Hyperspectral image aerial small target detection method and device

Abstract

The invention provides a hyperspectral image aerial small target detection method and device, and the method comprises the steps: determining a plurality of wavebands with the maximum contrast based on a local contrast feature pattern of each waveband in a target hyperspectral image, and obtaining the target hyperspectral image after the waveband selection; calculating the enhanced mahalanobis distance of the spatial-spectral variance ratio of each spectral vector in the target hyperspectral image after the band selection; and determining a small target detection result of the target hyperspectral image based on the Mahalanobis distance with the enhanced spatial-spectral variance ratio. Therefore, small targets in the air can be monitored, wavebands with differences between the targets and the background can be well screened out, redundant background interference is reduced, the detection precision is improved, and the false alarm rate is reduced.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of target detection, and in particular to a hyperspectral image aerial small target detection method and device. BACKGROUND

[0002] Aerial small target detection refers to the use of remote sensing images obtained by space-based satellites to realize the detection of aerial targets such as civil aviation aircraft, which is of great significance to the safety of air traffic and national defense. Current space-based aerial small target detection systems are mostly based on infrared sensors, but due to the few characteristics of aerial small targets, it brings certain difficulties to their detection. Hyperspectral images provide a new way for aerial small target detection because they can carry spatial and spectral information at the same time. Hyperspectral remote sensing is a technology that continuously subdivides the spectrum to reach one percent of the band width based on spectroscopy, obtaining complete and continuous spectral curves of observed ground objects. On this basis, two-dimensional spatial information and one-dimensional spectral information of the target can be obtained simultaneously, and target detection can be performed by analyzing the spatial information and spectral information of the target contained in the image.

[0003] Current band selection and anomaly detection algorithms are mostly for ground targets and cannot be well applied to aerial small target detection. Existing band selection methods mainly hope to reduce the dimension while retaining more information in the hyperspectral image, which makes the selected band subset contain less aerial small target information and more ground background information, which leads to a large amount of redundant information in the band subset that is irrelevant to the aerial small target detection task, which will affect the accuracy of subsequent anomaly detection. At the same time, existing band selection methods often rely on a large number of preset parameters, and the cost of selecting the optimal band subset is higher. Existing anomaly detection methods can be mainly divided into statistical-based methods, representation-based methods, decomposition-based methods, and deep learning-based methods. Statistical-based methods are often simple to calculate, but due to noise and target data distribution models that do not conform to their assumptions, they are easily affected by noise, especially by strip noise. Representation-based methods, decomposition-based methods, and deep learning-based methods often introduce a large number of parameters, and the calculation time is longer when processing large-size hyperspectral images, which is difficult to meet the efficient requirements of aerial small target detection tasks. Most importantly, existing anomaly detection algorithms are mostly for ground targets and are not suitable for direct detection of aerial small targets. SUMMARY

[0004] The application provides a hyperspectral image air small target detection method and device, which aims to solve the defects that the existing band selection and anomaly detection algorithm is not suitable for air small target detection, realizes monitoring of air small targets, can filter out the band of target and background difference, reduces redundant background interference, improves detection precision, and reduces false alarm rate.

[0005] The application provides a hyperspectral image air small target detection method, which comprises the following steps: Based on the local contrast feature map of each band in the target hyperspectral image, a plurality of bands with the maximum contrast are determined, and a target hyperspectral image after band selection is obtained. The air-spectrum variance ratio enhanced Mahalanobis distance of each spectral vector in the target hyperspectral image after band selection is calculated. Based on the air-spectrum variance ratio enhanced Mahalanobis distance, a small target detection result of the target hyperspectral image is determined.

[0006] According to the hyperspectral image air small target detection method provided by the application, the local contrast feature map of each band in the target hyperspectral image is obtained in the following manner: For any band, based on the inter-band transfer coefficient matrix, the pre-set inflation rate and the local displacement operation, the local contrast feature matrix of the any band in four displacement directions is calculated. Based on the minimum value of the local contrast feature matrix of the any band in four displacement directions, the local contrast feature map matrix of the any band is determined. Wherein, in the case that the any band is the first band, the inter-band transfer coefficient matrix is a unit matrix; otherwise, the inter-band transfer coefficient matrix is updated based on the relationship between each pixel value in the local contrast feature map matrix of the previous band of the any band and the pre-set threshold value.

[0007] According to the hyperspectral image air small target detection method provided by the application, based on the local contrast feature map of each band in the target hyperspectral image, a plurality of bands with the maximum contrast are determined, and the method specifically comprises: For any band, the maximum value in the local contrast feature map matrix of the any band is taken as the local contrast intensity value of the any band. The local contrast intensity values of each band are sorted from large to small, and the first pre-set number of bands are taken as the plurality of bands with the maximum contrast.

