A method for UAV spectral recognition based on sparse reconstruction dual-domain joint matching

Abstract

This invention discloses a UAV spectral recognition method based on sparse reconstruction dual-domain joint matching, belonging to the field of target recognition and spectral imaging technology. The method includes employing a computational spectral imaging system; parallel processing of dual-domain data; construction of a dual-domain feature spectral library; and dual-domain joint matching and decision fusion. This invention addresses the problem that existing technologies inevitably produce artifacts, noise, and information distortion in the reconstructed spectral data. These distortions directly reduce the accuracy of subsequent recognition algorithms and fail to deeply integrate the sensing and recognition processes. This invention deeply integrates the recognition process with the sensing reconstruction process of computational spectral imaging. By performing dual verification in the reconstruction domain and sparse domain, it effectively overcomes the shortcomings of single-reconstruction-domain recognition methods, which are susceptible to artifacts and noise from the reconstruction algorithm, significantly improving the accuracy and robustness of UAV recognition.

Description

Technical Field

[0001] This invention relates to the field of unmanned aerial vehicle (UAV) technology, specifically to a UAV spectral recognition method based on sparse reconstruction dual-domain joint matching. Background Technology

[0002] In recent years, unmanned aerial vehicles (UAVs) characterized by being "low, slow, and small" have developed rapidly. Therefore, the effective detection and identification of UAVs has become an urgent technical problem to be solved.

[0003] Existing anti-drone technologies mainly include radar detection, visible light detection, and acoustic detection. Radar is not very effective at detecting drones with low radar cross section (RCS) and is easily affected by ground clutter; visible light cameras have difficulty distinguishing between drones and birds at long distances and are easily affected by factors such as lighting, weather (fog, haze), and camouflage; acoustic detection has a short range and a low signal-to-noise ratio in noisy urban environments.

[0004] As an emerging technology, hyperspectral imaging can acquire detailed spectral information of a target, forming a unique "spectral fingerprint," thereby identifying the target based on its material and providing a new solution for UAV detection. However, traditional hyperspectral imagers, such as pushbroom devices, require scanning imaging, making it difficult to capture fast-moving, maneuvering UAV targets. While snapshot hyperspectral imagers can provide instantaneous imaging, their development faces the challenge of the mutual constraint between spatial resolution and spectral resolution.

[0005] Computational spectral imaging technology, especially snapshot spectral imaging systems based on compressed sensing, uses a single two-dimensional detector to perform a snapshot measurement by optically encoding and compressing a three-dimensional hyperspectral data cube, and then reconstructs the complete spectral image through a reconstruction algorithm. Current methods typically process the reconstructed spectral image using traditional target recognition algorithms. This "reconstruct first, recognize later" sequential processing mode has fundamental flaws: First, compressed sensing reconstruction algorithms are not perfect, and the reconstructed spectral data inevitably contains artifacts, noise, and information distortion, which directly reduce the accuracy of subsequent recognition algorithms. Second, this mode completely ignores the compressed measurement values ​​and sparse domain coefficients, which contain rich feature information generated during the sensing process, failing to deeply integrate the sensing and recognition processes.

[0006] Therefore, developing a new UAV detection method that can be deeply coupled with the computational spectral sensing process, resist reconstruction errors, and achieve high-precision and high-robust identification has important theoretical significance and application value. Summary of the Invention

[0007] The purpose of this invention is to provide a UAV spectral recognition method based on sparse reconstruction dual-domain joint matching, which deeply integrates the recognition process with the sensor reconstruction process of computational spectral imaging. By performing dual verification in the reconstruction domain and sparse domain, the accuracy and robustness of UAV recognition are significantly improved, thus solving the problems mentioned in the background art.

[0008] To achieve the above objectives, the present invention provides the following technical solution: a UAV spectral recognition method based on sparse reconstruction dual-domain joint matching, comprising:

[0009] A computational spectral imaging system is used to perform optical encoding compression sampling on the scene under test, which includes the UAV, to obtain a single frame of two-dimensional compressed spectral measurement value.

