Camouflaged target detection method and device based on spectral variation augmentation

Abstract

The present invention discloses a camouflaged target detection method and device based on spectral variation augmentation. The method first obtains a background spectral variation dictionary and a target spectral variation dictionary based on a hyperspectral image. The background spectral variation dictionary and the target spectral variation dictionary are then merged to form a joint spectral variation dictionary. The joint spectral variation dictionary is then used to expand the background dictionary and the target dictionary, respectively, to form an extended background dictionary and an extended target dictionary. Based on collaborative representation, the extended background dictionary and the extended target dictionary are used to reconstruct a test sample, respectively. The reconstruction errors of the extended background dictionary and the extended target dictionary are calculated. The reconstruction errors of the extended background dictionary and the extended target dictionary are then used to determine whether the test sample is a camouflaged target. This method increases the diversity of spectral variations in the dictionary, addressing the limited diversity of spectral variations in the dictionary caused by a small number of samples and spectral variations, thereby improving the detection accuracy of the method.

Description

Technical Field

[0001] The present invention relates to the field of computer vision technology, and in particular to a camouflaged target detection method and device based on spectral variation augmentation. Background Art

[0002] Camouflaged target detection technology can support and assist in scenarios such as military reconnaissance, security monitoring, and autonomous driving. Target camouflage refers to making a target object difficult to detect by humans or machines through concealment, coatings, or camouflage netting. With the continuous development of material technology and the continuous improvement of camouflage methods, the difficulty of target object identification is gradually increasing.

[0003] Visible light or multispectral images have a wide spectral range in a single band. Camouflage technology can alter the spectral characteristics of a target, resulting in the appearance of "different objects with the same spectrum" between the camouflaged target and the background. Hyperspectral remote sensing technology not only acquires spatial information about the target, but also its optical properties in multiple bands, providing more refined data support for camouflaged target detection. Hyperspectral remote sensing acquires high-resolution spectral information of a substance in different bands, providing discriminant information and spectral identity of the substance in the spectral dimension. The difference in spectral identity between the camouflaged target and surrounding objects provides a basis for camouflaged target detection.

[0004] In the prior art, based on the difference in spectral identity between the camouflaged target and the surrounding objects, the document "W. Li and Q. Du. Collaborative Representation for Hyperspectral Anomaly Detection [J]. IEEE Transactions on Geoscience and Remote Sensing, 2015, 53 (3): 1463-1474" discloses a hyperspectral image anomaly detection method based on collaborative representation to detect camouflaged targets. This method assumes that each pixel in the background can be approximately represented by its spatial neighborhood, while the abnormal pixel cannot. The collaborative constraint of the representation is strengthened by minimizing the l2-norm of the representation weight vector, and the contribution of each neighboring pixel is adjusted using a distance-weighted regularization matrix. This method can adaptively model the background and determine whether a pixel is abnormal by reconstruction error, thus avoiding the need to train a classification model. However, the anomaly detection accuracy is limited due to the lack of target samples. The target detection algorithm based on collaborative representation has a significantly improved detection accuracy when there are sufficient background samples and target samples. However, in actual scenarios, it is difficult to provide a large number of prior spectra of camouflaged targets. The number of target samples that can be obtained is small, and due to the influence of materials and imaging environment, there will be certain spectral variations between the spectral characteristics of the same ground object. The small number of samples and spectral variations will lead to limited spectral variation diversity of the dictionary in the target detection method based on collaborative representation, reducing the detection accuracy of the algorithm. Summary of the Invention

[0005] Based on this, it is necessary to provide a camouflaged target detection method and device based on spectral change augmentation that can overcome the problems of small sample size and spectral change and has high detection accuracy to address the above technical problems.

[0006] In a first aspect, the present invention provides a camouflaged target detection method based on spectral variation augmentation, comprising:

[0007] Obtaining a background spectrum change dictionary and a target spectrum change dictionary according to a hyperspectral image, wherein the hyperspectral image includes a training sample and a test sample, and the training sample includes a background sample and a target sample;

[0008] Merging the background spectrum change dictionary and the target spectrum change dictionary to form a joint spectrum change dictionary;

[0009] The background dictionary and target dictionary are respectively expanded using the joint spectral variation dictionary to form an extended background dictionary and an extended target dictionary. The background dictionary is the pixel set of all background samples, and the target dictionary is the set of all target samples.