[0008] According to the present invention, a method for detecting small aerial targets in hyperspectral images calculates the Mahalanobis distance with enhanced spatial-spectral variance ratio for each spectral vector corresponding to each band in the hyperspectral image of the target after band selection. Specifically, this includes: Based on the target hyperspectral image after band selection, calculate the spectral vector matrix and the mean vector of all spectral vectors in the spectral vector matrix. Based on the mean vector, calculate the covariance matrix of the spectral vector matrix, and calculate the standard deviation and local spatial standard deviation of each spectral vector; Based on the mean vector, the covariance matrix, the standard deviation of each spectral vector, and the local spatial standard deviation of each spectral vector, the Mahalanobis distance with enhanced spatial-spectral variance ratio for each spectral vector is calculated.

[0009] According to the hyperspectral image aerial small target detection method provided by the present invention, the Mahalanobis distance with enhanced spatial-spectral variance ratio for each spectral vector is calculated according to the following formula: In the formula, Indicates the first spectral vectors The spatial spectral variance ratio is enhanced by the Mahalanobis distance; Indicates the first spectral vectors The local spatial standard deviation; Indicates the first spectral vectors Standard deviation; It is a mean vector; It is the covariance matrix of the spectral vector matrix.

[0010] According to the hyperspectral image aerial small target detection method provided by the present invention, the local spatial standard deviation of each spectral vector is calculated in the following manner: The standard deviation filter kernel is determined based on a pre-set expansion rate; Based on the standard deviation filter kernel, the first band of the target hyperspectral image matrix after band selection is subjected to standard deviation filtering to obtain the local spatial standard deviation of each spectral vector.

[0011] The present invention also provides a hyperspectral image aerial small target detection device, comprising the following modules: The selection module is used to determine the multiple bands with the highest contrast based on the local contrast feature map of each band in the target hyperspectral image, and to obtain the target hyperspectral image after band selection. The calculation module is used to calculate the Mahalanobis distance with enhanced spatial-spectral variance ratio for each spectral vector in the target hyperspectral image after band selection. The detection module is used to determine the small target detection result of the target hyperspectral image based on the Mahalanobis distance enhanced by the spatial-spectral variance ratio.

[0012] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the hyperspectral image aerial small target detection method as described above.

[0013] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the hyperspectral image aerial small target detection method as described above.

[0014] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the hyperspectral image aerial small target detection method as described above.

[0015] The hyperspectral image aerial small target detection method and apparatus provided by this invention determines multiple bands with the highest contrast based on the local contrast feature map of each band in the target hyperspectral image, and obtains the target hyperspectral image after band selection. This can effectively filter out bands with differences between the target and the background. Then, the Mahalanobis distance with enhanced spatial-spectral variance ratio of each spectral vector in the target hyperspectral image after band selection is calculated. Based on the Mahalanobis distance with enhanced spatial-spectral variance ratio, the small target detection result of the target hyperspectral image is determined. This can enhance the suppression of background and noise, thereby realizing the monitoring of small aerial targets, reducing redundant background interference, improving detection accuracy, and reducing false alarm rate. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in this invention 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 some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0017] Figure 1 This is a flowchart illustrating the hyperspectral image aerial small target detection method provided by the present invention.

[0018] Figure 2 This is a flowchart illustrating the hyperspectral image aerial small target detection method with combined band selection and anomaly detection provided by the present invention.

[0019] Figure 3 This is a schematic diagram illustrating the calculation method of the cyclic displacement strategy provided by the present invention.

[0020] Figure 4 This is a schematic diagram of the structure of the hyperspectral image aerial small target detection device provided by the present invention.

[0021] Figure 5 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0023] Figure 1 This is a flowchart illustrating the hyperspectral image aerial small target detection method provided by the present invention, as shown below. Figure 1 As shown, the method includes the following steps: Step 100: Based on the local contrast feature map of each band in the target hyperspectral image, determine the multiple bands with the largest contrast and obtain the target hyperspectral image after band selection.

[0024] Step 101: Calculate the Mahalanobis distance with enhanced spatial-spectral variance ratio for each spectral vector in the target hyperspectral image after band selection.

[0025] Step 102: Determine the small target detection result of the target hyperspectral image based on the Mahalanobis distance enhanced by the spatial-spectral variance ratio.

[0026] Specifically, the first step is to select the bands that best represent the target features from the hyperspectral image. In this embodiment of the invention, the local contrast feature map of each band can be determined based on the target hyperspectral image.