[0010] Dual-domain parallel data processing: Dual-domain parallel processing is performed on the acquired single-frame two-dimensional compressed spectral measurement values;

[0011] Constructing a dual-domain feature spectral library: Pre-construct a feature library containing standard samples of UAVs and background ground features;

[0012] Dual-domain joint matching and decision fusion: Perform dual-domain joint matching on the pixels to be identified in the hyperspectral data cube;

[0013] Spatial clustering and target confirmation: Spatial clustering analysis is performed on all points identified as drone pixels to ultimately confirm and output the location of the drone target.

[0014] Preferably, the step of performing optical encoding compression sampling on the test scene including the UAV to obtain a single-frame two-dimensional compressed spectral measurement value specifically includes:

[0015] The light emitted by the target object S passes through a variable resolution optical system and is incident on a pixel-level random filter array;

[0016] The filter array randomly modulates the spectral information at different spatial locations, and the modulated light signal is sensed by the detector array to form a single frame of two-dimensional compressed spectral measurement image.

[0017] The 3D hyperspectral data cube of the scene under test is vectorized into a one-dimensional column vector, as shown in the following expression:

[0018]

[0019]

[0020] in, This is the one-dimensional column vector obtained after vectorizing the three-dimensional hyperspectral data cube. This represents the dimension of the one-dimensional column vector after vectorization. The number of bands in the spectral dimension. The number of image rows in the spatial dimension. The number of image columns in the spatial dimension;

[0021] The compressed measurement values ​​obtained from the two-dimensional detector are vectorized into a one-dimensional column vector. ,in The imaging process can then be represented by the following formula:

[0022]

[0023]

[0024] in, It is a spectral signal. This is a one-dimensional column vector derived from the vectorized compressed measurement values ​​obtained on the two-dimensional detector. For the sensing matrix, This represents a one-dimensional column vector after the compressed measurement values ​​have been vectorized. Dimensions This is additive noise in the system.

[0025] Preferably, the dual-domain parallel processing of the acquired single-frame two-dimensional compressed spectral measurement values ​​specifically includes:

[0026] After obtaining the vectorized one-dimensional column vector of the compressed measurement values ​​obtained from the two-dimensional detector, the vectorized one-dimensional column vector of the compressed measurement values ​​obtained from the two-dimensional detector is processed in parallel across two domains.

[0027] The processing relies on the sparsity prior of the hyperspectral signal, i.e., the hyperspectral signal... In sparse base The following is sparsely represented:

[0028]

[0029] in, Hyperspectral signal In sparse base The sparse coefficient vector below, most of whose elements are zero or close to zero;

[0030] After processing, the processing path is divided into path A and path B. Reconstruction domain processing is performed in path A, and sparse domain processing is performed in path B.

[0031] Preferably, the reconstruction domain processing in path A specifically includes:

[0032] Based on compressed sensing theory, a preset sparse basis and reconstruction algorithm are selected to reconstruct two-dimensional compressed spectral measurements into a three-dimensional hyperspectral data cube.

[0033] The sparsity coefficient is calculated using the following formula:

[0034]

[0035] in, Let be the estimated value of the sparse coefficient vector to be solved. Indicates to Find the minimum point of the function. The square of the L2 norm, It is an L1 norm. Here, H is the regularization parameter, and H represents the sensing matrix. Represents a sparse base;

[0036] The reconstructed hyperspectral data cube is obtained by performing an inverse transform on the sparse coefficients, and the calculation formula is as follows:

[0037]

[0038] in, For hyperspectral data cubes;

[0039] Sparse domain processing is performed on path B, specifically including:

[0040] During the reconstruction algorithm, the sparse coefficient vector of the spectral signal corresponding to each spatial pixel under the sparse basis is extracted. As a characteristic of its sparse domain.

[0041] Preferably, the pre-constructed feature library contains standard samples of drones and background ground features, and the library includes:

[0042] A two-domain feature spectrum library is pre-constructed;

[0043] A reconstruction domain feature sub-library is constructed. In a controlled laboratory, the standard spectral reflectance curves and their enhancement features of UAVs and background samples are measured, and the measured data are stored in the reconstruction domain feature sub-library.

