A method and device for detecting small infrared targets based on spherical space and multi-directional derivatives

Abstract

The present invention discloses a method and device for detecting small infrared targets based on spheroidization space and multi-directional derivatives. The method comprises: (1) establishing a spheroidization space based on the original thermal infrared image sequence #imgabs0#, mapping the thermal infrared image to the spheroidization space for data spheroidization, and obtaining a spheroidization image sequence #imgabs1#; (2) transforming the sequence #imgabs2# from the spatial domain to the frequency domain, and constructing a frequency domain filter f σ To suppress low-frequency signals, the image data is transformed into the spatial domain to obtain image D f ; (3) Based on image D f The gradient information is used to calculate the weight matrix W based on the multi-directional fusion derivatives, and the weight sequence #imgabs3# (4) is obtained. The weight sequence #imgabs4# is used to weight the spherical image sequence #imgabs5# and output the small target detection result sequence #imgabs6# to realize thermal infrared small target detection. The present invention utilizes the designed infrared small target detection method based on spherical space and multi-directional derivatives to effectively improve the detection performance of thermal infrared small targets.

Description

Technical Field

[0001] The invention relates to the field of image processing, and in particular to a method and a device for detecting small infrared targets based on spheroidized space and multi-directional derivatives. Background Art

[0002] In recent years, infrared small target detection has attracted more and more attention in various fields such as military and civilian, especially in air defense and marine surveillance. Due to the need for long-distance detection, the targets of interest in thermal infrared images are usually 2×2 to 9×9 in size. Their "weak" and "small" characteristics make infrared small target detection a challenging task, which faces the following challenges:

[0003] 1) Due to long-range imaging and background variability, the signal clutter of infrared images is relatively low, which indicates that the contrast between small targets and the background is low, increasing the difficulty of target detection;

[0004] 2) There are many interferences, such as inherent noise and structural noise, which may cause false detection to a certain extent, and the background suppression ability is weak;

[0005] 3) The infrared search and tracking system has a high time-efficiency performance for the infrared small target detection method, which indicates that the infrared small target detection method needs to improve the computational efficiency while ensuring accuracy to meet the needs of real-time applications.

[0006] Generally, infrared dim small target detection methods can be divided into traditional methods and deep learning-based methods. Background consistency methods assume that the infrared background is highly correlated and design filters to extract small targets, such as Max-Median, Max-Mean, and wavelet transform. However, in complex scenes, these filters will not only enhance the target, but also highlight other background components, causing false alarms.

[0007] Methods based on the human visual system (HVS) aim to detect objects based on the contrast between the object and its surrounding neighborhood. The most representative HVS method is the local contrast metric (LCM). Subsequently, different LCM variants have been proposed, such as global sparse weighted LCM (GSWLCM), energy local uncertainty metric (ELUM), and gradient filter-based gradient contrast metric (GCF-CM). However, these detectors are highly dependent on the contrast metric between the object and the background, which can easily cause false detection in complex scenes.

[0008] Detection methods based on low-rank sparse decomposition rely on the following assumptions: the background exhibits low-rank characteristics because most of the background parts can be composed of similar image patches; the target exhibits sparsity because it only occupies a small number of pixels. Therefore, low-rank sparse decomposition methods can separate target components from the original infrared image, such as LogTFNN, NRAM, METTR, STT-TRNR, STTM-TSNM and 4D-TR. However, many methods do not fully utilize or destroy the spatiotemporal information in a given sequence, resulting in background shrinkage or target loss.

[0009] With the continuous improvement of hardware equipment and the continuous development of deep learning technology, more and more infrared small target detection methods based on deep learning have been proposed, including KCPNet, APAFNet, STDMANet and MSAFFNet. However, these methods have high requirements for sample data and the workload of making sample labels is huge. In addition, infrared small target detection methods based on deep learning focus on the target itself and lack robustness under different backgrounds.