[0027] Local contrast is measured by calculating the difference between a pixel and its neighborhood to indicate whether the pixel is significant. This difference can be variations in brightness, spectral values, texture, etc. In hyperspectral images, the calculation of local contrast can take into account the variations of each pixel across different spectral bands. For example, if the spectral value of a pixel changes significantly within a small neighborhood, then this area may be a region of interest.

[0028] After determining the local contrast feature map of each band, the bands with the highest contrast can be selected as the bands that best reflect the target, and a hyperspectral image of the target after band selection can be constructed.

[0029] This step allows us to select the bands with the highest contrast and the most representative bands, reducing data dimensionality and improving the accuracy of target detection.

[0030] After obtaining the band-selected hyperspectral image of the target, the Mahalanobis distance of the spatial-spectral variance ratio enhancement for each spectral vector in the band-selected target hyperspectral image can be calculated. For the band-selected target hyperspectral image, the spectral response of each pixel constitutes a spectral vector, that is, a spectral vector contains the spectral values ​​of that pixel in multiple bands.

[0031] Mahalanobis distance is a measure of the difference between a data point and the center of the distribution. It considers not only the difference between the data point and the mean, but also the covariance structure of the data distribution and the correlation between different variables.

[0032] Mahalanobis distance provides a more accurate similarity measure by calculating the difference between the pixel spectral vector and the mean vector of all pixels and adjusting it according to the covariance structure of the data. It is suitable for the correlation between different bands in hyperspectral data.

[0033] The spatial-spectral variance ratio is a metric used to measure the difference in variation of pixels across the spatial and spectral dimensions. It identifies pixels with large spectral and spatial variations by comparing the spectral standard deviation (which describes spectral variation) with the local spatial standard deviation (which describes variation within the spatial neighborhood).

[0034] A high spatial-spectral variance ratio usually indicates that the pixel has a large spatial variation or a significant spectral variation compared to surrounding pixels, which may be a characteristic of a target or anomaly region.

[0035] In this embodiment of the invention, the spatial-spectral variance ratio-enhanced Mahalanobis distance combines the spatial-spectral variance ratio with the traditional Mahalanobis distance. It can introduce the spatial-spectral variance ratio as a weighting factor on the basis of the Mahalanobis distance to obtain the spatial-spectral variance ratio-enhanced Mahalanobis distance, which enhances the sensitivity of the Mahalanobis distance to pixels with large spatial and spectral variations.

[0036] Therefore, by calculating the Mahalanobis distance enhanced by the spatial-spectral variance ratio of each spectral vector, small target regions in the target hyperspectral image can be detected, and small target detection results can be obtained.

[0037] For example, a high Mahalanobis distance value might indicate an anomaly or target region in an image. Therefore, an appropriate threshold can be selected based on the Mahalanobis distance with enhanced spatial-spectral variance ratio to distinguish between target and background regions. Pixels with a Mahalanobis distance greater than the set threshold are considered target regions, thus obtaining small target detection results.

[0038] The hyperspectral image aerial small target detection method provided by this invention determines multiple bands with the highest contrast based on the local contrast feature map of each band in the target hyperspectral image, and obtains a target hyperspectral image after band selection. This method can effectively filter out bands with differences between the target and the background. Then, it calculates the Mahalanobis distance of the spatial-spectral variance ratio enhancement for each spectral vector in the target hyperspectral image after band selection. Based on the Mahalanobis distance of the spatial-spectral variance ratio enhancement, the small target detection result of the target hyperspectral image is determined. This method can enhance the suppression of background and noise, thereby realizing the monitoring of small aerial targets, reducing redundant background interference, improving detection accuracy, and reducing false alarm rate.

[0039] According to the hyperspectral image aerial small target detection method provided by the present invention, the local contrast feature map of each band in the hyperspectral image of the target is obtained according to the following method: For any band, the local contrast feature matrix of any band in four displacement directions is calculated based on the inter-band transfer coefficient matrix, the pre-set expansion rate, and local displacement operation. The local contrast feature matrix of any band is determined based on the minimum value of the local contrast feature matrix in the four displacement directions. In the case where any band is the first band, the inter-band transfer coefficient matrix is ​​an identity matrix; otherwise, the inter-band transfer coefficient matrix is ​​updated based on the relationship between each pixel value in the local contrast feature map matrix of the previous band of any band and a pre-set threshold.

[0040] Specifically, the spatial contrast characteristics of each band can be represented by calculating the spectral differences and spatial variations within a local region, thereby obtaining the local contrast feature map of each band in the target hyperspectral image.