[0010] Based on the collaborative representation, the test samples are reconstructed using the extended background dictionary and the extended target dictionary respectively, and the reconstruction error of the extended background dictionary and the reconstruction error of the extended target dictionary are calculated;

[0011] Whether the test sample is a disguised target is determined based on the reconstruction error of the extended background dictionary and the reconstruction error of the extended target dictionary.

[0012] In one embodiment, obtaining a background spectrum variation dictionary and a target spectrum variation dictionary according to a hyperspectral image includes:

[0013] Acquire hyperspectral images;

[0014] An unsupervised clustering algorithm is used to cluster the background samples and obtain several categories and cluster centers of each category;

[0015] Construct a background spectrum change dictionary based on the difference between the background sample and the cluster center of the category to which the background sample belongs;

[0016] The target spectrum variation dictionary is constructed based on the difference between the target sample and the target sample mean.

[0017] In one embodiment, a cascade operation is used to merge the background spectrum variation dictionary and the target spectrum variation dictionary to form a joint spectrum variation dictionary.

[0018] In one embodiment, using the joint spectrum variation dictionary to expand the background dictionary and the target dictionary to form the expanded background dictionary and the expanded target dictionary includes:

[0019] Combining the joint spectral variation dictionary with the background dictionary to form an extended background dictionary;

[0020] The joint spectral variation dictionary is combined with the target dictionary to form an extended target dictionary.

[0021] In one embodiment, reconstructing the test sample using the extended background dictionary and the extended target dictionary based on the collaborative representation, and calculating the reconstruction error of the extended background dictionary and the reconstruction error of the extended target dictionary includes:

[0022] Calculate the collaborative representation coefficients corresponding to the extended background dictionary and the extended target dictionary;

[0023] Use the extended background dictionary and the extended target dictionary to collaboratively represent the test samples respectively;

[0024] The reconstruction error of the extended background dictionary and the reconstruction error of the extended target dictionary are calculated based on the collaborative representation of the test sample by the test sample, the extended background dictionary and the extended target dictionary.

[0025] In one embodiment, the reconstruction error of the extended background dictionary and the reconstruction error of the extended target dictionary are calculated based on the collaborative representation of the test sample by the test sample, the extended background dictionary and the extended target dictionary.

[0026] r t (x)=||xD te γ te ||2

[0027] r b (x)=||xA be γ be ||2 (1)

[0028] In the formula, x represents the test sample, r t (x) represents the reconstruction error of the expanded target dictionary, r b (x) represents the reconstruction error of the extended background dictionary, D te r te Denotes the collaborative representation of the test sample using the extended target dictionary, D be r be Denotes the collaborative representation of the test sample using the extended background dictionary, D te Denotes the extended target dictionary, D be Represents the extended background dictionary, r te represents the collaborative representation coefficient corresponding to the extended target dictionary, r be Indicates the collaborative representation coefficient corresponding to the extended background dictionary.

[0029] In one embodiment, the calculation method of the collaborative representation coefficient corresponding to the extended target dictionary is as follows:

[0030] γ te =D te T D te +λI -1 D te T x (2)

[0031] The calculation method of the collaborative representation coefficient corresponding to the extended background dictionary is:

[0032] γ be =D be T D be +λI -1 D be T x (3)

[0033] Where λ is the regularization parameter, λ∈(0.001,0.01,0.1), and I represents the identity matrix.

[0034] In one embodiment, determining whether a test sample is a disguised target based on a reconstruction error of a background dictionary and a reconstruction error of a target dictionary includes:

[0035] Calculate the difference between the reconstruction error of the expanded background dictionary and the reconstruction error of the expanded target dictionary;

[0036] The difference between the reconstruction error of the extended background dictionary and the reconstruction error of the extended target dictionary and the error threshold is determined. If the difference between the reconstruction error of the extended background dictionary and the reconstruction error of the extended target dictionary is greater than or equal to the error threshold, the test sample is judged to be a disguised target. If the difference between the reconstruction error of the extended background dictionary and the reconstruction error of the extended target dictionary is less than the error threshold, the test sample is judged to be background.