[0010] In summary, infrared small target detection is a challenging task, and it is urgent to solve problems such as insufficient detection capability of weak and small targets and insufficient background suppression capability. Summary of the invention

[0011] In view of the deficiencies in the prior art, the purpose of the present invention is to provide a thermal infrared small target detection method and device based on spheroidized space and multi-directional derivatives, which fully utilize the low-frequency characteristics of background components in infrared images, the obvious gradient difference between the target and the background, etc., and can use the construction of spheroidized space and high-pass filter to perform data spheroidization and high-pass filtering on the infrared image to greatly suppress the background components; and use the multi-directional gradient method to calculate the gradient values ​​of the image in different directions, effectively capture the target, enhance the target's significance, and realize infrared small target detection; in addition, the target detection ability, background suppression ability and comprehensive performance of the method of the present invention have been effectively verified.

[0012] To achieve the above object, the present invention provides the following technical solutions:

[0013] The present invention discloses a method for detecting small infrared targets based on spheroidized space and multi-directional derivatives, which is characterized by comprising the following steps:

[0014] Step 1): Based on the original thermal infrared image sequence Establishing spherical space Ω DS , map the infrared image D to the spherical space Ω DS Spherize the data to obtain a spherical image sequence

[0015] Step 2): Spherize the image sequence Each frame of the spheroidized image D DS , using fast Fourier transform to convert D DS Transform from spatial domain to frequency domain to obtain Construct frequency domain filter f σ To inhibit The low-frequency signal in the frequency domain is used to obtain the image data in the frequency domain Then Transform to the spatial domain to obtain the high-pass filtered image D f ;

[0016] Step 3): Based on high-pass filtering image D f Gradient features, define the main direction and secondary direction, and calculate each frame D f The corresponding weight matrix W based on multi-directional fusion derivatives;

[0017] Step 4): For the spherical image sequence Use the weight matrix W to weight the corresponding spherical image D DS , get the target detection result T, and output the small target detection result sequence Each frame image in the sequence is a small target detection result image, realizing thermal infrared small target detection.

[0018] The present invention also discloses an infrared small target detection device based on spheroidized space and multi-directional derivatives for implementing the method, which comprises:

[0019] The data spheroidization module establishes a spheroidization space, maps the original infrared image sequence to the spheroidization space for data spheroidization, and obtains a spheroidized image sequence;

[0020] The high-pass filter module transforms each frame of the spheroidized image in the spheroidized image sequence from the spatial domain to the frequency domain, constructs a frequency domain filter to suppress low-frequency signals, and then transforms it to the spatial domain to obtain a high-pass filtered image;

[0021] Based on the weight information extraction module of multi-directional fusion derivatives, the weight matrix of multi-directional fusion derivatives is designed and calculated by using the gradient information of high-pass filtered images;

[0022] The spheroidization data weighting module weights the spheroidization image with the weight matrix to obtain the infrared small target detection result;

[0023] The target detection result output module is used to output the infrared small target detection result sequence.

[0024] The beneficial effects of the present invention are:

[0025] 1) The present invention makes full use of the low-frequency characteristics of the background components in the infrared image, and can perform data spheroidization and high-pass filtering on the infrared image through the construction of spheroidization space and high-pass filter, thereby suppressing the background components and enhancing high-frequency information such as small targets in the image;

[0026] 2) The present invention makes full use of the obvious gradient difference between the target and the background. By calculating and fusing the gradient values ​​of the image in different directions, the designed weight matrix of the multi-directional fusion derivative can effectively enhance the saliency of the target, greatly suppress the background, noise and other components, and realize infrared weak target detection. The background suppression ability, target detection ability and comprehensive effectiveness of this method have been verified. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 The data spherification process diagram of the present invention, (a) represents the original data distribution, (b) represents the data distribution after centralization, and (c) represents the data distribution diagram after mapping to the spherification space;

[0028] Figure 2 A schematic diagram of the main direction and the secondary direction defined in the present invention;

[0029] Figure 3 It is a structural schematic diagram of the infrared small target detection device of the present invention;

[0030] Figure 4 This is a sample frame image of a thermal infrared image sequence used for experimental testing;

[0031] Figure 5 The thermal infrared small target detection result of the example frame of the thermal infrared image sequence;