[0041] First, an inter-band transfer coefficient matrix can be initialized. This matrix is ​​used during hyperspectral image processing to adjust the interactions between different bands, thereby controlling the enhancement or suppression of bands. In this way, the target region can be enhanced based on the local contrast characteristics of each band, while suppressing background noise, thus improving the target detection performance of the image.

[0042] The initial inter-band transfer coefficient matrix is ​​an identity matrix, meaning that the initial transfer coefficient between all bands is 1, indicating no enhancement or suppression.

[0043] It is understood that in this embodiment of the invention, images of each band need to be processed sequentially. Therefore, after processing the image of the first band using an initialized inter-band transfer coefficient matrix to obtain the local contrast feature map of the first band, the inter-band transfer coefficient matrix used for any band starting from the second band will be updated according to the relationship between each pixel value in the local contrast feature map matrix of the previous band and a preset threshold. For any band starting from the second band, the image of that band can be processed according to the updated inter-band transfer coefficient matrix, thereby enhancing the suspected target area and suppressing the suspected background area.

[0044] In this embodiment of the invention, if the local contrast of some pixels in the local contrast feature map matrix is ​​greater than a threshold, the transmission coefficient at the corresponding position in the inter-band transmission coefficient matrix is ​​increased by a fixed value (enhancement), and conversely, it is decreased by a fixed value (suppression).

[0045] After processing an image of a specific band using the inter-band transfer coefficient matrix, the spatial contrast of that band can be calculated using a local shift operation. A shift operation involves moving pixel values ​​in the image a certain distance along a specific direction (such as up, down, left, right, or diagonal), and the shift distance can be determined by a pre-set dilation rate.

[0046] For example, define a set of displacement directions. , representing the four directions of the image: top left, top, top right, and left.

[0047] For each band, a cyclic shift operation can be performed on each pixel of the target hyperspectral image, i.e., according to... The set is translated pixel by pixel to obtain a new displacement image matrix. There are two corresponding displacement results in each straight line direction, as shown in the following formula: Then, the local contrast of each pixel needs to be calculated. Therefore, after obtaining the displacement image matrix of any band, the target hyperspectral image and the displacement image matrix in the same straight line direction can be compared. and Take the differences separately, then perform a Cartesian product. ), to obtain the local contrast feature matrix in this direction. As shown in the following formula: Since local contrast in different directions reflects the changes in an image in different spatial directions, and the minimum value can preserve the areas with stronger changes, the local contrast matrix in each direction is calculated. Then, it can be based on the preset expansion rate. The minimum value among multiple directions is selected as the local contrast feature map of this band, thus obtaining the local contrast feature map matrix of this band, as shown in the following formula: This step helps capture local changes in the image in different directions and selects the most significant changes as the contrast features of that band.

[0048] According to the present invention, a method for detecting small aerial targets using hyperspectral images is provided, which determines multiple bands with the highest contrast based on the local contrast feature map of each band in the hyperspectral image of the target. Specifically, the method includes: For any band, the maximum value in the local contrast feature map matrix of any band is taken as the local contrast intensity value of any band. The local contrast intensity values ​​of each band are sorted from largest to smallest, and the first preset number of bands are selected as the bands with the highest contrast.

[0049] Specifically, after obtaining the local contrast feature map of each band, the contrast index of each band can be calculated. In this embodiment of the invention, for any band, the maximum value in the local contrast feature map matrix of that band can be used as the local contrast intensity value of that band, that is, the contrast index of that band.

[0050] The contrast indices (local contrast intensity values) of all bands are sorted in descending order from largest to smallest, and the top N bands with the highest contrast are selected. These bands have the highest contrast and best reflect the information of the target area or anomaly area.

[0051] According to the present invention, a method for detecting small aerial targets in hyperspectral images calculates the Mahalanobis distance with enhanced spatial-spectral variance ratio for each spectral vector in the hyperspectral image of the target after band selection, specifically including: Based on the target hyperspectral image after band selection, calculate the spectral vector matrix and the mean vector of all spectral vectors in the spectral vector matrix. Based on the mean vector, calculate the covariance matrix of the spectral vector matrix, and calculate the standard deviation and local spatial standard deviation of each spectral vector; The Mahalanobis distance with enhanced spatial-spectral variance ratio for each spectral vector is calculated based on the mean vector, covariance matrix, standard deviation of each spectral vector, and local spatial standard deviation of each spectral vector.