[0044] A sparse domain feature sub-library is constructed by projecting each standard spectral curve through the same sparse basis as that in the hyperspectral data cube to obtain its ideal sparse coefficient vector, which is then stored in the sparse domain feature sub-library. The formula for calculating the sparse coefficient vector is as follows:

[0045]

[0046] in, It is a sparse coefficient vector. for transpose, This is the standard spectral curve.

[0047] Preferably, the step of performing dual-domain joint matching on the pixels to be identified in the hyperspectral data cube specifically includes:

[0048] Reconstruction domain matching: Extract the spectral curve of the pixel to be identified in the reconstructed hyperspectral data cube, and perform similarity matching with the reconstruction domain feature sub-library to obtain the reconstruction domain matching score.

[0049] Sparse domain matching: Extract the sparse coefficient vector of the pixel to be identified, and perform similarity matching with the sparse domain feature sub-library to obtain the sparse domain matching score.

[0050] Decision fusion: Design fusion rules, combine the reconstructed domain matching score and the sparse domain matching score, calculate the final comprehensive confidence score, and determine whether the pixel belongs to the drone based on the preset threshold.

[0051] Preferably, the reconstructed domain matching specifically includes:

[0052] Using the spectral angle matching algorithm, calculate Compared with a standard spectrum in the library The similarity is the cosine of the angle between their vectors:

[0053]

[0054] in, Similarity score for spectral domain matching. For the first in the spectral library A standard spectral vector, The spectral vector to be matched, The angle between the spectral vector to be matched and the standard spectral vector cosine value, Let be the Euclidean norm of the spectral vector to be matched. For the first in the spectral library The Euclidean norm of a standard spectral vector.

[0055] Preferably, the sparse domain matching specifically includes:

[0056] Using cosine similarity, calculate With a certain ideal sparsity coefficient in the library Similarity:

[0057]

[0058] in, For similarity scores in sparse domain matching, Let be the sparse coefficient vector to be matched. The first in the spectral library An ideal sparse coefficient vector Let be the Euclidean norm of the sparse coefficient vector to be matched. The first in the spectral library The Euclidean norm of an ideal sparse coefficient vector.

[0059] Preferably, the decision fusion specifically includes:

[0060] The final confidence score is obtained by weighted fusion of the highest similarity scores from the spectral domain matching and the sparse domain matching.

[0061]

[0062] in, The final confidence score is calculated as follows. To integrate weights, Similarity score for spectral domain matching. Similarity score for sparse domain matching;

[0063] like If the value exceeds a preset threshold, the pixel is identified as a drone pixel.

[0064] Preferably, the step of performing spatial clustering analysis on all points identified as drone pixels to ultimately confirm and output the location of the drone target specifically includes:

[0065] For all point sets identified as drone pixels, the DBSCAN clustering algorithm is used for spatial clustering to form candidate target patches;

[0066] The two-dimensional detector can detect the farthest distance of the drone;

[0067] The area and aspect ratio morphological parameters of each patch are calculated. The area and aspect ratio morphological parameters of each patch are compared with the expected physical size of the UAV at the farthest distance. Unreasonable noise and false alarm targets are eliminated, and finally the UAV's position information is confirmed and output.

[0068] Compared with the prior art, the beneficial effects of the present invention are:

[0069] 1. This invention effectively overcomes the shortcomings of single reconstruction domain recognition methods, which are susceptible to artifacts and noise from reconstruction algorithms, by performing dual matching and verification in the reconstruction domain and sparse domain. The complementarity of the dual-domain information significantly improves the reliability of the recognition results and reduces the false alarm rate and false negative rate.

[0070] 2. This invention, by matching in the sparse domain of the signal, is equivalent to directly comparing the inherent structural genes of the signal, which can uncover deep identification features that are difficult to characterize by traditional spectral curves, thus improving the ability to distinguish similar materials.

[0071] 3. This invention, by tightly coupling with the front-end sensing process of computational spectral imaging, makes full use of the intermediate information of the compressed sensing process. It is a smart sensing solution that deeply integrates software and hardware, rather than a simple back-end algorithm application.

[0072] 4. This invention achieves rapid response and identification of targets while ensuring accuracy through dual-domain parallel processing and matching, and can be combined with a fast pre-screening mechanism. Attached Figure Description

[0073] Figure 1 The above is an overall flowchart of a UAV spectral recognition method provided in an embodiment of the present invention.