[0032] Figure 6 It is a thermal infrared image example frame and the small target detection result image of the example frame by this method, ELUM and LogTFNN and the corresponding three-dimensional grayscale visualization image;

[0033] Figure 7 The small target detection results and the corresponding three-dimensional grayscale visualization image of this example frame by METTR, STT-TRNR, STTM-TSNM and 4D-TR. DETAILED DESCRIPTION

[0034] In order to make the purpose, technical solutions and advantages of the present invention clearer, the present invention is described in detail below in conjunction with specific embodiments. Specific embodiments are described below to simplify the present invention. However, it should be recognized that the present invention is not limited to the illustrated embodiments, and various modifications of the present invention are possible without departing from the basic principles, and these equivalent forms also fall within the scope defined by the claims attached to the present application.

[0035] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the accompanying drawings in the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0036] The steps of the infrared small target detection method based on spheroidized space and multi-directional derivatives in this embodiment of the present invention mainly include:

[0037] Step 1: Based on the original thermal infrared image sequence Establishing spherical space Ω DS , map the infrared image D to the spherical space Ω DS Spherize the data to obtain a spherical image sequence

[0038] Specifically, for the infrared image sequence It contains n3 frames of infrared images, and the infrared image frames are stacked frame by frame in the order of frames from the time dimension. Get the infrared image tensor Expand by time dimension Get the matrix It can be expressed as:

[0039]

[0040] Among them, reshape(·) represents the tensor reconstruction operator. Reshape into a matrix of size [n1n2,n3] Specifically in the embodiment, n1 and n2 are both set to 256, and n3 is set to 50;

[0041] right The corresponding covariance matrix C is characteristically decomposed as shown below:

[0042]

[0043] Where μ represents The column mean vector of , (·) T represents the matrix transpose operation, V represents the eigenvector matrix, and ∑ represents the diagonal matrix;

[0044] Based on V and ∑ obtained by eigendecomposition, construct the spherical space Ω DS , the corresponding projection matrix P is:

[0045]

[0046] Wherein, k represents a scaling factor. Specifically in the embodiment, I represents the identity matrix,;

[0047] The projection matrix P can be used to convert the matrix Mapping to spherical space Ω DS , as follows:

[0048]

[0049] in, express In tensor form in spheroidized space, Ω DS (·) represents the spherical space mapping operator, and the matrix is ​​transformed by the projection matrix P Projection to the constructed spherical space Ω DS , and we get the matrix And reshape it into a tensor of scale [n1, n2, n3] The data distribution changes during data spherification are as follows: Figure 1 As shown;

[0050] From the tensor Extract each frontal slice D DS , and obtain the spheroidized image sequence

[0051] Step 2: Spherize the image sequence Each frame of the spheroidized image D DS , using fast Fourier transform to convert D DS Transform from spatial domain to frequency domain to obtain Construct frequency domain filter f σ To inhibit The low-frequency signal in the frequency domain is used to obtain the image data in the frequency domain Then Transform to the spatial domain to obtain the high-pass filtered image D f ;

[0052] Specifically, for the spheroidized image sequence obtained in step 1) Using fast Fourier transform, the sequence Each frame of the spheroidized image D DS Mapping from the spatial domain to the frequency domain, we get And there is Among them, fft(·) is the fast Fourier transform operator; Centralize and get centralized data And there is Among them, fftshift(·) is the data centering operator;

[0053] Construct a two-dimensional high-pass filter f σ To suppress centralized data The low-frequency components of the data for

[0054]

[0055] Among them, f σ is a two-dimensional high-pass filter, ρ represents the variance of the filter, "*" represents the dot product operator in the frequency domain, and in the embodiment, σ is taken as 100;

[0056] Furthermore, Perform a decentralized operation and finally map it to the spatial domain to obtain a high-pass filtered image D f

[0057]

[0058] Among them, ifft(·) is the inverse fast Fourier transform operator, and ifftshift(·) is the data decentralization operator.