[0052] Specifically, for the target hyperspectral image after band selection ,in and These are the height and width of the image, respectively. This is the number of selected bands. Each pixel has... The spectral values ​​of each band form a spectral vector. For ease of processing, the spectral values ​​of each pixel can be arranged into a matrix, i.e., a spectral vector matrix, where each column represents the spectral vector corresponding to a band.

[0053] After obtaining the spectral vector matrix, the mean vector of all spectral vectors in the spectral vector matrix can be calculated. It is obtained by averaging the spectral vector of each band and is used to reflect the average spectral characteristics of the entire image.

[0054] After obtaining the mean vector, the covariance matrix of the spectral vector matrix can be calculated based on the mean, as well as the standard deviation and local spatial standard deviation of each spectral vector.

[0055] The covariance matrix can be calculated using the following formula: No. spectral vectors The standard deviation can be calculated using the following formula: Alternatively, the local spatial standard deviation of each spectral vector can be calculated as follows: The standard deviation filter kernel is determined based on the pre-set expansion rate. ,Right now ; Based on the standard deviation filter kernel, the first band of the target hyperspectral image matrix after band selection is subjected to standard deviation filtering to obtain the local spatial standard deviation of each spectral vector, which is calculated according to the following formula: Finally, the Mahalanobis distance with enhanced spatial-spectral variance ratio can be calculated using the mean vector, covariance matrix, standard deviation of the spectral vector, and local spatial standard deviation.

[0056] Optionally, the Mahalanobis distance for the spatial spectral variance ratio enhancement of each spectral vector is calculated according to the following formula: In the formula, Indicates the first spectral vectors The spatial spectral variance ratio is enhanced by the Mahalanobis distance; Indicates the first spectral vectors The local spatial standard deviation; Indicates the first spectral vectors Standard deviation; It is a mean vector; It is the covariance matrix of the spectral vector matrix.

[0057] The following examples, through specific application scenarios, further illustrate the hyperspectral image aerial small target detection method provided by the present invention.

[0058] Figure 2 This is a flowchart illustrating the hyperspectral image aerial small target detection method with joint band selection and anomaly detection provided by the present invention, as shown below. Figure 2 As shown, this embodiment proposes a novel fast detection strategy for small aerial targets using band selection and anomaly detection. The processing can be divided into two stages: Phase 1: The Local Contrast-Based Band Selection Algorithm (LC-UBSM) reduces the dimensionality of hyperspectral data and increases the separability between the target and the background. In the first stage, for the hyperspectral image to be processed, firstly, for each band in the hyperspectral image, a cyclic shift strategy is used to obtain its corresponding local contrast feature map; simultaneously, based on the band continuity, the possible target and background location information provided by the feature map of the previous band is used to enhance the corresponding target region and suppress the corresponding background region in the subsequent band; finally, the maximum value in the feature map corresponding to each band is selected as the index of that band and sorted to obtain the required bands.

[0059] Phase Two: Achieving rapid target detection using the Joint Spatial and Spectral Variance Ratio-based RX anomaly detection algorithm (CSSV-RX). In the second stage, for the band subset with high separability between the target and background obtained in the first stage, a standard deviation filter kernel is used to calculate the local standard deviation of each pixel in the first band in the spatial dimension; in the spectral dimension, the standard deviation of the corresponding spectral vector for each pixel is calculated. Simultaneously, the Mahalanobis distance of the corresponding spectral vector for each pixel is calculated. Finally, the ratio of the spatial local standard deviation to the spectral standard deviation is used as the coefficient of the Mahalanobis distance to obtain the response of all pixels, i.e., the detection result map.

[0060] The specific algorithm is as follows: I. Calculation Instructions 1. Addition, subtraction, scalar multiplication, division, and assignment operations are conventional mathematical operations and programming languages.

[0061] 2. The dimension is An identity matrix of a certain size.

[0062] 3. Cartesian product Indicates the matrix and Multiplying each element in the matrix yields the matrix. .

[0063] 4. Cyclic displacement operation Figure 3 This is a schematic diagram of the calculation method for the cyclic displacement strategy provided by the present invention, as shown below. Figure 3 As shown, the cyclic shift operation represents the matrix... Towards Displacement in direction, i.e., matrix Split and reassemble, for example, the original matrix Move the distance to the right and down respectively. get : 5. Find the minimum value between the two matrices. Represents a comparison matrix sum matrix For each element in the matrix, select the minimum value to form a new matrix. .

[0064] 6. Indicates the matrix Become OK List.

[0065] 7. Matrix transpose. Representation matrix For matrix The transpose of .

[0066] 8. Find the mean of the matrix. Represents finding a matrix mean vector .

