A target recognition method based on UAV hyperspectral images

Abstract

This invention belongs to the field of image processing technology and discloses a target recognition method based on UAV hyperspectral images. The method includes: S1, acquisition and preprocessing of camouflage material spectra; S2, acquisition and preprocessing of airborne hyperspectral data; and S3, selection of a pure background dictionary A based on spatial spectral features. b S4. Reconstruct the spectrum of the pixel to be tested using the background dictionary, and use the reconstruction error to determine whether the pixel belongs to the target. This invention establishes constraints for selecting the background dictionary based on the homogeneity of the image and the Euclidean distance from the laboratory spectrum. It fully utilizes the spatial-spectral characteristics of airborne hyperspectral images to select pure background pixels for sparse representation of the background dictionary, which can improve the accuracy of target detection by the sparse representation method. Based on the reconstruction error of the spectrum of the pixel to be tested, the spatial location of the target disguised by the camouflage net can be accurately detected, thereby solving the technical problem of inaccurate detection of camouflage targets.

Description

Technical Field

[0001] This invention relates to the field of image processing technology, and more specifically to a target recognition method based on hyperspectral images from unmanned aerial vehicles (UAVs). Background Technology

[0002] Hyperspectral target recognition is achieved by analyzing the spectral differences between the background and the target. Currently, for targets that exhibit "different but similar spectra" with the background, existing hyperspectral target detection methods include the Reed-XiaoLi algorithm (RX), the Constrained Energy Minimization (CEM) algorithm, the Matched Filtering (SMF) algorithm, Orthogonal Subspace Projection (OSP) using a subspace model, and sparse representation. However, existing algorithms such as CEM, RX, SMF, and OSP cannot accurately determine the number and location of camouflaged targets, especially in jungles and deserts where camouflage nets are used. Sparse representation, compared to the other methods, delves deeper into the differences between the target and background spectra by constructing a dictionary and reconstructing the pixel spectra using the basis vectors within the dictionary.

[0003] The purity of the dictionary in a sparse representation model affects the accuracy of target recognition. Currently, the construction of background dictionaries typically only utilizes the spectral information of hyperspectral images, with limited application of spatial information. The most common background dictionary method involves visually selecting some targets and using a sliding dual-window approach to select the pixels of the outer window of the targets as basis vectors in the background dictionary. However, this method is not suitable for targets that have low visual difference from the background.

[0004] Existing target recognition technologies can effectively identify typical ground features, but for camouflaged targets with small spectral and spatial differences, they cannot fully extract the spatial-spectral information in the image, resulting in inaccurate detection of camouflaged targets. Therefore, this paper proposes a target recognition method to address the problem of identifying camouflage materials. Summary of the Invention

[0005] The present invention aims to provide a target recognition method based on UAV hyperspectral images to solve the technical problem in the prior art that when extracting camouflaged targets in hyperspectral images using sparse representation algorithms, the selection of the background dictionary is affected by the "different objects with the same spectrum" of the camouflaged targets, resulting in the inability to accurately detect camouflaged targets.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] A target recognition method based on UAV hyperspectral images includes the following steps:

[0008] S1. Acquisition and preprocessing of the spectrum of camouflage materials;

[0009] S2. Acquisition and preprocessing of airborne hyperspectral data;

[0010] S3. Selecting a clean background dictionary based on spatial spectral features ;

[0011] S4. Reconstruct the spectrum of the pixel to be measured using the background dictionary, and use the reconstruction error to determine whether the pixel to be measured belongs to the target.

[0012] Furthermore, the spectrum of the reconstructed pixel is given by the following formula:

[0013]

[0014] In the formula, y represents the spectrum of image pixels. Represents the background dictionary. This represents a sparse vector.

[0015] Furthermore, the specific steps for selecting the background dictionary are as follows:

[0016] S31. Calculate the Euclidean distance between the laboratory spectrum and the spectrum of each pixel in the image;

[0017] S32. Calculate the homogeneity value of each pixel in the preprocessed airborne hyperspectral image;

[0018] S33. Establish a background dictionary, select constraints, and obtain the background dictionary.

[0019] Furthermore, the formula for calculating the Euclidean distance between the laboratory spectrum and the spectrum of each pixel in the image is as follows:

[0020]

[0021] In the formula This represents the Euclidean distance between a pixel in the image and the prior spectrum. Represents the pixels in the image at the th Values ​​for each band, Indicates the prior spectrum of the first The value for each band, where k represents the number of bands.

[0022] Furthermore, the formula for calculating the homogeneity value of each pixel is as follows:

[0023]

[0024] In the formula Representing the The mean value of pixels within an n×n window across each band. Represents the nth element within an n×n sliding window. The value of the u-th pixel in the i-th band This represents the homogeneity value of a pixel.