[0059] Step 3: Filter image D based on high pass f Gradient features, define the main direction and secondary direction, and calculate each frame D f The corresponding weight matrix W based on multi-directional fusion derivatives;

[0060] Specifically, the high-pass filtered image D f The gradient information is in, is the gradient operator, G x For image D f The horizontal derivative of G y For image D f The vertical derivative of the image D f The directional derivative matrix along direction θ is

[0061] G θ =G x cosθ+G y sinθ (7)

[0062] Set 0, π / 2, π, 3π / 2 as the main directions, π / 4, 3π / 4, 5π / 4, 7π / 4 as the secondary directions. The schematic diagram of the main direction and the secondary direction is as follows Figure 2 As shown, the main direction derivative matrix is ​​G0, G π / 2 , G π , G 3π / 2 , the secondary directional derivative matrix is ​​G π / 4 , G 3π / 4 , G 5π / 4 , G 7π / 4 ; For any position point (x, y) in the image, take it as the center and define the multi-directional fusion derivative Q of (x, y) in each quadrant i (x, y):

[0063]

[0064] Among them, a1 and a2 represent the harmonic weight coefficients, Represents the square of the directional derivative value of (x, y) in the θ direction, from which the multi-directional fusion derivative matrix Q can be obtained i , i = 1, 2, 3, 4, multi-directional fusion derivative Q in each quadrant i (x, y) is the multi-directional fusion derivative matrix Q i The value at the position (x, y). Specifically in the embodiment, a1 and a2 are taken as 1;

[0065] Infrared weak targets usually cause significant changes in the grayscale value of the local area. In order to highlight the area with significant grayscale value changes and suppress the area with slow grayscale value changes, the multi-directional fusion derivative matrix Q i , define the weight matrix W based on multi-directional fusion derivatives, the weight W(x, y) at the point (x, y) is

[0066]

[0067] Where u(·) is the unit step function, k i is the adaptive threshold, calculated as

[0068] k i =mean(Q i )+γ·var(Q i ) (10)

[0069] Among them, mean(Q i ) represents the matrix Q i The mean value of var(Q i ) is the matrix Q i The variance of γ is 3 in the specific embodiment.

[0070] Step 4: Spherized Image Sequence Use the weight matrix W to weight the corresponding spherical image D DS , get the target detection result T, and output the small target detection result sequence Realize thermal infrared small target detection;

[0071] Specifically, for the spheroidized image sequence Each frame of the spheroidized image D DS , use the corresponding weight matrix W for weighted calculation, and get the target detection result T corresponding to each frame image:

[0072] T=W⊙D DS (11) Where “⊙” is the Hadamard product;

[0073] This outputs a sequence of small target detection results Realize thermal infrared small target detection.

[0074] Corresponding to the aforementioned embodiment of an infrared small target detection method based on spheroidized space and multi-directional derivatives, the present invention also provides an embodiment of an infrared small target detection device based on spheroidized space and multi-directional derivatives.

[0075] Figure 3 FIG. 1 is a block diagram of an infrared small target detection device based on spherical space and multi-directional derivatives according to an exemplary embodiment. Figure 3 As shown, the device comprises:

[0076] The data spheroidization module establishes a spheroidization space, maps the original infrared image sequence to the spheroidization space for data spheroidization, and obtains a spheroidized image sequence;

[0077] The high-pass filter module transforms each frame of the spheroidized image in the spheroidized image sequence from the spatial domain to the frequency domain, constructs a frequency domain filter to suppress low-frequency signals, and then transforms it to the spatial domain to obtain a high-pass filtered image;

[0078] Based on the weight information extraction module of multi-directional fusion derivatives, the weight matrix of multi-directional fusion derivatives is designed and calculated by using the gradient information of high-pass filtered images;

[0079] The spheroidization data weighting module weights the spheroidization image with the weight matrix to obtain the infrared small target detection result;

[0080] The target detection result output module is used to output the infrared small target detection result sequence.

[0081] Regarding the device in the above embodiment, the specific manner in which each module performs operations has been described in detail in the embodiment of the method, and will not be elaborated here.