[0067] 9. Find the inverse matrix of the matrix. Represents finding a matrix inverse matrix .

[0068] 10. Standard Deviation Filtering .

[0069] Represents a matrix Perform standard deviation filtering, with a filter kernel size of . Even using identity pair matrix of size Perform filtering, and use the standard deviation of all elements in the filter kernel as the value of the center point.

[0070] 11. Find the standard deviation of the vector. The returned matrix The standard deviation.

[0071] II. Input Instructions 1. It is a hyperspectral image in the form of a three-dimensional matrix with H rows, W columns, and B bands.

[0072] 2. The expansion rate is set to be close to the target size.

[0073] 3. This is the inter-band transfer coefficient, used to control the magnitude of the influence of the previous band on the next band.

[0074] 4. The threshold is used to control whether the previous band enhances or suppresses each pixel in the next band.

[0075] III. Calculation Process Phase 1: Contrast-Based Band Selection Algorithm (LC-UBSM) Step 1. Input: The hyperspectral image to be processed. Expansion rate Its size is close to that of an aerial target, and its inter-band transmission coefficient is... and threshold It is used to control the proportion of transmission between wavebands.

[0076] Step 2. Initialize the inter-band transfer coefficient matrix as an identity matrix. .

[0077] Step 3. Calculate the local contrast feature map for each band by iterating through the bands.

[0078] for do: Step 4. Apply the inter-band transfer coefficient matrix obtained from the previous band to the next band to enhance the suspected target area and suppress the suspected background area: .

[0079] Step 5. For the first bands Perform a cyclic displacement operation as follows: Step 6. Use a matrix The matrix obtained by displacement in the same straight direction and By subtracting the components separately and then performing a Cartesian product, we obtain the local contrast feature matrix in that direction: .

[0080] Step 7. Take the minimum value of the local contrast matrix in the four directions to obtain the dilation rate. At that time, the first Local contrast feature map matrix for each band: .

[0081] Step 8. Determine the number of bands. Is it greater than 2 and less than the number of bands? .

[0082] if do: Step 9. Calculate the inter-band transfer coefficient matrix by iterating through the bands.

[0083] for do: Step 10. Determine whether the value of each pixel in the local contrast feature map is greater than the threshold.

[0084] if do: Step 11. Set the inter-band transfer coefficient to enhanced: .

[0085] else: Step 12. Set the inter-band transfer coefficient to suppressed: .

[0086] end if end for end if Step 13. Take the first... The maximum value in the local contrast feature map matrix of the i-th band is used as the i-th band. The index for each band characterizes the local contrast intensity of that band.

[0087] end for Step 14. Sort all bands according to their local contrast ratio, and select the band with the highest contrast ratio. The bands are the result of band selection: Step 15. Select a subset of output bands. .

[0088] Phase Two: RX Anomaly Detection Algorithm Based on Joint Spatial Spectral Variance (CSSV-RX) Step 1. Input: A subset of hyperspectral images after band selection in Phase 1. Expansion rate .

[0089] Step 2. Calculate the number of pixels contained in each band: .

[0090] Step 3. Convert the matrix Become OK List: .

[0091] Step 4. Find the matrix. Find the mean of all spectral vectors in the matrix, and then transpose the mean vector: .

[0092] Step 5. Find the matrix. Covariance matrix: .

[0093] Step 6. For the matrix The first band is subjected to standard deviation filtering, and the filter kernel size is [value missing]. : Step 7. Iterate through the spectrum and calculate the Mahalanobis distance for the spatial spectral variance ratio enhancement for each spectral vector.

[0094] for do: Step 8. Find the first... Standard deviation of each spectral vector: .

[0095] Step 9. Find the first... The Mahalanobis distance enhanced by the spatial-spectral variance ratio of each spectral vector: end for Step 10. Output the spatial-spectral variance ratio-enhanced Mahalanobis distance for each spectral vector. .

[0096] The hyperspectral image aerial small target detection device provided by the present invention is described below. The hyperspectral image aerial small target detection device described below can be referred to in correspondence with the hyperspectral image aerial small target detection method described above.

[0097] Figure 4 This is a schematic diagram of the structure of the hyperspectral image aerial small target detection device provided by the present invention, as shown below. Figure 4 As shown, the device includes the following modules: The selection module 400 is used to determine multiple bands with the highest contrast based on the local contrast feature map of each band in the target hyperspectral image, and to obtain the target hyperspectral image after band selection. Calculation module 410 is used to calculate the Mahalanobis distance with enhanced spatial-spectral variance ratio for each spectral vector in the target hyperspectral image after band selection. The detection module 420 is used to determine the small target detection result of the target hyperspectral image based on the Mahalanobis distance enhanced by the spatial-spectral variance ratio.