[0025] Furthermore, the specific constraints for selecting the background dictionary are as follows: ,in The spectrum of pixels with the largest Euclidean distance values ​​from the camouflage net measured in the laboratory is taken as 85% of the total. value, Cells representing high homogeneity are represented by a cumulative percentage of 15%. value.

[0026] Furthermore, the specific method for determining whether a pixel belongs to the target using reconstruction error involves first using the acquired background dictionary. Determine the optimal coefficients of the sparse vectors of the background dictionary. When the reconstruction coefficients are taken as the optimal coefficients At that time, the output value of the SRSS model is compared with the set threshold β to determine whether it belongs to the target.

[0027] Furthermore, the optimal coefficients of the sparse vector The formula is as follows:

[0028]

[0029] In the formula This is the optimal sparse vector. express The 0 norm, i.e., the number of non-zero elements, and L, which is the upper limit of the number of sparse coefficients, are used to solve for the sparse vector using the orthogonal matching pursuit algorithm. .

[0030] Furthermore, the formula for the output value of the SRSS model is as follows:

[0031]

[0032] In the formula The representative model output value is determined by setting a threshold β when the detector... If the output value is greater than β, it means that the target exists; otherwise, the target does not exist.

[0033] In summary, the present invention has the following beneficial effects:

[0034] This invention establishes background dictionary selection constraints based on the homogeneity of the image and the Euclidean distance from the laboratory spectrum. It fully utilizes the spatial-spectral features of airborne hyperspectral images to select pure background pixels for sparse representation of the background dictionary, which can improve the accuracy of target detection by the sparse representation method. Based on the reconstruction error of the spectrum of the pixel to be detected, the spatial location of the target disguised by the camouflage net can be accurately detected. Attached Figure Description

[0035] Figure 1This is a flowchart illustrating the selection process for the background dictionary in this invention. Detailed Implementation

[0036] The present invention will now be described in further detail with reference to the accompanying drawings and embodiments:

[0037] The present invention provides a target recognition method based on UAV hyperspectral images, comprising the following steps:

[0038] Acquisition and preprocessing of spectra of camouflage materials;

[0039] Spectral measurements were performed on grassland camouflage materials and desert camouflage materials. The obtained spectral curves were smoothed by SG and resampled to obtain spectral data of the two camouflage materials in the range of 400-1000 nm.

[0040] Acquisition and preprocessing of airborne hyperspectral data;

[0041] Airborne hyperspectral data were collected from grassland camouflage targets deployed in grassland camouflage environments and desert camouflage targets deployed in desert camouflage environments. After SG smoothing and spectral resampling, hyperspectral images with a spectral range of 400-1000 nm and 150 bands were obtained.

[0042] When using an overcomplete dictionary to sparsely represent a hyperspectral image, the spectral vector of each pixel in the image can be reconstructed using a small number of atoms in the overcomplete target or background dictionary. However, due to the high spectral similarity between the camouflaged target and the background, selecting camouflaged target pixels in the image is extremely difficult, and the prior number of target spectra cannot meet the requirements of the target dictionary. Therefore, this invention uses a background dictionary. The spectrum of the pixel to be measured is reconstructed as shown in formula (1).

[0043] (1)

[0044] In the formula, y represents the spectrum of image pixels. Represents the background dictionary. This represents a sparse vector.

[0045] Regarding formula (1), this paper proposes a sparse representation of background dictionary using spatial and spectral features (SRSS). By establishing background dictionary selection constraints based on the homogeneity of the image and the Euclidean distance from the laboratory spectrum, the SRSS method fully utilizes the spatial and spectral features of airborne hyperspectral images to select pure background pixels for sparse representation of the background dictionary. The reconstruction error is used to determine whether the pixel to be measured belongs to the target.

[0046] Background Dictionary The construction process is as follows Figure 1 As shown. Among them. Representing row P A complete dictionary of columns, where P represents the number of bands in the spectrum. This represents the number of background dictionaries.

[0047] (1) First, calculate the Euclidean distance between the target spectrum measured in the laboratory and the spectrum of each pixel in the image to determine the difference between the pixel spectrum of the image and the target spectrum measured in the laboratory. The formula is shown in Equation (2).

[0048] (2)

[0049] In the formula This represents the Euclidean distance between a pixel in the image and the prior spectrum. Represents the pixels in the image at the th Values ​​for each band, Indicates the prior spectrum of the first The value for each band, where k represents the number of bands.

[0050] Secondly, an n×n sliding window (n is an odd number, and a 5×5 sliding window is chosen in this paper) is set for the image, and the homogeneity value of each pixel is calculated. The smaller the homogeneity value, the higher the homogeneity of the pixel, and the greater the probability that it belongs to the same land feature as the surrounding pixels. The calculation is shown in formula (3).

[0051] (3)

[0052] In the formula Representing the The mean value of pixels within an n×n window across each band. Represents the nth element within an n×n sliding window. The value of the u-th pixel in the i-th band This represents the homogeneity value of a pixel.