[0037] In one embodiment, an unsupervised clustering algorithm is used to cluster the background samples into 20 to 40 categories. The unsupervised clustering algorithm is a K-Means clustering algorithm, a DBSCAN clustering algorithm, or a hierarchical clustering algorithm.

[0038] In a second aspect, the present invention also provides a camouflaged target detection device based on spectral variation augmentation, comprising a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the steps of the camouflaged target detection method based on spectral variation augmentation are implemented.

[0039] The aforementioned camouflaged target detection method and device based on spectral variation augmentation first obtains a background spectral variation dictionary and a target spectral variation dictionary from a hyperspectral image. The background spectral variation dictionary and the target spectral variation dictionary are then merged to form a joint spectral variation dictionary. The joint spectral variation dictionary is then used to expand the background dictionary and the target dictionary, respectively, to form an extended background dictionary and an extended target dictionary. Based on collaborative representation, the extended background dictionary and the extended target dictionary are then used to reconstruct a test sample, respectively. The reconstruction error of the extended background dictionary and the reconstruction error of the extended target dictionary are calculated. Finally, based on the reconstruction errors of the extended background dictionary and the extended target dictionary, the test sample is determined to be a camouflaged target. This method merges the background spectral variation dictionary and the target spectral variation dictionary to form a joint spectral variation dictionary, achieving the migration of spectral variations between the target and the background. Using the joint spectral variation dictionary to expand the background dictionary and the target dictionary increases the diversity of spectral variations in the dictionary, increases the number of samples, and solves the spectral variation problem, thereby improving the detection accuracy of the method. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 A flowchart of a camouflaged target detection method based on spectral augmentation provided by an embodiment of the present invention;

[0041] Figure 2A flowchart of a camouflaged target detection method based on spectral augmentation provided by an embodiment of the present invention;

[0042] Figure 3 A flowchart of a camouflaged target detection method based on spectral augmentation provided by an embodiment of the present invention;

[0043] Figure 4 A flowchart of a camouflaged target detection method based on spectral augmentation provided by an embodiment of the present invention;

[0044] Figure 5 This is a flow chart of a camouflaged target detection method based on spectral variation augmentation provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0045] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0046] In one embodiment, according to Figure 1 As shown, Figure 1 This is one of the flow charts of the camouflaged target detection method based on spectral variation augmentation provided by an embodiment of the present invention, which includes the following steps:

[0047] S101 . Obtain a background spectrum change dictionary and a target spectrum change dictionary according to a hyperspectral image, wherein the hyperspectral image includes training samples and test samples, and the training samples include background samples and target samples.

[0048] In this embodiment, the hyperspectral image is obtained by a satellite, and all samples in the hyperspectral image are divided into training samples and test samples, where the training samples include background and target samples and their corresponding labels (background / target), and the test samples include unlabeled background samples and target samples to be tested.

[0049] S102: Merge the background spectrum change dictionary and the target spectrum change dictionary to form a joint spectrum change dictionary.

[0050] The joint spectral change dictionary realizes the migration of spectral changes between target and background by combining the target and background spectral change dictionaries.

[0051] S103 , using the joint spectral variation dictionary to expand the background dictionary and the target dictionary respectively to form an expanded background dictionary and an expanded target dictionary, where the background dictionary is a pixel set of all background samples, and the target dictionary is a set of all target samples.

[0052] The background dictionary and target dictionary are expanded respectively by using the joint spectral variation dictionary to increase the diversity of spectral variation of the background dictionary and target dictionary, thereby increasing the number of samples and overcoming the spectral variation problem.

[0053] S104 , reconstructing the test sample using the extended background dictionary and the extended target dictionary based on the collaborative representation, and calculating the reconstruction error of the extended background dictionary and the reconstruction error of the extended target dictionary.