[0082] For the device embodiment, since it basically corresponds to the method embodiment, the relevant parts can refer to the partial description of the method embodiment. The device embodiment described above is only schematic. The various modules in the device are a kind of logical function division. There may be other division methods in actual implementation. For example, multiple modules can be combined or integrated into another unit. Another point is that the connection between the modules shown or discussed can be a communication connection through some interfaces, which can be electrical or other forms. Some or all of the modules can be selected according to actual needs to achieve the purpose of the present application. Ordinary technicians in this field can understand and implement it without paying creative labor. The following takes the disclosed real thermal infrared image sequence as an example to illustrate the specific implementation method to reflect the technical effect of the present invention. The specific steps in the embodiment will not be repeated.

[0083] Example

[0084] The accompanying drawings shown in the embodiments of the present invention can make the purpose, technical solutions and advantages of the present invention more clearly described. It should be noted that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention. All equivalent replacements, improvements, etc. made within the method ideas and principles provided by the present invention should be included in the protection scope of the present invention.

[0085] In this embodiment, the effectiveness of the thermal infrared small target detection algorithm will be verified through a public infrared image sequence. The selected thermal infrared image sequence is described as follows: the number of frames is 100, the frame image size is 256×256, the weak flying object moves slowly, and there are bright white buildings, linear bright textures and strong noise in the background. Figure 4 shown. Figure 6 It is a thermal infrared image example frame and the small target detection result image of the example frame by the method, ELUM and LogTFNN respectively and the corresponding three-dimensional grayscale visualization image; Figure 7 The small target detection results and the corresponding 3D grayscale visualization images of the example frame are respectively obtained by METTR, STT-TRNR, STTM-TSNM and 4D-TR.

[0086] The sources of the comparison methods are as follows:

[0087] ELUM is from "E. Zhao, W. Zheng, M. Li, H. Sun, and J. Wang, “Infrared small target detection using local component uncertainty measure with consistency assessment,” IEEE Geosci. Remote Sens. Lett., vol. 19, pp. 1–5, 2022."

[0088] LogTFNN is from "X. Kong, C. Yang, S. Cao, C. Li, and Z. Peng, “Infrared small target detection via nonconvex tensor fibered rank approximation,” IEEE Trans. Geosci. Remote Sens., vol. 60, pp. 1–21, 2021."

[0089] METTR is from "Z. Cao, X. Kong, Q. Zhu, S. Cao, and Z. Peng, “Infrared dim target"

[0090] detection via mode-k1k2 extension tensor tubal rank under complex ocean environment,” ISPRS-J. Photogramm. Remote Sens., vol. 181, pp. 167–190, 2021."

[0091] STT-TRNR is from "H. Yi, C. Yang, R. Qie, J. Liao, F. Wu, T. Pu, and Z. Peng, “Spatial temporal tensor ring norm regularization for infrared small target detection,” IEEE Geosci. Remote Sens. Lett., 2023."

[0092] STTM-TSNM comes from "D.Pang, P.Ma, Y.Feng, T.Shan, R.Tao, and Q.Jin, "Tensorspectral k-support norm minimization for detecting infrared dim and smalltarget against urban backgrounds," 2023."

[0093] 4D-TR comes from "F.Wu, H.Yu, A.Liu, J.Luo, and Z.Peng, "Infrared small target detection using spatio-temporal 4d tensor train and ring unfolding," 2023."

[0094] From the qualitative analysis, it can be seen that the selected contrast algorithms can detect small targets, but the background suppression ability is weak, especially LogTFNN, METTR and STTM-TSNM, which still retain strong residual background in the detection results; while the method of the present invention can completely suppress the background and enhance the target significance during the small target detection process.

[0095] In order to evaluate the superiority of the proposed method more objectively and truly, the detection performance of the algorithm is evaluated from a quantitative perspective through the 3D-ROC evaluation index system. The 3D-ROC evaluation index system comes from "C.-I. Chang, "An effective evaluation tool for hyperspectral target detection: 3D receiver operating characteristic curve analysis," IEEE Trans. Geosci. Remote Sens., vol. 59, no. 6, pp. 5131–5153, 2020." This evaluation index system will evaluate the algorithm from the perspective of target detection ability, background suppression ability and the comprehensive effectiveness of the detector, as follows:

[0096] Table 1 3D-ROC evaluation index system

[0097]

[0098] Table 2 shows the AUC results of the thermal infrared image sequence obtained by the 3D-ROC evaluation index system. The bold and underlined values ​​represent the corresponding optimal AUC values ​​and suboptimal AUC values, respectively. Quantitative analysis of the data shows that although the background suppression ability of STTM-TSNM is the best among all methods, its target detection ability is weak, with an AUC of (D,F) Only 0.9700; on the contrary, the method proposed in the present invention has (F,τ) AUC SNPR It is the second best among the compared methods, while all other indicators are optimal, with the best target detection capability and comprehensive performance.