[0074] Figure 2 This is a schematic diagram of the compressed sensing spectral imaging and reconstruction system used in an embodiment of the present invention.

[0075] Figure 3 This is a schematic diagram of the structure of the dual-domain feature spectral library constructed in an embodiment of the present invention.

[0076] Figure 4 This is a conceptual block diagram of dual-domain joint matching and decision fusion in an embodiment of the present invention. Detailed Implementation

[0077] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0078] To address the issue that existing technologies inevitably produce artifacts, noise, and information distortion in reconstructed spectral data, which directly reduce the accuracy of subsequent recognition algorithms and fail to deeply integrate the sensing and recognition processes, please refer to [the relevant documentation / reference]. Figures 1-4 This embodiment provides the following technical solution:

[0079] A UAV spectral recognition method based on sparse reconstruction dual-domain joint matching includes:

[0080] Step 1: Using a computational spectral imaging system: Optically encode and compress the scene under test, including the UAV, to obtain a single frame of two-dimensional compressed spectral measurement value;

[0081] Step 2: Parallel processing of dual-domain data: Perform parallel processing of the acquired single-frame two-dimensional compressed spectral measurement values ​​in dual-domain.

[0082] Step 3: Construct a dual-domain feature spectral library: Pre-construct a feature library containing standard samples of UAVs and background ground features;

[0083] Step 4: Dual-domain joint matching and decision fusion: Perform dual-domain joint matching on the pixels to be identified in the hyperspectral data cube;

[0084] Step 5: Spatial Clustering and Target Confirmation: Perform spatial clustering analysis on all points identified as drone pixels, and finally confirm and output the location of the drone target.

[0085] Optical encoding compression sampling is performed on the test scene including drones to obtain single-frame two-dimensional compressed spectral measurements, specifically including:

[0086] The light emitted by the target object S passes through a variable resolution optical system and is incident on a pixel-level random filter array;

[0087] The filter array randomly modulates the spectral information at different spatial locations, and the modulated light signal is sensed by the detector array to form a single frame of two-dimensional compressed spectral measurement image.

[0088] The 3D hyperspectral data cube of the scene under test is vectorized into a one-dimensional column vector, as shown in the following expression:

[0089]

[0090]

[0091] in, This is the one-dimensional column vector obtained after vectorizing the three-dimensional hyperspectral data cube. This represents the dimension of the one-dimensional column vector after vectorization. The number of bands in the spectral dimension. The number of image rows in the spatial dimension. The number of image columns in the spatial dimension;

[0092] The compressed measurement values ​​obtained from the two-dimensional detector are vectorized into a one-dimensional column vector. ,in The imaging process can then be represented by the following formula:

[0093]

[0094]

[0095] in, It is a spectral signal. This is a one-dimensional column vector derived from the vectorized compressed measurement values ​​obtained on the two-dimensional detector. Let be the sensing matrix, representing the entire end-to-end optical transformation process from the optical system and coded aperture to the detector integration. This represents a one-dimensional column vector after the compressed measurement values ​​have been vectorized. Dimensions This is additive noise in the system.

[0096] The specific steps for performing dual-domain parallel processing on the acquisition of single-frame two-dimensional compressed spectral measurements include:

[0097] Obtain the one-dimensional column vector of the compressed measurement values ​​obtained from the two-dimensional detector. Then, the vectorized one-dimensional column vector of the compressed measurement values ​​obtained from the two-dimensional detector is subjected to a two-domain parallel processing flow.

[0098] The processing relies on the sparsity prior of the hyperspectral signal, i.e., the hyperspectral signal... In sparse base The following is sparsely represented:

[0099]

[0100] in, Hyperspectral signal In sparse base The sparse coefficient vector below, most of whose elements are zero or close to zero;

[0101] After processing, the processing path is divided into path A and path B. Reconstruction domain processing is performed in path A, and sparse domain processing is performed in path B.

[0102] The reconstruction of the domain in path A includes:

[0103] Based on compressed sensing theory, a preset sparse basis and reconstruction algorithm are selected to reconstruct two-dimensional compressed spectral measurements into a three-dimensional hyperspectral data cube.