[0054] S105 , judging whether the test sample is a disguised target based on the reconstruction error of the extended background dictionary and the reconstruction error of the extended target dictionary.

[0055] Preferably, a detection discriminator can be used to determine whether the test sample is a disguised target, and the reconstruction error of the background dictionary and the reconstruction error of the target dictionary are input into the discriminator, and the discriminator outputs a detection result, which is the target or the background.

[0056] The camouflaged target detection method based on spectral change augmentation in this embodiment realizes the migration of spectral changes between the target and the background by combining the target and background spectral change dictionaries. The background dictionary and the target dictionary are respectively expanded using the combined spectral change dictionary to increase the diversity of the spectral changes of the background dictionary and the target dictionary, fully mine the prior information, and enhance the spectral changes, thereby increasing the number of samples, overcoming the spectral change problem, and improving the accuracy of camouflaged target detection.

[0057] In an optional embodiment, if Figure 2 As shown, Figure 2 This is one of the flow diagrams of the camouflaged target detection method based on spectral change augmentation provided by an embodiment of the present invention. This embodiment involves an optional method of obtaining a background spectral change dictionary and a target spectral change dictionary based on a hyperspectral image. Based on the above embodiment, step S101 includes:

[0058] S201. Acquire hyperspectral images

[0059] S202: clustering the background samples using an unsupervised clustering algorithm to obtain sample sets of several categories and cluster centers of each category.

[0060] Specifically, in this embodiment, an unsupervised clustering algorithm is used to cluster the background samples into 20 to 40 categories. The unsupervised clustering algorithm is a K-Means clustering algorithm, a DBSCAN clustering algorithm, or a hierarchical clustering algorithm.

[0061] For different hyperspectral images, the number of categories into which background samples are clustered is also different. The unsupervised clustering algorithm can directly obtain the categories of background samples and the cluster centers corresponding to the categories.

[0062] S203 : Construct a background spectrum variation dictionary based on the difference between the background sample and the cluster center of the category to which the background sample belongs.

[0063] Specifically, for a category of background samples, the difference between all background sample pixels in the category and their corresponding cluster centers is calculated one by one as the spectral change of the category. The spectral change combination of all categories is used to form a background spectral change dictionary.

[0064] S204: Construct a target spectrum change dictionary according to the difference between the target sample and the target sample mean.

[0065] Specifically, the difference between each target sample and the target sample mean is calculated, and all the differences are combined to form a target spectrum change dictionary.

[0066] In one embodiment, a cascade operation is used to merge the background spectrum variation dictionary and the target spectrum variation dictionary to form a joint spectrum variation dictionary.

[0067] In an optional embodiment, using a joint spectral variation dictionary to expand the background dictionary and the target dictionary to form an extended background dictionary and an extended target dictionary respectively includes: combining the joint spectral variation dictionary with the background dictionary to form an extended background dictionary, and combining the joint spectral variation dictionary with the target dictionary to form an extended target dictionary. It should be noted that the combination in this embodiment is a cascade, i.e., the joint spectral variation dictionary is cascaded with the background dictionary to form the extended background dictionary, and the joint spectral variation dictionary is cascaded with the target dictionary to form the extended target dictionary.

[0068] In a preferred embodiment, Figure 3 As shown, Figure 3 This is one of the flow diagrams of the camouflaged target detection method based on spectral variation augmentation provided by an embodiment of the present invention. This embodiment involves an optional method of reconstructing a test sample using an extended background dictionary and an extended target dictionary based on collaborative representation, and calculating the reconstruction error of the extended background dictionary and the reconstruction error of the extended target dictionary. Based on the above embodiment, step S104 includes:

[0069] S301: Calculate collaborative representation coefficients corresponding to the extended background dictionary and the extended target dictionary.