[0098] According to the hyperspectral image aerial small target detection device provided by the present invention, the local contrast feature map of each band in the hyperspectral image of the target is obtained according to the following method: For any band, the local contrast feature matrix of any band in four displacement directions is calculated based on the inter-band transfer coefficient matrix, the pre-set expansion rate, and local displacement operation. The local contrast feature matrix of any band is determined based on the minimum value of the local contrast feature matrix in the four displacement directions. In the case where any band is the first band, the inter-band transfer coefficient matrix is ​​an identity matrix; otherwise, the inter-band transfer coefficient matrix is ​​updated based on the relationship between each pixel value in the local contrast feature map matrix of the previous band of any band and a pre-set threshold.

[0099] According to the present invention, a hyperspectral image aerial small target detection device determines multiple bands with the highest contrast based on the local contrast feature map of each band in the hyperspectral image of the target, specifically including: For any band, the maximum value in the local contrast feature map matrix of any band is taken as the local contrast intensity value of any band. The local contrast intensity values ​​of each band are sorted from largest to smallest, and the first preset number of bands are selected as the bands with the highest contrast.

[0100] According to the present invention, a hyperspectral image aerial small target detection device calculates the Mahalanobis distance with enhanced spatial-spectral variance ratio for each spectral vector in the hyperspectral image of the target after band selection, specifically including: Based on the target hyperspectral image after band selection, calculate the spectral vector matrix and the mean vector of all spectral vectors in the spectral vector matrix. Based on the mean vector, calculate the covariance matrix of the spectral vector matrix, and calculate the standard deviation and local spatial standard deviation of each spectral vector; The Mahalanobis distance with enhanced spatial-spectral variance ratio for each spectral vector is calculated based on the mean vector, covariance matrix, standard deviation of each spectral vector, and local spatial standard deviation of each spectral vector.

[0101] According to the hyperspectral image aerial small target detection device provided by the present invention, the Mahalanobis distance with enhanced spatial-spectral variance ratio for each spectral vector is calculated according to the following formula: In the formula, Indicates the first spectral vectors The spatial spectral variance ratio is enhanced by the Mahalanobis distance; Indicates the first spectral vectors The local spatial standard deviation; Indicates the first spectral vectors Standard deviation; It is a mean vector; It is the covariance matrix of the spectral vector matrix.

[0102] According to the hyperspectral image aerial small target detection device provided by the present invention, the local spatial standard deviation of each spectral vector is calculated in the following manner: The standard deviation filter kernel is determined based on a pre-set expansion rate; Based on the standard deviation filter kernel, the first band of the target hyperspectral image matrix after band selection is filtered by standard deviation to obtain the local spatial standard deviation of each spectral vector.

[0103] Figure 5 This is a schematic diagram of the structure of the electronic device provided by the present invention, such as... Figure 5 As shown, the electronic device may include: a processor 510, a communications interface 520, a memory 530, and a communication bus 540, wherein the processor 510, the communications interface 520, and the memory 530 communicate with each other via the communication bus 540. The processor 510 can call logical instructions in the memory 530 to execute a hyperspectral image aerial small target detection method, which includes: Based on the local contrast feature map of each band in the target hyperspectral image, the multiple bands with the highest contrast are determined, and the target hyperspectral image after band selection is obtained. Calculate the Mahalanobis distance with enhanced spatial-spectral variance ratio for each spectral vector in the target hyperspectral image after band selection; Based on Mahalanobis distance enhanced by spatial-spectral variance ratio, the small target detection results of the target hyperspectral image are determined.

[0104] Furthermore, the logical instructions in the aforementioned memory 530 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0105] On the other hand, the present invention also provides a computer program product, the computer program product comprising a computer program that can be stored on a non-transitory computer-readable storage medium, wherein when the computer program is executed by a processor, the computer is able to execute the hyperspectral image aerial small target detection method provided by the above methods, the method comprising: Based on the local contrast feature map of each band in the target hyperspectral image, the multiple bands with the highest contrast are determined, and the target hyperspectral image after band selection is obtained. Calculate the Mahalanobis distance with enhanced spatial-spectral variance ratio for each spectral vector in the target hyperspectral image after band selection; Based on Mahalanobis distance enhanced by spatial-spectral variance ratio, the small target detection results of the target hyperspectral image are determined.