[0104] The sparsity coefficient is calculated using the following formula:

[0105]

[0106] in, Let be the estimated value of the sparse coefficient vector to be solved. Indicates to Find the minimum point of the function. The square of the L2 norm represents the data fidelity term, ensuring that the reconstructed signal matches the measured value; It is the L1 norm, used to facilitate solutions. sparsity; It is a regularization parameter used to balance the fidelity and sparsity terms, and H represents the sensing matrix. This represents a sparse base.

[0107] The reconstructed hyperspectral data cube is obtained by performing an inverse transform on the sparse coefficients, and the calculation formula is as follows:

[0108]

[0109] in, The hyperspectral data cube serves as the data foundation for reconstructing domain matching.

[0110] Sparse domain processing is performed on path B, specifically including:

[0111] During the reconstruction algorithm, the sparse coefficient vector of the spectral signal corresponding to each spatial pixel under the sparse basis is extracted. , as a characteristic of its sparse domain;

[0112] Path B (Sparse Domain Processing): The sparse coefficient vector obtained in the process of solving the above optimization problem. It is itself a highly condensed and structured description of the original spectral signal. We directly use the sparse coefficient vector corresponding to each pixel. Extract it as its feature in the sparse domain.

[0113] A feature library containing standard samples of drones and background ground features is pre-built. This library includes:

[0114] A two-domain feature spectrum library is pre-constructed;

[0115] Construct a feature sub-library of the reconstruction domain. In a controlled laboratory, measure the standard spectral reflectance curves of UAVs and background samples and their enhanced features (such as differential spectra, absorption feature parameters, etc.), and store the measured data in the feature sub-library of the reconstruction domain.

[0116] A sparse domain feature sub-library is constructed by projecting each standard spectral curve through the same sparse basis as that in the hyperspectral data cube to obtain its ideal sparse coefficient vector, which is then stored in the sparse domain feature sub-library. The formula for calculating the sparse coefficient vector is as follows:

[0117]

[0118] in, It is a sparse coefficient vector. for The transpose of (for orthogonal basis). These are standard spectral curves. This constitutes a sparse domain sub-library.

[0119] For the pixels to be identified in the hyperspectral data cube, perform dual-domain joint matching, specifically including:

[0120] Reconstruction domain matching: Extract the spectral curve of the pixel to be identified in the reconstructed hyperspectral data cube, and perform similarity matching with the reconstruction domain feature sub-library to obtain the reconstruction domain matching score.

[0121] Sparse domain matching: Extract the sparse coefficient vector of the pixel to be identified, and perform similarity matching with the sparse domain feature sub-library to obtain the sparse domain matching score.

[0122] Decision fusion: Design fusion rules, combine the reconstructed domain matching score and the sparse domain matching score, calculate the final comprehensive confidence score, and determine whether the pixel belongs to the drone based on the preset threshold.

[0123] Reconstructing domain matching specifically includes:

[0124] Using the spectral angle matching algorithm, calculate Compared with a standard spectrum in the library The similarity is the cosine of the angle between their vectors:

[0125]

[0126] in, Similarity score for spectral domain matching. The value range is [-1, 1], and the closer it is to 1, the better the match. For the first in the spectral library A standard spectral vector, The spectral vector to be matched, The angle between the spectral vector to be matched and the standard spectral vector cosine value, Let be the Euclidean norm of the spectral vector to be matched. For the first in the spectral library The Euclidean norm of a standard spectral vector.

[0127] Sparse domain matching, specifically including:

[0128] Using cosine similarity, calculate With a certain ideal sparsity coefficient in the library Similarity:

[0129]

[0130] in, For similarity scores in sparse domain matching, The range of is also [-1, 1], and the closer it is to 1, the better it matches the sparse structure. Let be the sparse coefficient vector to be matched. The first in the spectral library An ideal sparse coefficient vector Let be the Euclidean norm of the sparse coefficient vector to be matched. The first in the spectral library The Euclidean norm of an ideal sparse coefficient vector.

[0131] Decision integration specifically includes:

[0132] The final confidence score is obtained by weighted fusion of the highest similarity scores from the spectral domain matching and the sparse domain matching.