[0099] Based on the above qualitative and quantitative analysis, the thermal infrared small target detection method based on spherical space and multi-directional fusion derivatives proposed in the present invention has superior target detection capability, background suppression capability and comprehensive effectiveness, and is practical.

[0100] Table 2 Quantitative indicators of the detection results of thermal infrared image example sequences using the present invention, ELUM, LogTFNN, METTR, STT-TRNR, STTM-TSNM and 4D-TR

[0101] method <![CDATA[AUC (D，F) ]]> <![CDATA[AUC (D，τ) ]]> <![CDATA[AUC (F，τ) ]]> <![CDATA[AUC TD ]]> <![CDATA[AUC BS ]]> <![CDATA[AUC SNPR ]]> <![CDATA[AUC TDBS ]]> <![CDATA[AUC ODP ]]> The present invention 1.0000 0.9982 <![CDATA[ 1.2919e-3 ]]> 1.9982 0.9987 <![CDATA[ 7.7271e2 ]]> 0.9969 1.9969 ELUM 1.0000 0.9388 1.3790e-3 1.9388 <![CDATA[ 0.9986 ]]> 6.8073e2 0.9374 1.9374 LogTFNN <![CDATA[ 0.9999 ]]> 0.9455 2.8477e-3 1.9454 0.9971 3.3203e2 <![CDATA[ 0.9427 ]]> <![CDATA[ 1.9426 ]]> METTR 1.0000 <![CDATA[ 0.9722 ]]> 3.0329e-1 <![CDATA[ 1.9722 ]]> 0.6967 3.2055 0.6689 1.6689 STT-TRNR 0.9599 0.9508 8.0452e-2 1.9107 0.8794 1.1819e1 0.8704 1.8303 STTM-TSNM 0.9700 0.9335 1.0585e-3 1.9035 0.9689 8.8196e2 0.9325 1.9025 4D-TR 1.0000 0.9672 5.8890e-2 1.9672 0.9411 1.6424e1 0.9083 1.9083

[0102] The accompanying drawings shown in the embodiments of the present invention can make the purpose, technical solutions and advantages of the present invention more clearly described. It should be noted that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention. All equivalent replacements, improvements, etc. made within the method ideas and principles provided by the present invention should be included in the protection scope of the present invention.

Claims

1. A method for detecting small infrared targets based on spherical space and multi-directional derivatives, characterized in that: The steps include: Step 1): Based on the original thermal infrared image sequence , establish a spherical space , the thermal infrared image Mapping to spherical space Spherize the data to obtain a spherical image sequence ; The step 1) is specifically as follows: For the original thermal infrared image sequence , which contains Frame infrared image, stacking infrared image frames frame by frame in the order of frames from the time dimension , get the infrared image tensor , expand according to the time dimension , and we get the matrix , which can be expressed as: （1） in, Represents a tensor reshaping operator, Refactored to Matrix of size ; right The corresponding covariance matrix Perform eigendecomposition as shown below: （2） in, express The column mean vector of represents the matrix transpose operation, represents the eigenvector matrix, represents a diagonal matrix; Based on the eigendecomposition and , construct spherical space , the projection matrix corresponding to the spherical space for: （3） in, represents the scaling factor, represents the identity matrix; Using the projection matrix The matrix Mapping to spherical space , as follows: （4） in, express In tensor form in spheroidized space, Represents the spherical space mapping operator, which is composed of the projection matrix The matrix Projection to the constructed spherical space , and we get the matrix ( ), and refactor it into Scaled tensors ; From the tensor Extract each frontal slice , and obtain the spherical image sequence ; Step 2): Spheroidize the image sequence Each frame of the spherical image , using the fast Fourier transform Transform from spatial domain to frequency domain to obtain , construct frequency domain filter To inhibit The low-frequency signal in the frequency domain is used to obtain the image data in the frequency domain , and then Transform to the spatial domain to obtain a high-pass filtered image ; Step 3): Filter the image based on high pass Gradient features, define the main direction and secondary direction, and calculate each frame The corresponding weight matrix based on multi-directional fusion derivatives ; Step 4): For the spherical image sequence , using the weight matrix Weighted corresponding spherical image , get the target detection result , output small target detection result sequence , realizing thermal infrared small target detection.

