression noise;

[0086]

[0087] (72): Perform unsupervised abnormal target detection on the whole image to obtain detection results with a value range of 0 to 1. Set a segmentation threshold to obtain a segmented binary image of the target and the background.

[0088] Step 208: Obtain the detection and analysis results. The specific implementation process is as follows:

[0089] (81): By using different feature segmentation thresholds, the correspondence between camouflage level and feature segmentation threshold is obtained, and the camouflage level analysis results of abnormal targets are obtained.

Claims

1. A method for detecting and evaluating the effectiveness of masked targets using an inner and outer window spectral unmixing mechanism, characterized in that, Includes the following steps: Step 1, establish an internal and external window detection mechanism: Step 11: Initialize the size of the inner and outer windows. By inverting the size of the inner window area using a mask, delete any abnormal target pixels and obtain the background area data cube of the size of the outer window. Step 12: Obtain the environmental area data cube representing the size of the outer window; Step 13, combining edge regions in deep learning In the 0 mode, data filling of the edge region is completed. The step size is set, and the detection box of the outer window size is slid until the entire image is searched, thus establishing an unsupervised inner and outer window detection mechanism. Step 2: Obtain the endmember pure spectral dictionaries for the inner window background region and the outer window outer field region using the VCA unsupervised algorithm. Step 21: Using an endmember extraction algorithm, set the number of endmembers to be extracted, and extract the pure spectral dictionary of the background region after mask inversion. ; Step 22: Using an endmember extraction algorithm, set the number of endmembers to be extracted, and extract the pure spectral dictionary of the environmental region. ; Step 3: Using nonnegative least squares (NNLS), obtain the abundance estimation matrix for each pixel based on the hyperspectral data and the pure endmember spectral matrix. Step 31: Using a spectral unmixing algorithm, calculate the abundance estimation matrix of each pixel in the sliding window region relative to the background pure spectral matrix, based on the masked background region and the corresponding pure spectral matrix. ; Step 32: Using a spectral unmixing algorithm, based on the environmental region and the corresponding pure spectral matrix, calculate the abundance estimation matrix of each pixel in the sliding window region relative to the environmental pure spectral matrix. ; Step 4: Obtain the endmember spectral dictionary representation for each pixel: Step 41, based on the obtained background region abundance estimation matrix and corresponding pure spectrum dictionary The endmember spectral dictionary representation of each pixel in the sliding window region is calculated. ; Step 42, based on the obtained environmental area abundance estimation matrix and corresponding pure spectrum dictionary The endmember spectral dictionary representation of each pixel in the sliding window region is calculated. ; Step 5: Construct a detector to detect abnormal targets: Step 51: Multiply the endmember spectral dictionary representation and abundance estimation matrix corresponding to each pixel in the background and environment regions to obtain the endmember vector of that pixel; In the above formula, Indicates the first OK, Indicates the first List; Step 52: Subtract the reflectance of the corresponding pixel in the original preprocessed hyperspectral data from the endmember vectors of the background and environment regions to obtain the difference in expressed noise. In the above formula, Indicates the first in the original data line, number Characteristic representation of column elements; Indicates the adjustment factor; Step 6: Set the detection threshold and obtain the detection and analysis results. Step 61: Perform unsupervised abnormal target detection on the entire image to obtain detection results with a value range of 0 to 1. Set the segmentation threshold to obtain a segmented binary image of the target and background. Step 62: By using different feature segmentation thresholds, the correspondence between camouflage level and feature segmentation threshold is obtained, and then the camouflage level analysis results of abnormally occluded targets are obtained.

2. The method for masking target detection and effect evaluation using an inner and outer window spectral demixing mechanism according to claim 1, characterized in that: In step 11, the inner and outer window sizes are initialized to odd numbers.

3. The method for masking target detection and effect evaluation using an inner and outer window spectral demixing mechanism according to claim 1, characterized in that: In steps 21 and 22, the endmember extraction algorithm includes the VCA vertex component analysis algorithm.

Images