[0133]

[0134] in, The final confidence score is calculated as follows. To integrate weights, Similarity score for spectral domain matching. Similarity score for sparse domain matching;

[0135] like If the value exceeds a preset threshold, the pixel is identified as a drone pixel.

[0136] Spatial clustering analysis is performed on all points identified as drone pixels to ultimately confirm and output the location of the drone target, specifically including:

[0137] For all point sets identified as drone pixels, the DBSCAN clustering algorithm is used for spatial clustering to form candidate target patches;

[0138] The two-dimensional detector can detect the farthest distance of the drone;

[0139] The area and aspect ratio morphological parameters of each patch are calculated. The area and aspect ratio morphological parameters of each patch are compared with the expected physical size of the UAV at the farthest distance. Unreasonable noise and false alarm targets are eliminated, and finally the UAV's position information is confirmed and output.

[0140] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0141] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention.

Claims

1. A UAV spectral recognition method based on sparse reconstruction dual-domain joint matching, characterized in that, include: A computational spectral imaging system is used to perform optical encoding compression sampling on the scene under test, which includes the UAV, to obtain a single frame of two-dimensional compressed spectral measurement value. Dual-domain parallel data processing: Dual-domain parallel processing is performed on the acquired single-frame two-dimensional compressed spectral measurement values; The specific steps of performing dual-domain parallel processing on the acquired single-frame two-dimensional compressed spectral measurement values ​​include: After obtaining the vectorized one-dimensional column vector of the compressed measurement values ​​obtained from the two-dimensional detector, the vectorized one-dimensional column vector of the compressed measurement values ​​obtained from the two-dimensional detector is processed in parallel across two domains. The processing relies on the sparsity prior of the hyperspectral signal, i.e., the hyperspectral signal... In sparse base The following is sparsely represented: ； in, Hyperspectral signal In sparse base The sparse coefficient vector below, most of whose elements are zero or close to zero; After processing, the processing path is divided into path A and path B. Reconstruction domain processing is performed in path A, and sparse domain processing is performed in path B. The reconstruction of the domain in path A includes: Based on compressed sensing theory, a preset sparse basis and reconstruction algorithm are selected to reconstruct two-dimensional compressed spectral measurements into a three-dimensional hyperspectral data cube. The sparsity coefficient is calculated using the following formula: ； in, Let be the estimated value of the sparse coefficient vector to be solved. Indicates to Find the minimum point of the function. The square of the L2 norm, It is an L1 norm. It is a regularization parameter; The reconstructed hyperspectral data cube is obtained by performing an inverse transform on the sparse coefficients, and the calculation formula is as follows: ； in, For hyperspectral data cubes; Sparse domain processing is performed on path B, specifically including: During the reconstruction algorithm, the sparse coefficient vector of the spectral signal corresponding to each spatial pixel under the sparse basis is extracted. , as its sparse domain characteristic; Constructing a dual-domain feature spectral library: Pre-construct a feature library containing standard samples of UAVs and background ground features; Dual-domain joint matching and decision fusion: Perform dual-domain joint matching on the pixels to be identified in the hyperspectral data cube; Spatial clustering and target confirmation: Spatial clustering analysis is performed on all points identified as drone pixels to ultimately confirm and output the location of the drone target.

2. The UAV spectral recognition method based on sparse reconstruction dual-domain joint matching according to claim 1, characterized in that, The step of performing optical encoding compression sampling on the test scene including the drone to obtain a single-frame two-dimensional compressed spectral measurement value specifically includes: The light emitted by the target object S passes through a variable resolution optical system and is incident on a pixel-level random filter array; The filter array randomly modulates the spectral information at different spatial locations, and the modulated light signal is sensed by the detector array to form a single frame of two-dimensional compressed spectral measurement image. The 3D hyperspectral data cube of the scene under test is vectorized into a one-dimensional column vector, as shown in the following expression: ； ； in, This is the one-dimensional column vector obtained after vectorizing the three-dimensional hyperspectral data cube. This represents the dimension of the one-dimensional column vector after vectorization. The number of bands in the spectral dimension. The number of image rows in the spatial dimension. The number of image columns in the spatial dimension; The compressed measurement values ​​obtained from the two-dimensional detector are represented as a one-dimensional column vector. ,in The imaging process can then be represented by the following formula: ； ； in, This is a one-dimensional column vector derived from the vectorized compressed measurement values ​​obtained on the two-dimensional detector. For the sensing matrix, This represents a one-dimensional column vector after the compressed measurement values ​​have been vectorized. Dimensions This is additive noise in the system.