[0070] Specifically, expand the target dictionary D te The corresponding synergy representation coefficient is calculated as

[0071] γ te =D te T D te +λI -1 Dte T x (2)

[0072] Extended Background Dictionary D be The corresponding synergy representation coefficient is calculated as

[0073] γ be =D be T D be λI -1 D be T x (3)

[0074] Where λ is the regularization parameter, λ∈(0.001,0.01,0.1), and I represents the identity matrix.

[0075] S302: Use the extended background dictionary and the extended target dictionary to collaboratively represent the test samples.

[0076] It should be noted that the collaborative representation of the test sample is to reconstruct the test sample. Optionally, the collaborative representation of the test sample is

[0077]

[0078] Where, represents the collaborative representation of the test sample using the extended target dictionary, Denotes the collaborative representation of the test sample using the extended background dictionary.

[0079] S303 : Calculate the reconstruction error of the extended background dictionary and the reconstruction error of the extended target dictionary based on the collaborative representation of the test sample by the test sample, the extended background dictionary, and the extended target dictionary.

[0080] Specifically, in this embodiment, the reconstruction error of the extended background dictionary and the reconstruction error of the extended target dictionary are calculated based on the collaborative representation of the test sample, the extended background dictionary and the extended target dictionary.

[0081] r t (x)=||xD te γ te ||2

[0082] r b (x)=||xA be γ be ||2 (1)

[0083] In the formula, x represents the test sample, r t (x) represents the reconstruction error of the expanded target dictionary, r b (x) represents the reconstruction error of the extended background dictionary, Dte r te Denotes the collaborative representation of the test sample using the extended target dictionary, D be r be Denotes the collaborative representation of the test sample using the extended background dictionary, D te Denotes the extended target dictionary, D be Represents the extended background dictionary, r te represents the collaborative representation coefficient corresponding to the extended target dictionary, r be Indicates the collaborative representation coefficient corresponding to the extended background dictionary.

[0084] In one embodiment, Figure 4 As shown, Figure 4 This is one of the flow diagrams of the camouflaged target detection method based on spectral variation augmentation provided by an embodiment of the present invention. This embodiment involves an optional method of reconstructing a test sample using an extended background dictionary and an extended target dictionary based on collaborative representation, and calculating the reconstruction error of the extended background dictionary and the reconstruction error of the extended target dictionary. Based on the above embodiment, step S105 includes:

[0085] S401 : Calculate the difference between the reconstruction error of the extended background dictionary and the reconstruction error of the extended target dictionary.

[0086] S402. Determine the difference between the reconstruction error of the extended background dictionary and the reconstruction error of the extended target dictionary and the error threshold. If the difference between the reconstruction error of the extended background dictionary and the reconstruction error of the extended target dictionary is greater than or equal to the error threshold, the test sample is determined to be a disguised target. If the difference between the reconstruction error of the extended background dictionary and the reconstruction error of the extended target dictionary is less than the error threshold, the test sample is determined to be background.

[0087] In a specific embodiment, Figure 5 As shown, Figure 5 This is one of the flow diagrams of a disguised target detection method based on spectral variation augmentation provided by an embodiment of the present invention. In the disguised target detection method based on spectral variation augmentation of the present invention, a K-Means clustering algorithm is specifically used to divide background samples into different categories. The method includes the following steps:

[0088] (1) Obtain a hyperspectral image and convert the N b The pixel set of background samples is defined as the background dictionary B represents the number of bands.

[0089] (2) The background samples are divided into W categories using the K-Means clustering algorithm. After clustering, the pixel set of the i-th category is recorded as n iRepresents the number of samples in the i-th category after clustering, and the cluster center of the i-th category is c i , then the set of cluster centers is recorded as

[0090] (3) Construct a background spectrum change dictionary, for the pixel set of the i-th category after clustering Calculate pixels one by one The cluster center c of the category to which it belongs i The spectral difference between Then the set of spectral changes extracted from all background samples in the i-th category is recorded as The background spectrum variation dictionary constructed using background samples of all categories is recorded as

[0091] (4) Construct the target spectral change dictionary, assuming that the hyperspectral image provides N t target samples, then the target dictionary is constructed using the set of target samples, which is recorded as Average the spectra of all target samples in the dictionary to obtain the target sample mean The target spectrum change s is obtained by calculating the difference between each target sample and the target sample mean i , then the target spectrum change dictionary constructed by all target spectrum changes is recorded as

[0092] (5) Combine the background spectrum change dictionary and the target spectrum change dictionary through cascade operation to form a joint spectrum change dictionary V = V b S t .