[0106] In another aspect, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the hyperspectral image aerial small target detection method provided by the methods described above, the method comprising: Based on the local contrast feature map of each band in the target hyperspectral image, the multiple bands with the highest contrast are determined, and the target hyperspectral image after band selection is obtained. Calculate the Mahalanobis distance with enhanced spatial-spectral variance ratio for each spectral vector in the target hyperspectral image after band selection; Based on Mahalanobis distance enhanced by spatial-spectral variance ratio, the small target detection results of the target hyperspectral image are determined.

[0107] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0108] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0109] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for detecting small aerial targets using hyperspectral images, characterized in that, include: Based on the local contrast feature map of each band in the target hyperspectral image, the multiple bands with the highest contrast are determined, and the target hyperspectral image after band selection is obtained. Calculate the Mahalanobis distance with enhanced spatial-spectral variance ratio for each spectral vector in the target hyperspectral image after band selection; Based on the Mahalanobis distance enhanced by the spatial-spectral variance ratio, the small target detection result of the target hyperspectral image is determined.

2. The method for detecting small aerial targets using hyperspectral images according to claim 1, characterized in that, The local contrast feature map of each band in the target hyperspectral image is obtained according to the following method: For any band, the local contrast feature matrix of the arbitrary band in four displacement directions is calculated based on the inter-band transfer coefficient matrix, the pre-set expansion rate, and the local displacement operation. Based on the minimum value of the local contrast feature matrix of the arbitrary band in the four displacement directions, the local contrast feature map matrix of the arbitrary band is determined. Wherein, if the arbitrary band is the first band, the inter-band transfer coefficient matrix is ​​an identity matrix; otherwise, the inter-band transfer coefficient matrix is ​​updated based on the relationship between each pixel value in the local contrast feature map matrix of the previous band of the arbitrary band and a preset threshold.

3. The method for detecting small aerial targets using hyperspectral imaging according to claim 1 or 2, characterized in that, Based on the local contrast feature map of each band in the target hyperspectral image, the multiple bands with the highest contrast are identified, specifically including: For any band, the maximum value in the local contrast feature map matrix of the arbitrary band is taken as the local contrast intensity value of the arbitrary band. The local contrast intensity values ​​of each band are sorted from largest to smallest, and the first preset number of bands are selected as the bands with the highest contrast.

4. The method for detecting small aerial targets using hyperspectral images according to claim 1, characterized in that, Calculating the Mahalanobis distance with enhanced spatial-spectral variance ratio for each spectral vector in the target hyperspectral image after band selection, specifically including: Based on the target hyperspectral image after band selection, calculate the spectral vector matrix and the mean vector of all spectral vectors in the spectral vector matrix. Based on the mean vector, calculate the covariance matrix of the spectral vector matrix, and calculate the standard deviation and local spatial standard deviation of each spectral vector; Based on the mean vector, the covariance matrix, the standard deviation of each spectral vector, and the local spatial standard deviation of each spectral vector, the Mahalanobis distance with enhanced spatial-spectral variance ratio for each spectral vector is calculated.

5. The method for detecting small aerial targets using hyperspectral images according to claim 4, characterized in that, The Mahalanobis distance enhanced by the spatial-spectral variance ratio for each spectral vector is calculated according to the following formula: In the formula, Indicates the first spectral vectors The spatial spectral variance ratio is enhanced by the Mahalanobis distance; Indicates the first spectral vectors The local spatial standard deviation; Indicates the first spectral vectors Standard deviation; It is a mean vector; It is the covariance matrix of the spectral vector matrix.

6. The method for detecting small aerial targets using hyperspectral images according to claim 4 or 5, characterized in that, The local spatial standard deviation of each spectral vector is calculated as follows: The standard deviation filter kernel is determined based on a pre-set expansion rate; Based on the standard deviation filter kernel, the first band of the target hyperspectral image matrix after band selection is subjected to standard deviation filtering to obtain the local spatial standard deviation of each spectral vector.

7. A hyperspectral imaging device for detecting small aerial targets, characterized in that, include: The selection module is used to determine the multiple bands with the highest contrast based on the local contrast feature map of each band in the target hyperspectral image, and to obtain the target hyperspectral image after band selection. The calculation module is used to calculate the Mahalanobis distance with enhanced spatial-spectral variance ratio for each spectral vector in the target hyperspectral image after band selection. The detection module is used to determine the small target detection result of the target hyperspectral image based on the Mahalanobis distance enhanced by the spatial-spectral variance ratio.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the hyperspectral image aerial small target detection method as described in any one of claims 1 to 6.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the hyperspectral image aerial small target detection method as described in any one of claims 1 to 6.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the hyperspectral image aerial small target detection method as described in any one of claims 1 to 6.

Images