2. The infrared small target detection method based on spheroidized space and multi-directional derivatives according to claim 1 is characterized in that: The step 2) is specifically as follows: For the spheroidized image sequence obtained in step 1) , using fast Fourier transform to transform the sequence Each frame of the spherical image Mapping from the spatial domain to the frequency domain, we get , and there is ,in, is the fast Fourier transform operator; Centralize and get centralized data , and there is ,in, is the data centering operator; Construct a 2D high-pass filter To suppress centralized data The low-frequency components of the data are obtained for （5） in, is a two-dimensional high-pass filter, , represents the variance of the filter, and "*" represents the dot product operator in the frequency domain; Furthermore, Perform decentralized operations and finally map them to the spatial domain to obtain a high-pass filtered image. : （6） in, is the inverse fast Fourier transform operator, Decentralize operators for data.

3. The infrared small target detection method based on spheroidized space and multi-directional derivatives according to claim 1 is characterized in that: The step 3) is specifically as follows: High pass filtered image The gradient information is ,in, is the gradient operator, For images The horizontal derivative of For images The vertical derivative of Along direction The directional derivative matrix of is （7） set up The main direction, is a secondary direction, then the main direction derivative matrix is ​​the matrix , the secondary directional derivative matrix is ; For any point in the image , taking it as the center, defining Multi-directional fusion derivatives in each quadrant : （8） in, and represents the harmonic weight coefficient, express exist The square of the directional derivative value in the direction, thus obtaining the multi-directional fusion derivative matrix , multi-directional fusion derivatives in each quadrant is the multi-directional fusion derivative matrix In Location The value at ; Infrared weak targets usually cause significant changes in the grayscale value of the local area. In order to highlight the area with significant grayscale value changes and suppress the area with slow grayscale value changes, the multi-directional fusion derivative matrix , define the weight matrix based on multi-directional fusion derivatives ,point The weight of for （9） in, is a unit step function, is the adaptive threshold, calculated as （10） in, Representation Matrix The mean of For the matrix The variance of Represents a positive constant.

4. The infrared small target detection method based on spheroidized space and multi-directional derivatives according to claim 1 is characterized in that: The step 4) is specifically as follows: Spheroidized image sequence Each frame of the spherical image , using the corresponding weight matrix Perform weighted calculation to obtain the target detection result corresponding to each frame image for （11） in," ” is Hadamard; This outputs a sequence of small target detection results , each frame image in the sequence is a small target detection result image, realizing thermal infrared small target detection.

5. An infrared small target detection device based on spheroidized space and multi-directional derivatives for implementing the method of claim 1, characterized in that: include: The data spheroidization module establishes a spheroidization space, maps the original infrared image sequence to the spheroidization space for data spheroidization, and obtains a spheroidized image sequence; The high-pass filter module transforms each frame of the spheroidized image in the spheroidized image sequence from the spatial domain to the frequency domain, constructs a frequency domain filter to suppress low-frequency signals, and then transforms it to the spatial domain to obtain a high-pass filtered image; Based on the weight information extraction module of multi-directional fusion derivatives, the weight matrix of multi-directional fusion derivatives is designed and calculated by using the gradient information of high-pass filtered images; The spheroidization data weighting module weights the spheroidization image with the weight matrix to obtain the infrared small target detection result; The target detection result output module is used to output the infrared small target detection result sequence.

Images