[0053] (2) Establish background dictionary selection constraints. ,in The spectrum of pixels with the largest Euclidean distance values ​​from the camouflage net measured in the laboratory is taken as 85% of the total. value. Cells representing high homogeneity are represented by a cumulative percentage of 15%. value.

[0054] Then, using the obtained background dictionary Solving formula (4) yields the optimal coefficients of the sparse vectors of the background dictionary. .

[0055] (4)

[0056] In the formula This is the optimal sparse vector. express The 0 norm, i.e., the number of non-zero elements, and L, which is the upper limit of the number of sparse coefficients, are used to solve for the sparse vector using the orthogonal matching pursuit algorithm. When the reconstructed coefficients are obtained as the optimal coefficients. When the error is obtained, the magnitude of the background dictionary reconstruction pixel spectral vector error can be obtained through formula (5), which is the output value of the model SRSS.

[0057] (5)

[0058] In the formula The representative model output value is determined by setting a threshold β when the detector... If the output value is greater than β, it means that the target exists; otherwise, the target does not exist.

[0059] The above description is merely an embodiment of the present invention, and common knowledge such as specific technical solutions or characteristics in the solution is not described in detail here. It should be noted that those skilled in the art can make several modifications and improvements without departing from the technical solution of the present invention, and these should also be considered within the scope of protection of the present invention. These modifications and improvements will not affect the effectiveness of the implementation of the present invention or the practicality of the patent. The scope of protection claimed in this application shall be determined by the content of its claims, and the specific embodiments described in the specification can be used to interpret the content of the claims.

Claims

1. A target recognition method based on UAV hyperspectral images, characterized in that, Includes the following steps: S1. Acquisition and preprocessing of the spectrum of camouflage materials; S2. Acquisition and preprocessing of airborne hyperspectral data; S3. Selecting a clean background dictionary based on spatial spectral features ; Selecting a clean background dictionary The specific steps are as follows: S31. Calculate the Euclidean distance between the laboratory spectrum and the spectrum of each pixel in the image; S32. Calculate the homogeneity value of each pixel in the preprocessed airborne hyperspectral image; S33. Establish a background dictionary, select constraints, and obtain the background dictionary; S4. Reconstruct the spectrum of the pixel to be measured using the background dictionary, and use the reconstruction error to determine whether the pixel to be measured belongs to the target.

2. The target recognition method based on UAV hyperspectral images according to claim 1, characterized in that, The spectrum of the reconstructed pixel is given by the following formula: ； In the formula, y represents the spectrum of image pixels. Represents the background dictionary. This represents a sparse vector.

3. The target recognition method based on UAV hyperspectral images according to claim 1, characterized in that, The formula for calculating the Euclidean distance between the laboratory spectrum and the spectrum of each pixel in the image is as follows: ； In the formula This represents the Euclidean distance between a pixel in the image and the prior spectrum. Represents the pixels in the image at the th Values ​​for each band, Indicates the prior spectrum of the first The value for each band, where k represents the number of bands.

4. The target recognition method based on UAV hyperspectral images according to claim 1, characterized in that, The formula for calculating the homogeneity value of each pixel is as follows: ； In the formula Representing the The mean value of pixels within an n×n window across each band. Represents the nth element within an n×n sliding window. The value of the u-th pixel in the i-th band This represents the homogeneity value of a pixel.

5. The target recognition method based on UAV hyperspectral images according to claim 1, characterized in that, The specific constraints for selecting the background dictionary are as follows: ,in The spectrum of pixels with the largest Euclidean distance values ​​from the camouflage net measured in the laboratory is taken as 85% of the total. value, Cells representing high homogeneity are represented by a cumulative percentage of 15%. value.

6. The target recognition method based on UAV hyperspectral images according to claim 1, characterized in that, The specific method for determining whether a pixel belongs to the target using reconstruction error is to first use the acquired background dictionary. Determine the optimal coefficients of the sparse vectors of the background dictionary. When the reconstructed coefficients are the optimal coefficients At that time, the output value of the SRSS model is compared with the set threshold β to determine whether it belongs to the target.

7. The target recognition method based on UAV hyperspectral images according to claim 6, characterized in that, The optimal coefficients of the sparse vector The formula is as follows: ； In the formula This is the optimal sparse vector. express The 0 norm, i.e., the number of non-zero elements, and L, which is the upper limit of the number of sparse coefficients, are used to solve for the sparse vector using the orthogonal matching pursuit algorithm. .

8. A target recognition method based on UAV hyperspectral images according to claim 6 or 7, characterized in that, The formula for the output value of the SRSS model is as follows: ； In the formula The representative model output value is determined by setting a threshold β when the detector... If the output value is greater than β, it means that the target exists; otherwise, the target does not exist.

Images