[0093] (6) By extending the target dictionary and the background dictionary, the test samples are respectively Perform collaborative representation, solve the collaborative representation coefficients corresponding to the extended target dictionary and the extended background dictionary, and reconstruct the test samples.

[0094] (7) Determine whether the test sample is a disguised target based on the reconstruction error of the extended target dictionary and the extended background dictionary, and output the detection result.

[0095] Specifically, the following detectors are used to determine whether the test sample is a disguised target or background:

[0096]

[0097] Where δ represents the threshold range, which is usually set to 0.1 to 0.4. Alternatively, a δ that enables the false alarm rate or detection accuracy to reach a specified value can be selected through verification experiments.

[0098] The camouflaged target detection method based on spectral variation augmentation in this embodiment realizes the migration of spectral variations between the target and the background through a joint spectral variation dictionary. The background dictionary and the target dictionary are respectively expanded using the joint spectral variation dictionary to increase the diversity of spectral variations in the background dictionary and the target dictionary, and the prior information is fully mined. The problem of limited spectral variation diversity in the dictionary caused by spectral variation and a small number of samples is overcome through spectral variation augmentation, thereby improving the accuracy of camouflaged target detection.

[0099] Based on the same inventive concept, embodiments of the present application also provide a disguised target detection device based on spectral augmentation for implementing the aforementioned disguised target detection method based on spectral augmentation. The solution provided by this device is similar to the solution described in the aforementioned method. Therefore, the specific limitations of one or more points in the embodiments of the disguised target detection device based on spectral augmentation provided below can be found in the above-mentioned limitations of the disguised target detection method based on spectral augmentation, and will not be repeated here.

[0100] In one embodiment, a camouflaged target detection device based on spectral augmentation includes a memory and a processor. The memory stores a computer program. When the processor executes the computer program, the steps of the camouflaged target detection method based on spectral augmentation are implemented.

[0101] It should be understood that, although the steps in the flowcharts according to the embodiments described above are shown in sequence as indicated by the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction on the execution of these steps, and these steps may be executed in other orders. Moreover, at least a portion of the steps in the flowcharts according to the embodiments described above may include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but may be executed at different times. The execution order of these steps or stages is not necessarily sequential, but may be executed in turn or alternately with other steps or at least a portion of steps or stages in other steps.

[0102] The above-described embodiments merely represent several implementation methods of the present invention. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that a person skilled in the art would be able to make various modifications and improvements without departing from the spirit of the present invention, all of which fall within the scope of protection of the present invention. Therefore, the scope of protection of the present invention shall be determined by the appended claims.

Claims

1. A camouflaged target detection method based on spectral variation augmentation, characterized in that: include: Acquire a background spectrum change dictionary and a target spectrum change dictionary according to a hyperspectral image, wherein the hyperspectral image includes a training sample and a test sample, and the training sample includes a background sample and a target sample; Merging the background spectrum change dictionary and the target spectrum change dictionary to form a joint spectrum change dictionary; Using the joint spectral variation dictionary, the background dictionary and the target dictionary are respectively expanded to form an extended background dictionary and an extended target dictionary, wherein the background dictionary is a pixel set of all background samples, and the target dictionary is a set of all target samples; Based on the collaborative representation, the test samples are reconstructed using the extended background dictionary and the extended target dictionary respectively, and the reconstruction error of the extended background dictionary and the reconstruction error of the extended target dictionary are calculated; It is determined whether the test sample is a disguised target according to the reconstruction error of the extended background dictionary and the reconstruction error of the extended target dictionary.