3. The UAV spectral recognition method based on sparse reconstruction dual-domain joint matching according to claim 1, characterized in that, The pre-constructed feature library contains standard samples of drones and background ground features, and the library includes: A two-domain feature spectrum library is pre-constructed; A reconstruction domain feature sub-library is constructed. In a controlled laboratory, the standard spectral reflectance curves and their enhancement features of UAVs and background samples are measured, and the measured data are stored in the reconstruction domain feature sub-library. A sparse domain feature sub-library is constructed by projecting each standard spectral curve through the same sparse basis as that in the hyperspectral data cube to obtain its ideal sparse coefficient vector, which is then stored in the sparse domain feature sub-library. The formula for calculating the sparse coefficient vector is as follows: ； in, It is a sparse coefficient vector. for transpose, This is the standard spectral curve.

4. The UAV spectral recognition method based on sparse reconstruction dual-domain joint matching according to claim 1, characterized in that, The process of performing dual-domain joint matching on the pixels to be identified in the hyperspectral data cube specifically includes: Reconstruction domain matching: Extract the spectral curve of the pixel to be identified in the reconstructed hyperspectral data cube, and perform similarity matching with the reconstruction domain feature sub-library to obtain the reconstruction domain matching score; Sparse domain matching: Extract the sparse coefficient vector of the pixel to be identified, perform similarity matching with the sparse domain feature sub-library, and obtain the sparse domain matching score; Decision fusion: Design fusion rules, combine the reconstructed domain matching score and the sparse domain matching score, calculate the final comprehensive confidence score, and determine whether the pixel belongs to the drone based on the preset threshold.

5. The UAV spectral recognition method based on sparse reconstruction dual-domain joint matching according to claim 4, characterized in that, The reconstructed domain matching specifically includes: Using the spectral angle matching algorithm, calculate Compared with a standard spectrum in the library The similarity is the cosine of the angle between their vectors: ； in, Similarity score for spectral domain matching. For the first in the spectral library A standard spectral vector, The spectral vector to be matched, The angle between the spectral vector to be matched and the standard spectral vector The cosine value, Let be the Euclidean norm of the spectral vector to be matched. For the first in the spectral library The Euclidean norm of a standard spectral vector.

6. The UAV spectral recognition method based on sparse reconstruction dual-domain joint matching according to claim 4, characterized in that, The sparse domain matching specifically includes: Using cosine similarity, calculate With a certain ideal sparsity coefficient in the library Similarity: ； in, For similarity scores in sparse domain matching, Let be the sparse coefficient vector to be matched. The first in the spectral library An ideal sparse coefficient vector Let be the Euclidean norm of the sparse coefficient vector to be matched. The first in the spectral library The Euclidean norm of an ideal sparse coefficient vector.

7. The UAV spectral recognition method based on sparse reconstruction dual-domain joint matching according to claim 4, characterized in that, The decision fusion specifically includes: The highest similarity scores of the spectral domain matching and the sparse domain matching are weighted and fused to obtain the final confidence score. ； in, The final confidence score is calculated as follows. For weighting; like If the value exceeds a preset threshold, the pixel is identified as a drone pixel.

8. The UAV spectral recognition method based on sparse reconstruction dual-domain joint matching according to claim 1, characterized in that, The step of performing spatial clustering analysis on all points identified as drone pixels to ultimately confirm and output the location of the drone target specifically includes: For all point sets identified as drone pixels, the DBSCAN clustering algorithm is used for spatial clustering to form candidate target patches; The area and aspect ratio morphological parameters of each patch are calculated and compared with the expected physical size of the UAV at this detection distance. Unreasonable noise and false alarm targets are eliminated, and finally the UAV's position information is confirmed and output.

Images