2. The camouflaged target detection method based on spectrum augmentation according to claim 1, characterized in that: Obtaining background spectrum change dictionary and target spectrum change dictionary based on hyperspectral image includes: Acquire hyperspectral images; An unsupervised clustering algorithm is used to cluster background samples and obtain sample sets of several categories and cluster centers of each category; Construct a background spectrum change dictionary based on the difference between the background sample and the cluster center of the category to which the background sample belongs; The target spectrum variation dictionary is constructed based on the difference between the target sample and the target sample mean.

3. The camouflaged target detection method based on spectrum augmentation according to claim 2, characterized in that: A cascade operation is used to merge the background spectral change dictionary and the target spectral change dictionary to form a joint spectral change dictionary.

4. The camouflaged target detection method based on spectrum augmentation according to claim 3, characterized in that: The background dictionary and the target dictionary are respectively expanded using the joint spectral variation dictionary to form an expanded background dictionary and an expanded target dictionary including: combining the joint spectrum change dictionary and the background dictionary to form an extended background dictionary; The joint spectral variation dictionary and the target dictionary are combined to form an extended target dictionary.

5. The camouflaged target detection method based on spectrum augmentation according to claim 4, characterized in that: Based on the collaborative representation, the test samples are reconstructed using the extended background dictionary and the extended target dictionary respectively. The reconstruction error of the extended background dictionary and the reconstruction error of the extended target dictionary are calculated. Calculate the collaborative representation coefficients corresponding to the extended background dictionary and the extended target dictionary; Use the extended background dictionary and the extended target dictionary to collaboratively represent the test samples respectively; The reconstruction error of the extended background dictionary and the reconstruction error of the extended target dictionary are calculated based on the collaborative representation of the test sample by the test sample, the extended background dictionary and the extended target dictionary.

6. The camouflaged target detection method based on spectrum augmentation according to claim 5, characterized in that: The reconstruction error of the extended background dictionary and the reconstruction error of the extended target dictionary are calculated based on the collaborative representation of the test sample, the extended background dictionary and the extended target dictionary. In the formula, x represents the test sample, r t (x) represents the reconstruction error of the expanded target dictionary, r b (x) represents the reconstruction error of the extended background dictionary, D te r te Denotes the collaborative representation of the test sample using the extended target dictionary, D be r be Denotes the collaborative representation of the test sample using the extended background dictionary, D te Denotes the extended target dictionary, D be Represents the extended background dictionary, r te represents the collaborative representation coefficient corresponding to the extended target dictionary, r be Indicates the collaborative representation coefficient corresponding to the extended background dictionary.

7. The camouflaged target detection method based on spectrum augmentation according to claim 6, characterized in that: The calculation method of the collaborative representation coefficient corresponding to the extended target dictionary is: γ te =D te T D te +λI -1 D te T x (2) The calculation method of the collaborative representation coefficient corresponding to the extended background dictionary is: γ be =D be T D be +λI -1 D be T x (3) Where λ is the regularization parameter, λ∈(0.001,0.01,0.1), and I represents the identity matrix.

8. The camouflaged target detection method based on spectrum augmentation according to claim 5, characterized in that: Judging whether a test sample is a disguised target according to the reconstruction error of the extended background dictionary and the reconstruction error of the extended target dictionary includes: Calculate the difference between the reconstruction error of the expanded background dictionary and the reconstruction error of the expanded target dictionary; The difference between the reconstruction error of the extended background dictionary and the reconstruction error of the extended target dictionary and the error threshold is determined. If the difference between the reconstruction error of the extended background dictionary and the reconstruction error of the extended target dictionary is greater than or equal to the error threshold, the test sample is determined to be a disguised target. If the difference between the reconstruction error of the extended background dictionary and the reconstruction error of the extended target dictionary is less than the error threshold, the test sample is determined to be background.

9. The camouflaged target detection method based on spectrum augmentation according to claim 2, characterized in that: The background samples are clustered into 20 to 40 categories using an unsupervised clustering algorithm, wherein the unsupervised clustering algorithm is a K-Means clustering algorithm, a DBSCAN clustering algorithm, or a hierarchical clustering algorithm.

10. A camouflaged target detection device based on spectrum augmentation, comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 9 are implemented.

Images